From 6d6cebb4a77a0fde1c6fdf4ec57a21f72d1117b6 Mon Sep 17 00:00:00 2001
From: Ralph Castain <rhc@open-mpi.org>
Date: Tue, 14 Nov 2006 19:34:59 +0000
Subject: [PATCH] Bring over the update to terminate orteds that are generated
by a dynamic spawn such as comm_spawn. This introduces the concept of a job
"family" - i.e., jobs that have a parent/child relationship. Comm_spawn'ed
jobs have a parent (the one that spawned them). We track that relationship
throughout the lineage - i.e., if a comm_spawned job in turn calls
comm_spawn, then it has a parent (the one that spawned it) and a "root" job
(the original job that started things).
Accordingly, there are new APIs to the name service to support the ability to get a job's parent, root, immediate children, and all its descendants. In addition, the terminate_job, terminate_orted, and signal_job APIs for the PLS have been modified to accept attributes that define the extent of their actions. For example, doing a "terminate_job" with an attribute of ORTE_NS_INCLUDE_DESCENDANTS will terminate the given jobid AND all jobs that descended from it.
I have tested this capability on a MacBook under rsh, Odin under SLURM, and LANL's Flash (bproc). It worked successfully on non-MPI jobs (both simple and including a spawn), and MPI jobs (again, both simple and with a spawn).
This commit was SVN r12597.
---
ompi/attribute/attribute_predefined.c | 4 +-
ompi/communicator/comm.c | 2 +-
ompi/communicator/comm_dyn.c | 60 +-
ompi/communicator/communicator.h | 2 +-
ompi/mca/btl/mvapi/btl_mvapi_endpoint.c | 4 +-
ompi/mca/btl/openib/btl_openib_endpoint.c | 4 +-
ompi/mca/btl/tcp/btl_tcp_endpoint.c | 2 +-
ompi/mca/btl/udapl/btl_udapl_endpoint.c | 8 +-
ompi/mca/coll/sm/coll_sm_module.c | 2 +-
ompi/mca/pml/base/pml_base_module_exchange.c | 10 +-
ompi/mca/pml/dr/pml_dr.c | 4 +-
ompi/proc/proc.c | 26 +-
ompi/runtime/ompi_mpi_abort.c | 4 +-
ompi/runtime/ompi_mpi_init.c | 2 +-
ompi/runtime/ompi_mpi_io.c | 8 +-
orte/dss/dss_pack.c | 2 +-
orte/include/orte/orte_constants.h | 4 +-
orte/include/orte/orte_types.h | 1 +
orte/mca/errmgr/base/errmgr_base_receive.c | 4 +-
orte/mca/errmgr/hnp/errmgr_hnp.c | 17 +-
orte/mca/gpr/proxy/gpr_proxy_component.c | 6 +-
.../gpr_replica_deliver_notify_msg_api.c | 1 +
.../functional_layer/gpr_replica_cleanup_fn.c | 4 +-
.../gpr_replica_messaging_fn.c | 6 +-
.../gpr_replica_trig_ops_fn.c | 29 +-
orte/mca/gpr/replica/gpr_replica_component.c | 4 +-
orte/mca/iof/base/iof_base_close.c | 2 +-
orte/mca/iof/base/iof_base_endpoint.c | 11 +-
orte/mca/iof/proxy/iof_proxy.c | 18 +-
orte/mca/iof/proxy/iof_proxy_component.c | 4 +-
orte/mca/iof/proxy/iof_proxy_svc.c | 6 +-
orte/mca/iof/svc/iof_svc.c | 20 +-
orte/mca/iof/svc/iof_svc_component.c | 4 +-
orte/mca/iof/svc/iof_svc_proxy.c | 10 +-
orte/mca/iof/svc/iof_svc_pub.c | 12 +-
orte/mca/iof/svc/iof_svc_sub.c | 14 +-
orte/mca/ns/base/Makefile.am | 9 +-
orte/mca/ns/base/base.h | 202 ----
.../ns_data_type_compare_fns.c | 60 +-
.../data_type_support/ns_data_type_copy_fns.c | 21 +-
.../ns_data_type_packing_fns.c | 43 +-
.../ns_data_type_print_fns.c | 36 +-
.../ns_data_type_release_fns.c | 2 +-
.../data_type_support/ns_data_type_size_fns.c | 6 +-
.../ns_data_type_unpacking_fns.c | 27 +-
orte/mca/ns/base/ns_base_cell_fns.c | 232 ++++
orte/mca/ns/base/ns_base_diag_fns.c | 98 ++
orte/mca/ns/base/ns_base_general_fns.c | 78 ++
orte/mca/ns/base/ns_base_job_fns.c | 169 +++
orte/mca/ns/base/ns_base_local_fns.c | 678 -----------
orte/mca/ns/base/ns_base_open.c | 27 +-
orte/mca/ns/base/ns_base_vpid_name_fns.c | 397 ++++++
orte/mca/ns/base/ns_private.h | 285 +++++
orte/mca/ns/ns.h | 372 +++---
orte/mca/ns/ns_types.h | 54 +-
orte/mca/ns/proxy/Makefile.am | 33 +-
orte/mca/ns/proxy/configure.params | 3 +-
orte/mca/ns/proxy/{src => }/ns_proxy.h | 43 +-
orte/mca/ns/proxy/ns_proxy_cell_fns.c | 440 +++++++
.../ns/proxy/{src => }/ns_proxy_component.c | 64 +-
orte/mca/ns/proxy/ns_proxy_diag_fns.c | 313 +++++
orte/mca/ns/proxy/ns_proxy_general_fns.c | 495 ++++++++
orte/mca/ns/proxy/ns_proxy_job_fns.c | 526 ++++++++
orte/mca/ns/proxy/src/Makefile.extra | 23 -
orte/mca/ns/proxy/src/ns_proxy.c | 1071 -----------------
orte/mca/ns/replica/Makefile.am | 36 +-
orte/mca/ns/replica/configure.params | 3 +-
orte/mca/ns/replica/{src => }/ns_replica.c | 221 ----
orte/mca/ns/replica/{src => }/ns_replica.h | 140 ++-
orte/mca/ns/replica/ns_replica_cell_fns.c | 297 +++++
.../ns/replica/ns_replica_class_instances.h | 172 +++
orte/mca/ns/replica/ns_replica_component.c | 309 +++++
orte/mca/ns/replica/ns_replica_diag_fns.c | 322 +++++
orte/mca/ns/replica/ns_replica_general_fns.c | 403 +++++++
orte/mca/ns/replica/ns_replica_job_fns.c | 295 +++++
orte/mca/ns/replica/ns_replica_recv.c | 457 +++++++
orte/mca/ns/replica/ns_replica_support_fns.c | 120 ++
orte/mca/ns/replica/src/Makefile.extra | 24 -
.../mca/ns/replica/src/ns_replica_component.c | 702 -----------
orte/mca/odls/bproc/odls_bproc.c | 6 +-
orte/mca/odls/default/odls_default_module.c | 17 +-
orte/mca/oob/base/base.h | 17 +-
orte/mca/oob/base/oob_base_init.c | 3 -
orte/mca/oob/base/oob_base_recv.c | 4 +-
orte/mca/oob/base/oob_base_recv_nb.c | 6 +-
orte/mca/oob/base/oob_base_xcast.c | 2 +-
orte/mca/oob/oob.h | 6 +-
orte/mca/oob/oob_types.h | 19 -
orte/mca/oob/tcp/oob_tcp.c | 27 +-
orte/mca/oob/tcp/oob_tcp.h | 6 +-
orte/mca/oob/tcp/oob_tcp_addr.c | 7 +-
orte/mca/oob/tcp/oob_tcp_msg.c | 15 +-
orte/mca/oob/tcp/oob_tcp_peer.c | 22 +-
orte/mca/oob/tcp/oob_tcp_ping.c | 5 +-
orte/mca/oob/tcp/oob_tcp_recv.c | 20 +-
orte/mca/oob/tcp/oob_tcp_send.c | 13 +-
orte/mca/pls/base/pls_base_dmn_registry_fns.c | 87 +-
orte/mca/pls/base/pls_base_receive.c | 43 +-
orte/mca/pls/base/pls_private.h | 2 +-
orte/mca/pls/bproc/pls_bproc.c | 29 +-
orte/mca/pls/bproc/pls_bproc.h | 11 +-
orte/mca/pls/bproc/pls_bproc_state.c | 6 +-
orte/mca/pls/cnos/pls_cnos.c | 16 +-
orte/mca/pls/gridengine/pls_gridengine.h | 6 +-
.../pls/gridengine/pls_gridengine_module.c | 12 +-
orte/mca/pls/pls.h | 6 +-
orte/mca/pls/poe/pls_poe_module.c | 12 +-
orte/mca/pls/proxy/pls_proxy.c | 24 +-
orte/mca/pls/proxy/pls_proxy.h | 6 +-
orte/mca/pls/rsh/pls_rsh.h | 7 +-
orte/mca/pls/rsh/pls_rsh_component.c | 5 +
orte/mca/pls/rsh/pls_rsh_module.c | 17 +-
orte/mca/pls/slurm/pls_slurm_module.c | 16 +-
orte/mca/pls/tm/pls_tm_module.c | 18 +-
orte/mca/ras/base/ras_base_receive.c | 4 +-
orte/mca/rds/base/rds_base_receive.c | 4 +-
orte/mca/rmaps/base/rmaps_base_receive.c | 4 +-
.../rmgr_data_type_packing_fns.c | 9 +-
orte/mca/rmgr/base/rmgr_base_attribute_fns.c | 20 +
orte/mca/rmgr/base/rmgr_base_receive.c | 12 +-
orte/mca/rmgr/base/rmgr_base_stage_gate.c | 24 +-
orte/mca/rmgr/proxy/rmgr_proxy.c | 12 +-
orte/mca/rmgr/rmgr.h | 2 +-
orte/mca/rmgr/urm/rmgr_urm.c | 11 +-
orte/mca/rml/base/rml_base_open.c | 2 -
orte/mca/rml/rml.h | 8 -
orte/mca/rml/rml_types.h | 15 -
orte/mca/schema/schema_types.h | 10 +
orte/mca/sds/bproc/sds_bproc_module.c | 2 +-
orte/mca/sds/env/sds_env_module.c | 4 +-
orte/mca/sds/pipe/sds_pipe_module.c | 4 +-
orte/mca/sds/singleton/sds_singleton_module.c | 5 +-
orte/mca/sds/slurm/sds_slurm_module.c | 2 +-
orte/mca/smr/base/smr_base_get_proc_state.c | 5 +-
orte/mca/smr/base/smr_base_set_proc_state.c | 12 +-
orte/runtime/orte_init_stage1.c | 12 +-
orte/runtime/orte_restart.c | 6 +-
orte/runtime/orte_setup_hnp.c | 4 +-
orte/test/unit/ns/Makefile | 12 +
orte/test/unit/ns/ns_peers.c | 204 ++++
orte/test/unit/ns/ns_string_fns.c | 157 +++
orte/tools/orted/orted.c | 8 +-
orte/tools/orteprobe/orteprobe.c | 2 +-
orte/tools/orterun/orterun.c | 33 +-
144 files changed, 6909 insertions(+), 3847 deletions(-)
create mode 100644 orte/mca/ns/base/ns_base_cell_fns.c
create mode 100644 orte/mca/ns/base/ns_base_diag_fns.c
create mode 100644 orte/mca/ns/base/ns_base_general_fns.c
create mode 100644 orte/mca/ns/base/ns_base_job_fns.c
delete mode 100644 orte/mca/ns/base/ns_base_local_fns.c
create mode 100644 orte/mca/ns/base/ns_base_vpid_name_fns.c
create mode 100644 orte/mca/ns/base/ns_private.h
rename orte/mca/ns/proxy/{src => }/ns_proxy.h (75%)
create mode 100644 orte/mca/ns/proxy/ns_proxy_cell_fns.c
rename orte/mca/ns/proxy/{src => }/ns_proxy_component.c (84%)
create mode 100644 orte/mca/ns/proxy/ns_proxy_diag_fns.c
create mode 100644 orte/mca/ns/proxy/ns_proxy_general_fns.c
create mode 100644 orte/mca/ns/proxy/ns_proxy_job_fns.c
delete mode 100644 orte/mca/ns/proxy/src/Makefile.extra
delete mode 100644 orte/mca/ns/proxy/src/ns_proxy.c
rename orte/mca/ns/replica/{src => }/ns_replica.c (67%)
rename orte/mca/ns/replica/{src => }/ns_replica.h (50%)
create mode 100644 orte/mca/ns/replica/ns_replica_cell_fns.c
create mode 100644 orte/mca/ns/replica/ns_replica_class_instances.h
create mode 100644 orte/mca/ns/replica/ns_replica_component.c
create mode 100644 orte/mca/ns/replica/ns_replica_diag_fns.c
create mode 100644 orte/mca/ns/replica/ns_replica_general_fns.c
create mode 100644 orte/mca/ns/replica/ns_replica_job_fns.c
create mode 100644 orte/mca/ns/replica/ns_replica_recv.c
create mode 100644 orte/mca/ns/replica/ns_replica_support_fns.c
delete mode 100644 orte/mca/ns/replica/src/Makefile.extra
delete mode 100644 orte/mca/ns/replica/src/ns_replica_component.c
create mode 100644 orte/test/unit/ns/Makefile
create mode 100644 orte/test/unit/ns/ns_peers.c
create mode 100644 orte/test/unit/ns/ns_string_fns.c
diff --git a/ompi/attribute/attribute_predefined.c b/ompi/attribute/attribute_predefined.c
index 23226db6c5..fb1134ff75 100644
--- a/ompi/attribute/attribute_predefined.c
+++ b/ompi/attribute/attribute_predefined.c
@@ -307,9 +307,7 @@ void ompi_attr_create_predefined_callback(
/* Set some default values */
- if (ORTE_SUCCESS != orte_ns.get_jobid(&job, orte_process_info.my_name)) {
- return;
- }
+ job = ORTE_PROC_MY_NAME->jobid;
/* Query the gpr to find out how many CPUs there will be.
This will only return a non-empty list in a persistent
diff --git a/ompi/communicator/comm.c b/ompi/communicator/comm.c
index 9febed8bab..ded4517b36 100644
--- a/ompi/communicator/comm.c
+++ b/ompi/communicator/comm.c
@@ -1040,7 +1040,7 @@ int ompi_comm_determine_first ( ompi_communicator_t *intercomm, int high )
theirproc = intercomm->c_remote_group->grp_proc_pointers[0];
mask = ORTE_NS_CMP_CELLID | ORTE_NS_CMP_JOBID | ORTE_NS_CMP_VPID;
- rc = orte_ns.compare (mask, &(ourproc->proc_name), &(theirproc->proc_name));
+ rc = orte_ns.compare_fields(mask, &(ourproc->proc_name), &(theirproc->proc_name));
if ( 0 > rc ) {
flag = true;
}
diff --git a/ompi/communicator/comm_dyn.c b/ompi/communicator/comm_dyn.c
index 8118a76c58..c49d2496d5 100644
--- a/ompi/communicator/comm_dyn.c
+++ b/ompi/communicator/comm_dyn.c
@@ -312,7 +312,7 @@ orte_process_name_t *ompi_comm_get_rport (orte_process_name_t *port, int send_fi
if (NULL == rbuf) {
return NULL;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer(ORTE_RML_NAME_ANY, rbuf, tag))) {
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer(ORTE_NAME_WILDCARD, rbuf, tag))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(rbuf);
return NULL;
@@ -565,6 +565,16 @@ ompi_comm_start_processes(int count, char **array_of_commands,
/* cleanup */
if (NULL != base_prefix) free(base_prefix);
+ /* tell the RTE that we want to be a child of this process' job */
+ if (ORTE_SUCCESS != (rc = orte_rmgr.add_attribute(&attributes, ORTE_NS_USE_PARENT,
+ ORTE_JOBID, &(orte_process_info.my_name->jobid),
+ ORTE_RMGR_ATTR_OVERRIDE))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&attributes);
+ opal_progress_event_decrement();
+ return MPI_ERR_SPAWN;
+ }
+
/* tell the RTE that we want to the children to run inside of our allocation -
* don't go get one just for them
*/
@@ -857,7 +867,7 @@ void ompi_comm_disconnect_waitall (int count, ompi_comm_disconnect_obj **objs)
#define OMPI_COMM_MAXJOBIDS 64
void ompi_comm_mark_dyncomm (ompi_communicator_t *comm)
{
- int i, j, numjobids=0, rc;
+ int i, j, numjobids=0;
int size, rsize;
int found;
orte_jobid_t jobids[OMPI_COMM_MAXJOBIDS], thisjobid;
@@ -875,40 +885,34 @@ void ompi_comm_mark_dyncomm (ompi_communicator_t *comm)
of different jobids. */
grp = comm->c_local_group;
for (i=0; i< size; i++) {
- if (ORTE_SUCCESS != (rc = orte_ns.get_jobid(&thisjobid, &(grp->grp_proc_pointers[i]->proc_name)))) {
- ORTE_ERROR_LOG(rc);
- return;
- }
- found = 0;
- for ( j=0; j<numjobids; j++) {
- if ( thisjobid == jobids[j]) {
- found = 1;
- break;
+ thisjobid = grp->grp_proc_pointers[i]->proc_name.jobid;
+ found = 0;
+ for ( j=0; j<numjobids; j++) {
+ if (thisjobid == jobids[j]) {
+ found = 1;
+ break;
+ }
+ }
+ if (!found ) {
+ jobids[numjobids++] = thisjobid;
}
- }
- if (!found ) {
- jobids[numjobids++] = thisjobid;
- }
}
/* if inter-comm, loop over all processes in remote_group
and count number of different jobids */
grp = comm->c_remote_group;
for (i=0; i< rsize; i++) {
- if (ORTE_SUCCESS != (rc = orte_ns.get_jobid(&thisjobid, &(grp->grp_proc_pointers[i]->proc_name)))) {
- ORTE_ERROR_LOG(rc);
- return;
- }
- found = 0;
- for ( j=0; j<numjobids; j++) {
- if ( thisjobid == jobids[j]) {
- found = 1;
- break;
+ thisjobid = grp->grp_proc_pointers[i]->proc_name.jobid;
+ found = 0;
+ for ( j=0; j<numjobids; j++) {
+ if ( thisjobid == jobids[j]) {
+ found = 1;
+ break;
+ }
+ }
+ if (!found ) {
+ jobids[numjobids++] = thisjobid;
}
- }
- if (!found ) {
- jobids[numjobids++] = thisjobid;
- }
}
/* if number of joibds larger than one, set the disconnect flag*/
diff --git a/ompi/communicator/communicator.h b/ompi/communicator/communicator.h
index dc709b8826..607d00cba5 100644
--- a/ompi/communicator/communicator.h
+++ b/ompi/communicator/communicator.h
@@ -27,7 +27,7 @@
#include "mpi.h"
#include "ompi/group/group.h"
#include "ompi/mca/coll/coll.h"
-#include "orte/mca/oob/oob_types.h"
+#include "orte/mca/rml/rml_types.h"
#include "ompi/proc/proc.h"
#if defined(c_plusplus) || defined(__cplusplus)
diff --git a/ompi/mca/btl/mvapi/btl_mvapi_endpoint.c b/ompi/mca/btl/mvapi/btl_mvapi_endpoint.c
index 4f069b28a1..ca65ddb55b 100644
--- a/ompi/mca/btl/mvapi/btl_mvapi_endpoint.c
+++ b/ompi/mca/btl/mvapi/btl_mvapi_endpoint.c
@@ -579,7 +579,7 @@ static void mca_btl_mvapi_endpoint_recv(
opal_list_get_end(&mca_btl_mvapi_component.ib_procs);
ib_proc = (mca_btl_mvapi_proc_t*)opal_list_get_next(ib_proc)) {
- if(orte_ns.compare(ORTE_NS_CMP_ALL, &ib_proc->proc_guid, endpoint) == 0) {
+ if(orte_ns.compare_fields(ORTE_NS_CMP_ALL, &ib_proc->proc_guid, endpoint) == ORTE_EQUAL) {
bool found = false;
/* Try to get the endpoint instance of this proc */
@@ -690,7 +690,7 @@ static void mca_btl_mvapi_endpoint_recv(
void mca_btl_mvapi_post_recv()
{
orte_rml.recv_buffer_nb(
- ORTE_RML_NAME_ANY,
+ ORTE_NAME_WILDCARD,
ORTE_RML_TAG_DYNAMIC-1,
ORTE_RML_PERSISTENT,
mca_btl_mvapi_endpoint_recv,
diff --git a/ompi/mca/btl/openib/btl_openib_endpoint.c b/ompi/mca/btl/openib/btl_openib_endpoint.c
index cb317a9f42..49a46f2cd5 100644
--- a/ompi/mca/btl/openib/btl_openib_endpoint.c
+++ b/ompi/mca/btl/openib/btl_openib_endpoint.c
@@ -662,7 +662,7 @@ static void mca_btl_openib_endpoint_recv(
opal_list_get_end(&mca_btl_openib_component.ib_procs);
ib_proc = (mca_btl_openib_proc_t*)opal_list_get_next(ib_proc)) {
- if(orte_ns.compare(ORTE_NS_CMP_ALL, &ib_proc->proc_guid, endpoint) == 0) {
+ if(orte_ns.compare_fields(ORTE_NS_CMP_ALL, &ib_proc->proc_guid, endpoint) == ORTE_EQUAL) {
bool found = false;
/* Try to get the endpoint instance of this proc */
@@ -776,7 +776,7 @@ void mca_btl_openib_post_recv()
{
orte_rml.recv_buffer_nb(
- ORTE_RML_NAME_ANY,
+ ORTE_NAME_WILDCARD,
ORTE_RML_TAG_DYNAMIC-1,
ORTE_RML_PERSISTENT,
mca_btl_openib_endpoint_recv,
diff --git a/ompi/mca/btl/tcp/btl_tcp_endpoint.c b/ompi/mca/btl/tcp/btl_tcp_endpoint.c
index 1a5c38267e..d7ca834001 100644
--- a/ompi/mca/btl/tcp/btl_tcp_endpoint.c
+++ b/ompi/mca/btl/tcp/btl_tcp_endpoint.c
@@ -307,7 +307,7 @@ bool mca_btl_tcp_endpoint_accept(mca_btl_base_endpoint_t* btl_endpoint, struct s
if((btl_addr = btl_endpoint->endpoint_addr) != NULL &&
btl_addr->addr_inet.s_addr == addr->sin_addr.s_addr) {
mca_btl_tcp_proc_t *endpoint_proc = btl_endpoint->endpoint_proc;
- cmpval = orte_ns.compare(mask,
+ cmpval = orte_ns.compare_fields(mask,
&endpoint_proc->proc_ompi->proc_name,
&this_proc->proc_ompi->proc_name);
if((btl_endpoint->endpoint_sd < 0) ||
diff --git a/ompi/mca/btl/udapl/btl_udapl_endpoint.c b/ompi/mca/btl/udapl/btl_udapl_endpoint.c
index f996fe9d79..a4cac125e2 100644
--- a/ompi/mca/btl/udapl/btl_udapl_endpoint.c
+++ b/ompi/mca/btl/udapl/btl_udapl_endpoint.c
@@ -207,7 +207,7 @@ void mca_btl_udapl_endpoint_recv(int status, orte_process_name_t* endpoint,
opal_list_get_end(&mca_btl_udapl_component.udapl_procs);
proc = (mca_btl_udapl_proc_t*)opal_list_get_next(proc)) {
- if(0 == orte_ns.compare(ORTE_NS_CMP_ALL, &proc->proc_guid, endpoint)) {
+ if(ORTE_EQUAL == orte_ns.compare_fields(ORTE_NS_CMP_ALL, &proc->proc_guid, endpoint)) {
for(i = 0; i < proc->proc_endpoint_count; i++) {
ep = proc->proc_endpoints[i];
@@ -231,7 +231,7 @@ void mca_btl_udapl_endpoint_recv(int status, orte_process_name_t* endpoint,
void mca_btl_udapl_endpoint_post_oob_recv(void)
{
- orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY, ORTE_RML_TAG_DYNAMIC-1,
+ orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DYNAMIC-1,
ORTE_RML_PERSISTENT, mca_btl_udapl_endpoint_recv, NULL);
}
@@ -246,7 +246,7 @@ void mca_btl_udapl_endpoint_connect(mca_btl_udapl_endpoint_t* endpoint)
/* Nasty test to prevent deadlock and unwanted connection attempts */
/* This right here is the whole point of using the ORTE/RML handshake */
if((MCA_BTL_UDAPL_CONN_EAGER == endpoint->endpoint_state &&
- 0 > orte_ns.compare(ORTE_NS_CMP_ALL,
+ 0 > orte_ns.compare_fields(ORTE_NS_CMP_ALL,
&endpoint->endpoint_proc->proc_guid,
&ompi_proc_local()->proc_name)) ||
(MCA_BTL_UDAPL_CLOSED != endpoint->endpoint_state &&
@@ -370,7 +370,7 @@ static int mca_btl_udapl_endpoint_finish_eager(
OPAL_THREAD_UNLOCK(&endpoint->endpoint_lock);
/* Only one side does dat_ep_connect() */
- if(0 < orte_ns.compare(ORTE_NS_CMP_ALL,
+ if(0 < orte_ns.compare_fields(ORTE_NS_CMP_ALL,
&endpoint->endpoint_proc->proc_guid,
&ompi_proc_local()->proc_name)) {
diff --git a/ompi/mca/coll/sm/coll_sm_module.c b/ompi/mca/coll/sm/coll_sm_module.c
index bbbb955a90..1a6134de57 100644
--- a/ompi/mca/coll/sm/coll_sm_module.c
+++ b/ompi/mca/coll/sm/coll_sm_module.c
@@ -626,7 +626,7 @@ static int bootstrap_comm(ompi_communicator_t *comm)
empty_index = -1;
for (i = 0; i < mca_coll_sm_component.sm_bootstrap_num_segments; ++i) {
if (comm->c_contextid == bshe->smbhe_keys[i].mcsbck_cid &&
- 0 == orte_ns.compare(ORTE_NS_CMP_ALL,
+ ORTE_EQUAL == orte_ns.compare_fields(ORTE_NS_CMP_ALL,
rank0,
&bshe->smbhe_keys[i].mcsbck_rank0_name)) {
found = true;
diff --git a/ompi/mca/pml/base/pml_base_module_exchange.c b/ompi/mca/pml/base/pml_base_module_exchange.c
index 48e42aa53f..f527e164cc 100644
--- a/ompi/mca/pml/base/pml_base_module_exchange.c
+++ b/ompi/mca/pml/base/pml_base_module_exchange.c
@@ -426,10 +426,7 @@ static int mca_pml_base_modex_subscribe(orte_process_name_t* name)
OPAL_UNLOCK(&mca_pml_base_modex_lock);
/* otherwise - subscribe to get this jobid's contact info */
- if (ORTE_SUCCESS != (rc = orte_ns.get_jobid(&jobid, name))) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
+ jobid = name->jobid;
if (ORTE_SUCCESS != (rc = orte_schema.get_std_subscription_name(&sub_name,
OMPI_MODEX_SUBSCRIPTION, jobid))) {
@@ -520,10 +517,7 @@ int mca_pml_base_modex_send(
orte_byte_object_t bo;
orte_data_value_t value = ORTE_DATA_VALUE_EMPTY;
- if (ORTE_SUCCESS != (rc = orte_ns.get_jobid(&jobid, orte_process_info.my_name))) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
+ jobid = ORTE_PROC_MY_NAME->jobid;
if (ORTE_SUCCESS != (rc = orte_schema.get_job_segment_name(&segment, jobid))) {
ORTE_ERROR_LOG(rc);
diff --git a/ompi/mca/pml/dr/pml_dr.c b/ompi/mca/pml/dr/pml_dr.c
index d3a69a575a..5081d10692 100644
--- a/ompi/mca/pml/dr/pml_dr.c
+++ b/ompi/mca/pml/dr/pml_dr.c
@@ -172,9 +172,9 @@ int mca_pml_dr_add_procs(ompi_proc_t** procs, size_t nprocs)
processes, but will work for initial job start */
idx = ompi_pointer_array_add(&mca_pml_dr.endpoints,
(void*) endpoint);
- if(orte_ns.compare(ORTE_NS_CMP_ALL,
+ if(orte_ns.compare_fields(ORTE_NS_CMP_ALL,
orte_process_info.my_name,
- &(endpoint->proc_ompi->proc_name)) == 0) {
+ &(endpoint->proc_ompi->proc_name)) == ORTE_EQUAL) {
mca_pml_dr.my_rank = idx;
}
endpoint->local = endpoint->dst = idx;
diff --git a/ompi/proc/proc.c b/ompi/proc/proc.c
index 66d3139063..dba746c458 100644
--- a/ompi/proc/proc.c
+++ b/ompi/proc/proc.c
@@ -104,7 +104,7 @@ void ompi_proc_destruct(ompi_proc_t* proc)
int ompi_proc_init(void)
{
orte_process_name_t *peers;
- orte_std_cntr_t i, npeers, self, num_tokens;
+ orte_std_cntr_t i, npeers, num_tokens;
orte_jobid_t jobid;
char *segment, **tokens;
orte_data_value_t value = { {OBJ_CLASS(orte_data_value_t),0}, ORTE_NULL, NULL};
@@ -115,7 +115,7 @@ int ompi_proc_init(void)
OBJ_CONSTRUCT(&ompi_proc_lock, opal_mutex_t);
/* get all peers in this job */
- if(ORTE_SUCCESS != (rc = orte_ns.get_peers(&peers, &npeers, &self))) {
+ if(ORTE_SUCCESS != (rc = orte_ns.get_peers(&peers, &npeers, NULL))) {
opal_output(0, "ompi_proc_init: get_peers failed with errno=%d", rc);
return rc;
}
@@ -124,7 +124,7 @@ int ompi_proc_init(void)
for( i = 0; i < npeers; i++ ) {
ompi_proc_t *proc = OBJ_NEW(ompi_proc_t);
proc->proc_name = peers[i];
- if( i == self ) {
+ if( i == ORTE_PROC_MY_NAME->vpid ) {
ompi_proc_local_proc = proc;
proc->proc_flags |= OMPI_PROC_FLAG_LOCAL;
}
@@ -150,10 +150,7 @@ int ompi_proc_init(void)
return rc;
}
- if (ORTE_SUCCESS != (rc = orte_ns.get_jobid(&jobid, orte_process_info.my_name))) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
+ jobid = ORTE_PROC_MY_NAME->jobid;
/* find the job segment on the registry */
if (ORTE_SUCCESS != (rc = orte_schema.get_job_segment_name(&segment, jobid))) {
@@ -223,7 +220,7 @@ ompi_proc_t** ompi_proc_world(size_t *size)
for (proc = (ompi_proc_t*)opal_list_get_first(&ompi_proc_list);
proc != (ompi_proc_t*)opal_list_get_end(&ompi_proc_list);
proc = (ompi_proc_t*)opal_list_get_next(proc)) {
- if (0 == orte_ns.compare(mask, &proc->proc_name, &my_name)) {
+ if (ORTE_EQUAL == orte_ns.compare_fields(mask, &proc->proc_name, &my_name)) {
++count;
}
}
@@ -239,7 +236,7 @@ ompi_proc_t** ompi_proc_world(size_t *size)
for (proc = (ompi_proc_t*)opal_list_get_first(&ompi_proc_list);
proc != (ompi_proc_t*)opal_list_get_end(&ompi_proc_list);
proc = (ompi_proc_t*)opal_list_get_next(proc)) {
- if (0 == orte_ns.compare(mask, &proc->proc_name, &my_name)) {
+ if (ORTE_EQUAL == orte_ns.compare_fields(mask, &proc->proc_name, &my_name)) {
procs[count++] = proc;
}
}
@@ -298,7 +295,7 @@ ompi_proc_t * ompi_proc_find ( const orte_process_name_t * name )
for(proc = (ompi_proc_t*)opal_list_get_first(&ompi_proc_list);
proc != (ompi_proc_t*)opal_list_get_end(&ompi_proc_list);
proc = (ompi_proc_t*)opal_list_get_next(proc)) {
- if (0 == orte_ns.compare(mask, &proc->proc_name, name)) {
+ if (ORTE_EQUAL == orte_ns.compare_fields(mask, &proc->proc_name, name)) {
rproc = proc;
break;
}
@@ -319,7 +316,7 @@ ompi_proc_t * ompi_proc_find_and_add ( const orte_process_name_t * name, bool* i
for(proc = (ompi_proc_t*)opal_list_get_first(&ompi_proc_list);
proc != (ompi_proc_t*)opal_list_get_end(&ompi_proc_list);
proc = (ompi_proc_t*)opal_list_get_next(proc)) {
- if (0 == orte_ns.compare(mask, &proc->proc_name, name)) {
+ if (ORTE_EQUAL == orte_ns.compare_fields(mask, &proc->proc_name, name)) {
*isnew = false;
rproc = proc;
break;
@@ -395,10 +392,7 @@ static int setup_registry_callback(void)
orte_gpr_subscription_id_t id;
orte_jobid_t jobid;
- if (ORTE_SUCCESS != (rc = orte_ns.get_jobid(&jobid, &local->proc_name))) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
+ jobid = local->proc_name.jobid;
/* find the job segment on the registry */
if (ORTE_SUCCESS !=
@@ -534,7 +528,7 @@ static void callback(orte_gpr_notify_data_t *data, void *cbdata)
/* find the associated proc entry and update its
arch flag. If the nodename of this info is
my local host, also set the LOCAL flag. */
- if (0 == orte_ns.compare(mask, &name, &proc->proc_name)) {
+ if (ORTE_EQUAL == orte_ns.compare_fields(mask, &name, &proc->proc_name)) {
proc->proc_arch = arch;
if (0 == strcmp(str, orte_system_info.nodename)) {
proc->proc_flags |= OMPI_PROC_FLAG_LOCAL;
diff --git a/ompi/runtime/ompi_mpi_abort.c b/ompi/runtime/ompi_mpi_abort.c
index a8f872fef6..d3a8da6aae 100644
--- a/ompi/runtime/ompi_mpi_abort.c
+++ b/ompi/runtime/ompi_mpi_abort.c
@@ -133,7 +133,7 @@ ompi_mpi_abort(struct ompi_communicator_t* comm,
/* put all the local procs in the abort list */
for (i = 0 ; i < ompi_comm_size(comm) ; ++i) {
- if (0 != orte_ns.compare(ORTE_NS_CMP_ALL,
+ if (ORTE_EQUAL != orte_ns.compare_fields(ORTE_NS_CMP_ALL,
&comm->c_local_group->grp_proc_pointers[i]->proc_name,
orte_process_info.my_name)) {
assert(count <= nabort_procs);
@@ -147,7 +147,7 @@ ompi_mpi_abort(struct ompi_communicator_t* comm,
/* if requested, kill off remote procs too */
if (kill_remote_of_intercomm) {
for (i = 0 ; i < ompi_comm_remote_size(comm) ; ++i) {
- if (0 != orte_ns.compare(ORTE_NS_CMP_ALL,
+ if (ORTE_EQUAL != orte_ns.compare_fields(ORTE_NS_CMP_ALL,
&comm->c_remote_group->grp_proc_pointers[i]->proc_name,
orte_process_info.my_name)) {
assert(count <= nabort_procs);
diff --git a/ompi/runtime/ompi_mpi_init.c b/ompi/runtime/ompi_mpi_init.c
index 11a15bfae2..c616c45e3a 100644
--- a/ompi/runtime/ompi_mpi_init.c
+++ b/ompi/runtime/ompi_mpi_init.c
@@ -294,7 +294,7 @@ int ompi_mpi_init(int argc, char **argv, int requested, int *provided)
}
if (!set) {
char *vpid;
- orte_ns_base_get_vpid_string(&vpid, orte_process_info.my_name);
+ orte_ns.get_vpid_string(&vpid, orte_process_info.my_name);
opal_show_help("help-mpi-runtime",
"mpi_init:startup:paffinity-unavailable",
true, vpid);
diff --git a/ompi/runtime/ompi_mpi_io.c b/ompi/runtime/ompi_mpi_io.c
index cfba440bb4..15e6e928f7 100644
--- a/ompi/runtime/ompi_mpi_io.c
+++ b/ompi/runtime/ompi_mpi_io.c
@@ -17,7 +17,7 @@
*/
#include "ompi_config.h"
-#include "orte/mca/oob/oob.h"
+#include "orte/mca/ns/ns_types.h"
#include "orte/mca/iof/iof.h"
#include "ompi/constants.h"
@@ -40,7 +40,7 @@ int ompi_mpi_init_io(void)
close(fds[0]);
rc = mca_iof.iof_publish(
- MCA_OOB_NAME_SELF,
+ OMPI_PROC_MY_NAME,
MCA_IOF_SINK,
MCA_IOF_STDIN,
fds[1]);
@@ -56,7 +56,7 @@ int ompi_mpi_init_io(void)
close(fds[1]);
rc = mca_iof.iof_publish(
- MCA_OOB_NAME_SELF,
+ OMPI_PROC_MY_NAME,
MCA_IOF_SOURCE,
MCA_IOF_STDOUT,
fds[0]);
@@ -72,7 +72,7 @@ int ompi_mpi_init_io(void)
close(fds[1]);
rc = mca_iof.iof_publish(
- MCA_OOB_NAME_SELF,
+ OMPI_PROC_MY_NAME,
MCA_IOF_SOURCE,
MCA_IOF_STDERR,
fds[0]);
diff --git a/orte/dss/dss_pack.c b/orte/dss/dss_pack.c
index 13c8fb3284..aa5d0784fd 100644
--- a/orte/dss/dss_pack.c
+++ b/orte/dss/dss_pack.c
@@ -36,7 +36,7 @@ int orte_dss_pack(orte_buffer_t *buffer, void *src, orte_std_cntr_t num_vals,
int rc;
/* check for error */
- if (NULL == buffer || NULL == src) {
+ if (NULL == buffer) {
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
return ORTE_ERR_BAD_PARAM;
}
diff --git a/orte/include/orte/orte_constants.h b/orte/include/orte/orte_constants.h
index 4c5e1fd6b8..9d03b3d642 100644
--- a/orte/include/orte/orte_constants.h
+++ b/orte/include/orte/orte_constants.h
@@ -29,8 +29,8 @@ extern "C" {
#define ORTE_ERR_BASE OPAL_ERR_MAX
/* define the results values for comparisons so we can change them in only one place */
-#define ORTE_VALUE1_GREATER -1
-#define ORTE_VALUE2_GREATER +1
+#define ORTE_VALUE1_GREATER +1
+#define ORTE_VALUE2_GREATER -1
#define ORTE_EQUAL 0
enum {
diff --git a/orte/include/orte/orte_types.h b/orte/include/orte/orte_types.h
index 133f8f4413..f816090924 100644
--- a/orte/include/orte/orte_types.h
+++ b/orte/include/orte/orte_types.h
@@ -38,6 +38,7 @@ typedef uint8_t orte_data_type_t; /** data type indicators used in ORTE */
typedef int32_t orte_std_cntr_t; /** standard counters used in ORTE */
#define ORTE_STD_CNTR_T ORTE_INT32
#define ORTE_STD_CNTR_MAX INT32_MAX
+#define ORTE_STD_CNTR_MIN INT32_MIN
#define ORTE_STD_CNTR_INVALID -1
/* define a structure to hold generic byte objects */
diff --git a/orte/mca/errmgr/base/errmgr_base_receive.c b/orte/mca/errmgr/base/errmgr_base_receive.c
index b972ff9054..4800c486e1 100644
--- a/orte/mca/errmgr/base/errmgr_base_receive.c
+++ b/orte/mca/errmgr/base/errmgr_base_receive.c
@@ -50,7 +50,7 @@ int orte_errmgr_base_comm_start(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY,
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_ERRMGR,
ORTE_RML_PERSISTENT,
orte_errmgr_base_recv,
@@ -70,7 +70,7 @@ int orte_errmgr_base_comm_stop(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_ERRMGR))) {
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_ERRMGR))) {
ORTE_ERROR_LOG(rc);
}
recv_issued = false;
diff --git a/orte/mca/errmgr/hnp/errmgr_hnp.c b/orte/mca/errmgr/hnp/errmgr_hnp.c
index 1ee8e85d5e..ab422a05f2 100644
--- a/orte/mca/errmgr/hnp/errmgr_hnp.c
+++ b/orte/mca/errmgr/hnp/errmgr_hnp.c
@@ -23,6 +23,7 @@
#include <stdlib.h>
#include <stdarg.h>
+#include "opal/class/opal_list.h"
#include "opal/util/trace.h"
#include "opal/util/output.h"
@@ -59,6 +60,8 @@ int orte_errmgr_hnp_proc_aborted(orte_gpr_notify_message_t *msg)
NULL
};
orte_data_value_t dval = ORTE_DATA_VALUE_EMPTY;
+ opal_list_t attrs;
+ opal_list_item_t *item;
int rc;
OPAL_TRACE(1);
@@ -79,11 +82,15 @@ int orte_errmgr_hnp_proc_aborted(orte_gpr_notify_message_t *msg)
return rc;
}
- /* tell the pls to terminate the job */
- if (ORTE_SUCCESS != (rc = orte_pls.terminate_job(job))) {
+ /* tell the pls to terminate the job AND ALL ITS DESCENDANTS */
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ orte_rmgr.add_attribute(&attrs, ORTE_NS_INCLUDE_DESCENDANTS, ORTE_UNDEF, NULL, ORTE_RMGR_ATTR_OVERRIDE);
+ if (ORTE_SUCCESS != (rc = orte_pls.terminate_job(job, &attrs))) {
ORTE_ERROR_LOG(rc);
return rc;
}
+ while (NULL != (item = opal_list_remove_first(&attrs))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attrs);
/* orterun will only wakeup when all procs report terminated. The terminate_job
* function *should* have done that - however, it is possible during abnormal
@@ -142,8 +149,10 @@ int orte_errmgr_hnp_incomplete_start(orte_gpr_notify_message_t *msg)
return rc;
}
- /* tell the pls to terminate the job */
- if (ORTE_SUCCESS != (rc = orte_pls.terminate_job(job))) {
+ /* tell the pls to terminate the job - just kill this job, not any descendants since
+ * the job is just trying to start
+ */
+ if (ORTE_SUCCESS != (rc = orte_pls.terminate_job(job, NULL))) {
ORTE_ERROR_LOG(rc);
}
diff --git a/orte/mca/gpr/proxy/gpr_proxy_component.c b/orte/mca/gpr/proxy/gpr_proxy_component.c
index 5b187d2d03..7746a4dad6 100644
--- a/orte/mca/gpr/proxy/gpr_proxy_component.c
+++ b/orte/mca/gpr/proxy/gpr_proxy_component.c
@@ -233,7 +233,7 @@ orte_gpr_proxy_component_init(bool *allow_multi_user_threads, bool *have_hidden_
ORTE_ERROR_LOG(ret);
return NULL;
}
- if(ORTE_SUCCESS != (ret = orte_ns.copy_process_name(&orte_process_info.gpr_replica, &name))) {
+ if(ORTE_SUCCESS != (ret = orte_dss.copy((void**)&orte_process_info.gpr_replica, &name, ORTE_NAME))) {
ORTE_ERROR_LOG(ret);
return NULL;
}
@@ -299,7 +299,7 @@ int orte_gpr_proxy_module_init(void)
{
/* issue the non-blocking receive */
int rc;
- rc = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY, ORTE_RML_TAG_GPR_NOTIFY, ORTE_RML_PERSISTENT, orte_gpr_proxy_notify_recv, NULL);
+ rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_GPR_NOTIFY, ORTE_RML_PERSISTENT, orte_gpr_proxy_notify_recv, NULL);
if(rc < 0) {
ORTE_ERROR_LOG(rc);
return rc;
@@ -359,7 +359,7 @@ int orte_gpr_proxy_finalize(void)
}
/* All done */
- orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_GPR_NOTIFY);
+ orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_GPR_NOTIFY);
return ORTE_SUCCESS;
}
diff --git a/orte/mca/gpr/replica/api_layer/gpr_replica_deliver_notify_msg_api.c b/orte/mca/gpr/replica/api_layer/gpr_replica_deliver_notify_msg_api.c
index 9c783d8f4d..61d8bfff75 100755
--- a/orte/mca/gpr/replica/api_layer/gpr_replica_deliver_notify_msg_api.c
+++ b/orte/mca/gpr/replica/api_layer/gpr_replica_deliver_notify_msg_api.c
@@ -30,6 +30,7 @@
#include "orte/orte_constants.h"
#include "opal/util/trace.h"
+#include "orte/dss/dss.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/gpr/replica/api_layer/gpr_replica_api.h"
diff --git a/orte/mca/gpr/replica/functional_layer/gpr_replica_cleanup_fn.c b/orte/mca/gpr/replica/functional_layer/gpr_replica_cleanup_fn.c
index f557f8b12c..1098c3f043 100644
--- a/orte/mca/gpr/replica/functional_layer/gpr_replica_cleanup_fn.c
+++ b/orte/mca/gpr/replica/functional_layer/gpr_replica_cleanup_fn.c
@@ -88,9 +88,7 @@ int orte_gpr_replica_cleanup_proc_fn(orte_process_name_t *proc)
}
/* find the job segment */
- if (ORTE_SUCCESS != orte_ns.get_jobid(&jobid, proc)) {
- return ORTE_ERR_BAD_PARAM;
- }
+ jobid = proc->jobid;
if (ORTE_SUCCESS != orte_ns.convert_jobid_to_string(&jobidstring, jobid)) {
return ORTE_ERR_BAD_PARAM;
diff --git a/orte/mca/gpr/replica/functional_layer/gpr_replica_messaging_fn.c b/orte/mca/gpr/replica/functional_layer/gpr_replica_messaging_fn.c
index a8d139da4c..774fb4ae34 100644
--- a/orte/mca/gpr/replica/functional_layer/gpr_replica_messaging_fn.c
+++ b/orte/mca/gpr/replica/functional_layer/gpr_replica_messaging_fn.c
@@ -387,9 +387,7 @@ int orte_gpr_replica_define_callback(orte_gpr_notify_msg_type_t msg_type,
if (((NULL == recipient && NULL == cb->requestor) &&
(msg_type == cb->message->msg_type)) ||
(((NULL != recipient && NULL != cb->requestor) &&
- (0 == orte_ns.compare(ORTE_NS_CMP_ALL,
- recipient,
- cb->requestor))) &&
+ (ORTE_EQUAL == orte_dss.compare(recipient, cb->requestor, ORTE_NAME))) &&
(msg_type == cb->message->msg_type))) {
/* okay, a callback has been registered to send data to this
* recipient - return this location
@@ -421,7 +419,7 @@ int orte_gpr_replica_define_callback(orte_gpr_notify_msg_type_t msg_type,
if (NULL == recipient) {
cb->requestor = NULL;
} else {
- if (ORTE_SUCCESS != (rc = orte_ns.copy_process_name(&(cb->requestor), recipient))) {
+ if (ORTE_SUCCESS != (rc = orte_dss.copy((void**)&(cb->requestor), recipient, ORTE_NAME))) {
ORTE_ERROR_LOG(rc);
return rc;
}
diff --git a/orte/mca/gpr/replica/functional_layer/gpr_replica_trig_ops_fn.c b/orte/mca/gpr/replica/functional_layer/gpr_replica_trig_ops_fn.c
index 16ee7f5b4e..7339df0415 100644
--- a/orte/mca/gpr/replica/functional_layer/gpr_replica_trig_ops_fn.c
+++ b/orte/mca/gpr/replica/functional_layer/gpr_replica_trig_ops_fn.c
@@ -216,8 +216,13 @@ ADDREQ:
i < (sub->requestors)->size; i++) {
if (NULL != reqs[i]) {
j++;
+ if ((NULL == reqs[i]->requestor && NULL != requestor) ||
+ (NULL != reqs[i]->requestor && NULL == requestor)) {
+ continue;
+ }
if (reqs[i]->idtag == subscription->id &&
- 0 == orte_ns.compare(ORTE_NS_CMP_ALL, reqs[i]->requestor, requestor)) {
+ ((NULL == reqs[i]->requestor && NULL == requestor) ||
+ (ORTE_EQUAL == orte_dss.compare(reqs[i]->requestor, requestor, ORTE_NAME)))) {
/* found this requestor - do not add it again */
goto DONESUB;
}
@@ -234,8 +239,7 @@ ADDREQ:
}
if (NULL != requestor) {
- if (ORTE_SUCCESS != (rc = orte_ns.copy_process_name(&(req->requestor),
- requestor))) {
+ if (ORTE_SUCCESS != (rc = orte_dss.copy((void**)&(req->requestor), requestor, ORTE_NAME))) {
ORTE_ERROR_LOG(rc);
return rc;
}
@@ -546,8 +550,14 @@ ADDREQ:
i < (trig->attached)->size; i++) {
if (NULL != reqs[i]) {
j++;
+ /* if one is NULL and the other isn't, then they can't possibly match */
+ if ((NULL == reqs[i]->requestor && NULL != requestor) ||
+ (NULL != reqs[i]->requestor && NULL == requestor)) {
+ continue;
+ }
if (reqs[i]->idtag == trigger->id &&
- 0 == orte_ns.compare(ORTE_NS_CMP_ALL, reqs[i]->requestor, requestor)) {
+ ((NULL == reqs[i]->requestor && NULL == requestor) ||
+ (ORTE_EQUAL == orte_dss.compare(reqs[i]->requestor, requestor, ORTE_NAME)))) {
/* found this requestor - do not add it again */
goto DONETRIG;
}
@@ -562,8 +572,7 @@ ADDREQ:
}
if (NULL != requestor) {
- if (ORTE_SUCCESS != (rc = orte_ns.copy_process_name(&(req->requestor),
- requestor))) {
+ if (ORTE_SUCCESS != (rc = orte_dss.copy((void**)&(req->requestor), requestor, ORTE_NAME))) {
ORTE_ERROR_LOG(rc);
return rc;
}
@@ -657,8 +666,7 @@ orte_gpr_replica_remove_subscription(orte_process_name_t *requestor,
if (id == reqs[k]->idtag &&
((NULL == requestor && NULL == reqs[k]->requestor) ||
(NULL != requestor && NULL != reqs[k]->requestor &&
- 0 == orte_ns.compare(ORTE_NS_CMP_ALL,
- reqs[k]->requestor, requestor)))) {
+ ORTE_EQUAL == orte_dss.compare(reqs[k]->requestor, requestor, ORTE_NAME)))) {
/* this is the subscription */
sub = subs[i];
req = reqs[k];
@@ -763,8 +771,7 @@ orte_gpr_replica_remove_trigger(orte_process_name_t *requestor,
if (id == reqs[k]->idtag &&
((NULL == requestor && NULL == reqs[k]->requestor) ||
(NULL != requestor && NULL != reqs[k]->requestor &&
- 0 == orte_ns.compare(ORTE_NS_CMP_ALL,
- reqs[k]->requestor, requestor)))) {
+ ORTE_EQUAL == orte_dss.compare(reqs[k]->requestor, requestor, ORTE_NAME)))) {
/* this is the trigger */
trig = trigs[i];
req = reqs[k];
@@ -1330,7 +1337,7 @@ int orte_gpr_replica_purge_subscriptions(orte_process_name_t *proc)
}
OBJ_RELEASE(trig);
} else if (NULL != proc && NULL != trig[i]->requestor &&
- 0 == orte_ns.compare(ORTE_NS_CMP_ALL, proc, trig[i]->requestor)) {
+ ORTE_EQUAL == orte_dss.compare(Oproc, trig[i]->requestor, ORTE_NAME)) {
if (ORTE_SUCCESS != (rc = orte_pointer_array_set_item(orte_gpr_replica.triggers,
trig[i]->index, NULL))) {
ORTE_ERROR_LOG(rc);
diff --git a/orte/mca/gpr/replica/gpr_replica_component.c b/orte/mca/gpr/replica/gpr_replica_component.c
index 1363ef9524..c14975d484 100644
--- a/orte/mca/gpr/replica/gpr_replica_component.c
+++ b/orte/mca/gpr/replica/gpr_replica_component.c
@@ -311,7 +311,7 @@ int orte_gpr_replica_module_init(void)
/* issue the non-blocking receive */
if (!orte_gpr_replica_globals.isolate) {
int rc = orte_rml.recv_buffer_nb(
- ORTE_RML_NAME_ANY, ORTE_RML_TAG_GPR, ORTE_RML_PERSISTENT, orte_gpr_replica_recv, NULL);
+ ORTE_NAME_WILDCARD, ORTE_RML_TAG_GPR, ORTE_RML_PERSISTENT, orte_gpr_replica_recv, NULL);
if(rc < 0) {
ORTE_ERROR_LOG(rc);
return rc;
@@ -437,7 +437,7 @@ int orte_gpr_replica_finalize(void)
return ORTE_SUCCESS;
}
- orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_GPR);
+ orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_GPR);
return ORTE_SUCCESS;
}
diff --git a/orte/mca/iof/base/iof_base_close.c b/orte/mca/iof/base/iof_base_close.c
index b7da7c2230..ea42475c0c 100644
--- a/orte/mca/iof/base/iof_base_close.c
+++ b/orte/mca/iof/base/iof_base_close.c
@@ -59,7 +59,7 @@ int orte_iof_base_close(void)
OPAL_THREAD_UNLOCK(&orte_iof_base.iof_lock);
if (NULL != orte_iof_base.iof_service) {
- orte_ns.free_name(&(orte_iof_base.iof_service));
+ free(orte_iof_base.iof_service);
}
return ORTE_SUCCESS;
diff --git a/orte/mca/iof/base/iof_base_endpoint.c b/orte/mca/iof/base/iof_base_endpoint.c
index f3659408ea..96cf3224d8 100644
--- a/orte/mca/iof/base/iof_base_endpoint.c
+++ b/orte/mca/iof/base/iof_base_endpoint.c
@@ -42,7 +42,10 @@
#include <signal.h>
#endif /* HAVE_SIGNAL_H */
#include "opal/util/output.h"
+
+#include "orte/mca/ns/ns.h"
#include "orte/mca/rml/rml.h"
+
#include "orte/mca/iof/base/base.h"
#include "orte/mca/iof/base/iof_base_endpoint.h"
#include "orte/mca/iof/base/iof_base_fragment.h"
@@ -172,7 +175,7 @@ static void orte_iof_base_endpoint_read_handler(int fd, short flags, void *cbdat
hdr = &frag->frag_hdr;
hdr->hdr_common.hdr_type = ORTE_IOF_BASE_HDR_MSG;
hdr->hdr_msg.msg_src = endpoint->ep_name;
- hdr->hdr_msg.msg_proxy = *ORTE_RML_NAME_SELF;
+ hdr->hdr_msg.msg_proxy = *ORTE_PROC_MY_NAME;
hdr->hdr_msg.msg_tag = endpoint->ep_tag;
hdr->hdr_msg.msg_seq = endpoint->ep_seq;
hdr->hdr_msg.msg_len = frag->frag_len;
@@ -294,7 +297,7 @@ static orte_iof_base_endpoint_t* orte_iof_base_endpoint_lookup(
item != opal_list_get_end(&orte_iof_base.iof_endpoints);
item = opal_list_get_next(item)) {
orte_iof_base_endpoint_t* endpoint = (orte_iof_base_endpoint_t*)item;
- if(orte_ns.compare(ORTE_NS_CMP_ALL,proc,&endpoint->ep_name) == 0 &&
+ if(orte_ns.compare_fields(ORTE_NS_CMP_ALL,proc,&endpoint->ep_name) == 0 &&
endpoint->ep_tag == tag && endpoint->ep_mode == mode) {
OBJ_RETAIN(endpoint);
return endpoint;
@@ -428,7 +431,7 @@ int orte_iof_base_endpoint_delete(
while(item != opal_list_get_end(&orte_iof_base.iof_endpoints)) {
opal_list_item_t* next = opal_list_get_next(item);
orte_iof_base_endpoint_t* endpoint = (orte_iof_base_endpoint_t*)item;
- if(orte_ns.compare(mask,proc,&endpoint->ep_name) == 0 &&
+ if(orte_ns.compare_fields(mask,proc,&endpoint->ep_name) == 0 &&
endpoint->ep_tag == tag) {
OBJ_RELEASE(endpoint);
opal_list_remove_item(&orte_iof_base.iof_endpoints,&endpoint->super);
@@ -485,7 +488,7 @@ orte_iof_base_endpoint_t* orte_iof_base_endpoint_match(
item != opal_list_get_end(&orte_iof_base.iof_endpoints);
item = opal_list_get_next(item)) {
orte_iof_base_endpoint_t* endpoint = (orte_iof_base_endpoint_t*)item;
- if(orte_ns.compare(dst_mask,dst_name,&endpoint->ep_name) == 0) {
+ if(orte_ns.compare_fields(dst_mask,dst_name,&endpoint->ep_name) == 0) {
if(endpoint->ep_tag == dst_tag || endpoint->ep_tag == ORTE_IOF_ANY || dst_tag == ORTE_IOF_ANY) {
OBJ_RETAIN(endpoint);
OPAL_THREAD_UNLOCK(&orte_iof_base.iof_lock);
diff --git a/orte/mca/iof/proxy/iof_proxy.c b/orte/mca/iof/proxy/iof_proxy.c
index 7972454d26..5d27ea1eac 100644
--- a/orte/mca/iof/proxy/iof_proxy.c
+++ b/orte/mca/iof/proxy/iof_proxy.c
@@ -144,7 +144,7 @@ int orte_iof_proxy_push(
/* send a subscription to server on behalf of the destination */
rc = orte_iof_proxy_svc_subscribe(
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_NS_CMP_ALL,
dst_tag,
dst_name,
@@ -156,7 +156,7 @@ int orte_iof_proxy_push(
/* setup a local endpoint to reflect registration */
rc = orte_iof_base_endpoint_create(
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_IOF_SOURCE,
dst_tag,
fd);
@@ -184,7 +184,7 @@ int orte_iof_proxy_pull(
/* setup a local endpoint */
int rc;
rc = orte_iof_base_endpoint_create(
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_IOF_SINK,
src_tag,
fd);
@@ -195,7 +195,7 @@ int orte_iof_proxy_pull(
/* publish this endpoint */
rc = orte_iof_proxy_svc_publish(
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
src_tag);
if(rc != ORTE_SUCCESS) {
ORTE_ERROR_LOG(rc);
@@ -207,7 +207,7 @@ int orte_iof_proxy_pull(
src_name,
src_mask,
src_tag,
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_NS_CMP_ALL,
src_tag);
if(rc != ORTE_SUCCESS) {
@@ -247,7 +247,7 @@ int orte_iof_proxy_subscribe(
int rc;
/* create a local registration to reflect the callback */
- rc = orte_iof_base_callback_create(ORTE_RML_NAME_SELF,src_tag,cbfunc,cbdata);
+ rc = orte_iof_base_callback_create(ORTE_PROC_MY_NAME,src_tag,cbfunc,cbdata);
if(rc != ORTE_SUCCESS)
return rc;
@@ -256,7 +256,7 @@ int orte_iof_proxy_subscribe(
src_name,
src_mask,
src_tag,
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_NS_CMP_ALL,
src_tag);
return rc;
@@ -274,13 +274,13 @@ int orte_iof_proxy_unsubscribe(
src_name,
src_mask,
src_tag,
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_NS_CMP_ALL,
src_tag);
if(rc != ORTE_SUCCESS)
return rc;
/* remove local callback */
- return orte_iof_base_callback_delete(ORTE_RML_NAME_SELF,src_tag);
+ return orte_iof_base_callback_delete(ORTE_PROC_MY_NAME,src_tag);
}
diff --git a/orte/mca/iof/proxy/iof_proxy_component.c b/orte/mca/iof/proxy/iof_proxy_component.c
index da43b1273c..f6ac8f8f98 100644
--- a/orte/mca/iof/proxy/iof_proxy_component.c
+++ b/orte/mca/iof/proxy/iof_proxy_component.c
@@ -127,7 +127,7 @@ orte_iof_proxy_init(int* priority, bool *allow_multi_user_threads, bool *have_hi
mca_iof_proxy_component.proxy_iov[0].iov_len = 0;
rc = orte_rml.recv_nb(
- ORTE_RML_NAME_ANY,
+ ORTE_NAME_WILDCARD,
mca_iof_proxy_component.proxy_iov,
1,
ORTE_RML_TAG_IOF_SVC,
@@ -152,7 +152,7 @@ static int orte_iof_proxy_close(void)
int rc = ORTE_SUCCESS;
if (initialized) {
- rc = orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_IOF_SVC);
+ rc = orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_IOF_SVC);
}
return rc;
}
diff --git a/orte/mca/iof/proxy/iof_proxy_svc.c b/orte/mca/iof/proxy/iof_proxy_svc.c
index 00e543cc82..1456bf405a 100644
--- a/orte/mca/iof/proxy/iof_proxy_svc.c
+++ b/orte/mca/iof/proxy/iof_proxy_svc.c
@@ -58,7 +58,7 @@ int orte_iof_proxy_svc_publish(
hdr.hdr_common.hdr_type = ORTE_IOF_BASE_HDR_PUB;
hdr.hdr_common.hdr_status = 0;
hdr.hdr_pub.pub_name = *name;
- hdr.hdr_pub.pub_proxy = *ORTE_RML_NAME_SELF;
+ hdr.hdr_pub.pub_proxy = *ORTE_PROC_MY_NAME;
hdr.hdr_pub.pub_mask = ORTE_NS_CMP_ALL;
hdr.hdr_pub.pub_tag = tag;
ORTE_IOF_BASE_HDR_PUB_NTOH(hdr.hdr_pub);
@@ -96,7 +96,7 @@ int orte_iof_proxy_svc_unpublish(
hdr.hdr_common.hdr_type = ORTE_IOF_BASE_HDR_PUB;
hdr.hdr_common.hdr_status = 0;
hdr.hdr_pub.pub_name = *name;
- hdr.hdr_pub.pub_proxy = *ORTE_RML_NAME_SELF;
+ hdr.hdr_pub.pub_proxy = *ORTE_PROC_MY_NAME;
hdr.hdr_pub.pub_mask = mask;
hdr.hdr_pub.pub_tag = tag;
ORTE_IOF_BASE_HDR_PUB_NTOH(hdr.hdr_pub);
@@ -257,7 +257,7 @@ static void orte_iof_proxy_svc_msg(
unsigned char* data)
{
orte_iof_base_endpoint_t* endpoint;
- endpoint = orte_iof_base_endpoint_match(ORTE_RML_NAME_ANY, ORTE_NS_CMP_NONE, msg->msg_tag);
+ endpoint = orte_iof_base_endpoint_match(ORTE_NAME_WILDCARD, ORTE_NS_CMP_NONE, msg->msg_tag);
if(endpoint != NULL) {
orte_iof_base_endpoint_forward(endpoint,src,msg,data);
OBJ_RELEASE(endpoint);
diff --git a/orte/mca/iof/svc/iof_svc.c b/orte/mca/iof/svc/iof_svc.c
index f60db9c878..7fbedc943c 100644
--- a/orte/mca/iof/svc/iof_svc.c
+++ b/orte/mca/iof/svc/iof_svc.c
@@ -80,7 +80,7 @@ int orte_iof_svc_publish(
if(mode == ORTE_IOF_SINK) {
rc = orte_iof_svc_pub_create(
name,
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_NS_CMP_ALL,
tag);
}
@@ -106,7 +106,7 @@ int orte_iof_svc_unpublish(
int rc;
rc = orte_iof_svc_pub_delete(
name,
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
mask,
tag);
if(rc != ORTE_SUCCESS)
@@ -142,7 +142,7 @@ int orte_iof_svc_push(
/* setup a subscription */
rc = orte_iof_svc_sub_create(
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_NS_CMP_ALL,
dst_tag,
dst_name,
@@ -153,7 +153,7 @@ int orte_iof_svc_push(
/* setup a local endpoint to reflect registration */
rc = orte_iof_base_endpoint_create(
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_IOF_SOURCE,
dst_tag,
fd);
@@ -181,7 +181,7 @@ int orte_iof_svc_pull(
/* setup a local endpoint */
rc = orte_iof_base_endpoint_create(
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_IOF_SINK,
src_tag,
fd);
@@ -193,7 +193,7 @@ int orte_iof_svc_pull(
src_name,
src_mask,
src_tag,
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_NS_CMP_ALL,
src_tag);
return rc;
@@ -230,7 +230,7 @@ int orte_iof_svc_subscribe(
int rc;
/* create a local registration to reflect the callback */
- rc = orte_iof_base_callback_create(ORTE_RML_NAME_SELF,src_tag,cbfunc,cbdata);
+ rc = orte_iof_base_callback_create(ORTE_PROC_MY_NAME,src_tag,cbfunc,cbdata);
if(rc != ORTE_SUCCESS)
return rc;
@@ -239,7 +239,7 @@ int orte_iof_svc_subscribe(
src_name,
src_mask,
src_tag,
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_NS_CMP_ALL,
src_tag);
return rc;
@@ -257,12 +257,12 @@ int orte_iof_svc_unsubscribe(
src_name,
src_mask,
src_tag,
- ORTE_RML_NAME_SELF,
+ ORTE_PROC_MY_NAME,
ORTE_NS_CMP_ALL,
src_tag);
if(ORTE_SUCCESS != rc)
return rc;
/* cleanup any locally registered callback */
- return orte_iof_base_callback_delete(ORTE_RML_NAME_SELF,src_tag);
+ return orte_iof_base_callback_delete(ORTE_PROC_MY_NAME,src_tag);
}
diff --git a/orte/mca/iof/svc/iof_svc_component.c b/orte/mca/iof/svc/iof_svc_component.c
index 16a9d980c1..121347c6f7 100644
--- a/orte/mca/iof/svc/iof_svc_component.c
+++ b/orte/mca/iof/svc/iof_svc_component.c
@@ -120,7 +120,7 @@ static int orte_iof_svc_close(void)
OBJ_RELEASE(item);
}
OPAL_THREAD_UNLOCK(&mca_iof_svc_component.svc_lock);
- orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_IOF_SVC);
+ orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_IOF_SVC);
}
return ORTE_SUCCESS;
@@ -164,7 +164,7 @@ orte_iof_svc_init(int* priority, bool *allow_multi_user_threads, bool *have_hidd
mca_iof_svc_component.svc_iov[0].iov_len = 0;
rc = orte_rml.recv_nb(
- ORTE_RML_NAME_ANY,
+ ORTE_NAME_WILDCARD,
mca_iof_svc_component.svc_iov,
1,
ORTE_RML_TAG_IOF_SVC,
diff --git a/orte/mca/iof/svc/iof_svc_proxy.c b/orte/mca/iof/svc/iof_svc_proxy.c
index 37c2003522..ac1ad8b3c1 100644
--- a/orte/mca/iof/svc/iof_svc_proxy.c
+++ b/orte/mca/iof/svc/iof_svc_proxy.c
@@ -151,7 +151,7 @@ static void orte_iof_svc_proxy_msg(
continue;
/* source match */
- if(orte_ns.compare(sub->src_mask,&sub->src_name,&hdr->msg_src) == 0) {
+ if(orte_ns.compare_fields(sub->src_mask,&sub->src_name,&hdr->msg_src) == 0) {
if(mca_iof_svc_component.svc_debug > 1) {
opal_output(0, "[%lu,%lu,%lu] orte_iof_svc_proxy_msg: tag %d sequence %d\n",
ORTE_NAME_ARGS(&sub->src_name),hdr->msg_tag,hdr->msg_seq);
@@ -227,7 +227,7 @@ static void orte_iof_svc_proxy_ack(
orte_iof_svc_sub_t* sub = (orte_iof_svc_sub_t*)s_item;
opal_list_item_t *f_item;
- if (orte_ns.compare(sub->src_mask,&sub->src_name,&hdr->msg_src) != 0 ||
+ if (orte_ns.compare_fields(sub->src_mask,&sub->src_name,&hdr->msg_src) != 0 ||
sub->src_tag != hdr->msg_tag) {
continue;
}
@@ -238,8 +238,8 @@ static void orte_iof_svc_proxy_ack(
f_item = opal_list_get_next(f_item)) {
orte_iof_svc_fwd_t* fwd = (orte_iof_svc_fwd_t*)f_item;
orte_iof_svc_pub_t* pub = fwd->fwd_pub;
- if (orte_ns.compare(pub->pub_mask,&pub->pub_name,src) == 0 ||
- orte_ns.compare(ORTE_NS_CMP_ALL,&pub->pub_proxy,src) == 0) {
+ if (orte_ns.compare_fields(pub->pub_mask,&pub->pub_name,src) == 0 ||
+ orte_ns.compare_fields(ORTE_NS_CMP_ALL,&pub->pub_proxy,src) == 0) {
value.uval = hdr->msg_seq + hdr->msg_len;
orte_hash_table_set_proc(&fwd->fwd_seq,
&hdr->msg_src, &value.vval);
@@ -259,7 +259,7 @@ static void orte_iof_svc_proxy_ack(
*/
if(seq_min == hdr->msg_seq+hdr->msg_len) {
- if(orte_ns.compare(ORTE_NS_CMP_ALL,orte_process_info.my_name,&hdr->msg_src) == 0) {
+ if(orte_ns.compare_fields(ORTE_NS_CMP_ALL,orte_process_info.my_name,&hdr->msg_src) == 0) {
orte_iof_base_endpoint_t* endpoint;
/*
* Local delivery
diff --git a/orte/mca/iof/svc/iof_svc_pub.c b/orte/mca/iof/svc/iof_svc_pub.c
index 996e1fafa7..36739f67bd 100644
--- a/orte/mca/iof/svc/iof_svc_pub.c
+++ b/orte/mca/iof/svc/iof_svc_pub.c
@@ -48,8 +48,8 @@ int orte_iof_svc_pub_create(
item != opal_list_get_end(&mca_iof_svc_component.svc_published);
item = opal_list_get_next(item)) {
pub = (orte_iof_svc_pub_t*)item;
- if(orte_ns.compare(pub_mask,pub_name,&pub->pub_name) == 0 &&
- orte_ns.compare(ORTE_NS_CMP_ALL,pub_proxy,&pub->pub_proxy) == 0 &&
+ if(orte_ns.compare_fields(pub_mask,pub_name,&pub->pub_name) == 0 &&
+ orte_ns.compare_fields(ORTE_NS_CMP_ALL,pub_proxy,&pub->pub_proxy) == 0 &&
pub_tag == pub->pub_tag) {
OPAL_THREAD_UNLOCK(&mca_iof_svc_component.svc_lock);
return ORTE_SUCCESS;
@@ -96,8 +96,8 @@ orte_iof_svc_pub_t* orte_iof_svc_pub_lookup(
item != opal_list_get_end(&mca_iof_svc_component.svc_published);
item = opal_list_get_next(item)) {
orte_iof_svc_pub_t* pub = (orte_iof_svc_pub_t*)item;
- if (orte_ns.compare(ORTE_NS_CMP_ALL, &pub->pub_name,pub_name) == 0 &&
- orte_ns.compare(ORTE_NS_CMP_ALL, &pub->pub_proxy,pub_proxy) == 0 &&
+ if (orte_ns.compare_fields(ORTE_NS_CMP_ALL, &pub->pub_name,pub_name) == 0 &&
+ orte_ns.compare_fields(ORTE_NS_CMP_ALL, &pub->pub_proxy,pub_proxy) == 0 &&
pub->pub_mask == pub_mask &&
pub->pub_tag == pub_tag) {
return pub;
@@ -157,8 +157,8 @@ void orte_iof_svc_pub_delete_all(
opal_list_item_t* p_next = opal_list_get_next(p_item);
orte_iof_svc_pub_t* pub = (orte_iof_svc_pub_t*)p_item;
- if (orte_ns.compare(ORTE_NS_CMP_ALL, &pub->pub_name,name) == 0 ||
- orte_ns.compare(ORTE_NS_CMP_ALL, &pub->pub_proxy,name) == 0) {
+ if (orte_ns.compare_fields(ORTE_NS_CMP_ALL, &pub->pub_name,name) == 0 ||
+ orte_ns.compare_fields(ORTE_NS_CMP_ALL, &pub->pub_proxy,name) == 0) {
opal_list_item_t* s_item;
for(s_item = opal_list_get_first(&mca_iof_svc_component.svc_subscribed);
diff --git a/orte/mca/iof/svc/iof_svc_sub.c b/orte/mca/iof/svc/iof_svc_sub.c
index 103da39460..f2f7f39805 100644
--- a/orte/mca/iof/svc/iof_svc_sub.c
+++ b/orte/mca/iof/svc/iof_svc_sub.c
@@ -81,10 +81,10 @@ int orte_iof_svc_sub_create(
item = opal_list_get_next(item)) {
sub = (orte_iof_svc_sub_t*)item;
if (sub->src_mask == src_mask &&
- orte_ns.compare(sub->src_mask,&sub->src_name,src_name) == 0 &&
+ orte_ns.compare_fields(sub->src_mask,&sub->src_name,src_name) == 0 &&
sub->src_tag == src_tag &&
sub->dst_mask == dst_mask &&
- orte_ns.compare(sub->dst_mask,&sub->dst_name,dst_name) == 0 &&
+ orte_ns.compare_fields(sub->dst_mask,&sub->dst_name,dst_name) == 0 &&
sub->dst_tag == dst_tag) {
OPAL_THREAD_UNLOCK(&mca_iof_svc_component.svc_lock);
return ORTE_SUCCESS;
@@ -134,10 +134,10 @@ int orte_iof_svc_sub_delete(
opal_list_item_t* next = opal_list_get_next(item);
orte_iof_svc_sub_t* sub = (orte_iof_svc_sub_t*)item;
if (sub->src_mask == src_mask &&
- orte_ns.compare(sub->src_mask,&sub->src_name,src_name) == 0 &&
+ orte_ns.compare_fields(sub->src_mask,&sub->src_name,src_name) == 0 &&
sub->src_tag == src_tag &&
sub->dst_mask == dst_mask &&
- orte_ns.compare(sub->dst_mask,&sub->dst_name,dst_name) == 0 &&
+ orte_ns.compare_fields(sub->dst_mask,&sub->dst_name,dst_name) == 0 &&
sub->dst_tag == dst_tag) {
opal_list_remove_item(&mca_iof_svc_component.svc_subscribed, item);
OBJ_RELEASE(item);
@@ -159,9 +159,9 @@ int orte_iof_svc_sub_delete_all(
opal_list_item_t* next = opal_list_get_next(item);
orte_iof_svc_sub_t* sub = (orte_iof_svc_sub_t*)item;
if ((sub->src_mask == ORTE_NS_CMP_ALL &&
- orte_ns.compare(ORTE_NS_CMP_ALL,&sub->src_name,name) == 0) ||
+ orte_ns.compare_fields(ORTE_NS_CMP_ALL,&sub->src_name,name) == 0) ||
(sub->dst_mask == ORTE_NS_CMP_ALL &&
- orte_ns.compare(ORTE_NS_CMP_ALL,&sub->dst_name,name) == 0)) {
+ orte_ns.compare_fields(ORTE_NS_CMP_ALL,&sub->dst_name,name) == 0)) {
opal_list_remove_item(&mca_iof_svc_component.svc_subscribed, item);
OBJ_RELEASE(item);
}
@@ -283,7 +283,7 @@ bool orte_iof_svc_fwd_match(
orte_iof_svc_sub_t* sub,
orte_iof_svc_pub_t* pub)
{
- if (orte_ns.compare(sub->dst_mask,&sub->dst_name,&pub->pub_name) == 0 &&
+ if (orte_ns.compare_fields(sub->dst_mask,&sub->dst_name,&pub->pub_name) == 0 &&
sub->src_tag == pub->pub_tag) {
return true;
} else {
diff --git a/orte/mca/ns/base/Makefile.am b/orte/mca/ns/base/Makefile.am
index deed6eb0e6..9d99d64481 100644
--- a/orte/mca/ns/base/Makefile.am
+++ b/orte/mca/ns/base/Makefile.am
@@ -17,13 +17,18 @@
#
headers += \
- base/base.h
+ base/base.h \
+ base/ns_private.h
libmca_ns_la_SOURCES += \
base/ns_base_close.c \
base/ns_base_select.c \
base/ns_base_open.c \
- base/ns_base_local_fns.c \
+ base/ns_base_cell_fns.c \
+ base/ns_base_job_fns.c \
+ base/ns_base_vpid_name_fns.c \
+ base/ns_base_general_fns.c \
+ base/ns_base_diag_fns.c \
base/data_type_support/ns_data_type_compare_fns.c \
base/data_type_support/ns_data_type_copy_fns.c \
base/data_type_support/ns_data_type_print_fns.c \
diff --git a/orte/mca/ns/base/base.h b/orte/mca/ns/base/base.h
index 641e79a16a..262d977f18 100644
--- a/orte/mca/ns/base/base.h
+++ b/orte/mca/ns/base/base.h
@@ -42,38 +42,6 @@
extern "C" {
#endif
-/* default limits */
-#define ORTE_NS_ARRAY_MAX_SIZE INT_MAX
-#define ORTE_NS_ARRAY_BLOCK_SIZE 512
-/*
- * Internal definitions
- */
-typedef uint8_t orte_ns_cmd_bitmask_t;
-
-/*
- * packing type definitions
- */
-/* CAUTION - any changes here must also change corresponding
- * typedefs above
- */
-#define ORTE_NS_CMD ORTE_INT8
-
-/*
- * define flag values for remote commands - only used internally
- */
-#define ORTE_NS_CREATE_CELLID_CMD (int8_t)0x01
-#define ORTE_NS_GET_CELL_INFO_CMD (int8_t)0x02
-#define ORTE_NS_CREATE_JOBID_CMD (int8_t)0x03
-#define ORTE_NS_RESERVE_RANGE_CMD (int8_t)0x04
-#define ORTE_NS_ASSIGN_OOB_TAG_CMD (int8_t)0x08
-#define ORTE_NS_GET_JOB_PEERS_CMD (int8_t)0x0A
-#define ORTE_NS_DEFINE_DATA_TYPE_CMD (int8_t)0x10
-#define ORTE_NS_CREATE_MY_NAME_CMD (int8_t)0x20
-#define ORTE_NS_DUMP_CELLS_CMD (int8_t)0x21
-#define ORTE_NS_DUMP_JOBIDS_CMD (int8_t)0x22
-#define ORTE_NS_DUMP_TAGS_CMD (int8_t)0x23
-#define ORTE_NS_DUMP_DATATYPES_CMD (int8_t)0x24
-
/*
* function definitions
@@ -82,176 +50,6 @@ ORTE_DECLSPEC int orte_ns_base_open(void);
ORTE_DECLSPEC int orte_ns_base_select(void);
ORTE_DECLSPEC int orte_ns_base_close(void);
- /*
- * Base functions that are common to all implementations - can be overridden
- */
-
-ORTE_DECLSPEC int orte_ns_base_assign_cellid_to_process(orte_process_name_t* name);
-
-ORTE_DECLSPEC int orte_ns_base_create_process_name(orte_process_name_t **name,
- orte_cellid_t cell,
- orte_jobid_t job,
- orte_vpid_t vpid);
-
-ORTE_DECLSPEC int orte_ns_base_copy_process_name(orte_process_name_t **dest,
- orte_process_name_t* src);
-
-ORTE_DECLSPEC int orte_ns_base_convert_string_to_process_name(orte_process_name_t **name,
- const char* name_string);
-
-ORTE_DECLSPEC int orte_ns_base_get_proc_name_string(char **name_string,
- const orte_process_name_t* name);
-
-ORTE_DECLSPEC int orte_ns_base_get_vpid_string(char **vpid_string, const orte_process_name_t* name);
-
-ORTE_DECLSPEC int orte_ns_base_convert_vpid_to_string(char **vpid_string, const orte_vpid_t vpid);
-
-ORTE_DECLSPEC int orte_ns_base_convert_string_to_vpid(orte_vpid_t *vpid, const char* vpidstring);
-
-ORTE_DECLSPEC int orte_ns_base_get_jobid_string(char **jobid_string, const orte_process_name_t* name);
-
-ORTE_DECLSPEC int orte_ns_base_convert_jobid_to_string(char **jobid_string, const orte_jobid_t jobid);
-
-ORTE_DECLSPEC int orte_ns_base_convert_string_to_jobid(orte_jobid_t *jobid, const char* jobidstring);
-
-ORTE_DECLSPEC int orte_ns_base_get_cellid_string(char **cellid_string, const orte_process_name_t* name);
-
-ORTE_DECLSPEC int orte_ns_base_convert_string_to_cellid(orte_cellid_t *cellid, const char *cellidstring);
-
-ORTE_DECLSPEC int orte_ns_base_convert_cellid_to_string(char **cellid_string, const orte_cellid_t cellid);
-
-ORTE_DECLSPEC int orte_ns_base_get_vpid(orte_vpid_t *vpid, const orte_process_name_t* name);
-
-ORTE_DECLSPEC int orte_ns_base_get_jobid(orte_jobid_t *jobid, const orte_process_name_t* name);
-
-ORTE_DECLSPEC int orte_ns_base_get_cellid(orte_cellid_t *cellid, const orte_process_name_t* name);
-
-ORTE_DECLSPEC int orte_ns_base_compare(orte_ns_cmp_bitmask_t fields,
- const orte_process_name_t* name1,
- const orte_process_name_t* name2);
-
-ORTE_DECLSPEC int orte_ns_base_free_name(orte_process_name_t **name);
-
-ORTE_DECLSPEC int orte_ns_base_print_dump(orte_buffer_t *buffer);
-
-
-/* not available functions */
-ORTE_DECLSPEC int orte_ns_base_module_init_not_available(void);
-
-ORTE_DECLSPEC int orte_ns_base_create_cellid_not_available(orte_cellid_t *cellid,
- char *site, char *resource);
-
-ORTE_DECLSPEC int orte_ns_base_get_cell_info_not_available(orte_cellid_t cellid,
- char **site, char **resource);
-
-ORTE_DECLSPEC int orte_ns_base_create_jobid_not_available(orte_jobid_t *jobid);
-
-ORTE_DECLSPEC int orte_ns_base_get_vpid_range_not_available(orte_jobid_t job,
- orte_vpid_t range,
- orte_vpid_t *startvpid);
-
-ORTE_DECLSPEC int orte_ns_base_derive_vpid(orte_vpid_t *vpid,
- orte_vpid_t base_vpid,
- int offset);
-
-ORTE_DECLSPEC int orte_ns_base_assign_rml_tag_not_available(orte_rml_tag_t *tag, char *name);
-
-ORTE_DECLSPEC int orte_ns_base_define_data_type_not_available(
- const char *name,
- orte_data_type_t *type);
-
-ORTE_DECLSPEC int orte_ns_base_create_my_name_not_available(void);
-
-ORTE_DECLSPEC int orte_ns_base_get_job_peers_not_available(orte_process_name_t **procs,
- orte_std_cntr_t *num_procs, orte_jobid_t job);
-
-ORTE_DECLSPEC int orte_ns_base_dump_cells_not_available(void);
-ORTE_DECLSPEC int orte_ns_base_dump_jobs_not_available(void);
-ORTE_DECLSPEC int orte_ns_base_dump_tags_not_available(void);
-ORTE_DECLSPEC int orte_ns_base_dump_datatypes_not_available(void);
-
-/* Base functions used everywhere */
-ORTE_DECLSPEC int orte_ns_base_get_peers(orte_process_name_t **procs,
- orte_std_cntr_t *num_procs, orte_std_cntr_t *self);
-
-ORTE_DECLSPEC int orte_ns_base_pack_name(orte_buffer_t *buffer, void *src,
- orte_std_cntr_t num_vals, orte_data_type_t type);
-
-ORTE_DECLSPEC int orte_ns_base_pack_cellid(orte_buffer_t *buffer, void *src,
- orte_std_cntr_t num_vals, orte_data_type_t type);
-
-ORTE_DECLSPEC int orte_ns_base_pack_jobid(orte_buffer_t *buffer, void *src,
- orte_std_cntr_t num_vals, orte_data_type_t type);
-
-ORTE_DECLSPEC int orte_ns_base_pack_vpid(orte_buffer_t *buffer, void *src,
- orte_std_cntr_t num_vals, orte_data_type_t type);
-
-ORTE_DECLSPEC int orte_ns_base_unpack_name(orte_buffer_t *buffer, void *dest,
- orte_std_cntr_t *num_vals, orte_data_type_t type);
-
-ORTE_DECLSPEC int orte_ns_base_unpack_cellid(orte_buffer_t *buffer, void *dest,
- orte_std_cntr_t *num_vals, orte_data_type_t type);
-
-ORTE_DECLSPEC int orte_ns_base_unpack_jobid(orte_buffer_t *buffer, void *dest,
- orte_std_cntr_t *num_vals, orte_data_type_t type);
-
-ORTE_DECLSPEC int orte_ns_base_unpack_vpid(orte_buffer_t *buffer, void *dest,
- orte_std_cntr_t *num_vals, orte_data_type_t type);
-
-/*
- * copy functions
- */
-
-int orte_ns_base_copy_name(orte_process_name_t **dest, orte_process_name_t *src, orte_data_type_t type);
-
-int orte_ns_base_copy_vpid(orte_vpid_t **dest, orte_vpid_t *src, orte_data_type_t type);
-
-int orte_ns_base_copy_cellid(orte_cellid_t **dest, orte_cellid_t *src, orte_data_type_t type);
-
-int orte_ns_base_copy_jobid(orte_jobid_t **dest, orte_jobid_t *src, orte_data_type_t type);
-
-/*
- * compare functions
- */
-
-int orte_ns_base_compare_name(orte_process_name_t *value1,
- orte_process_name_t *value2,
- orte_data_type_t type);
-
-
-int orte_ns_base_compare_vpid(orte_vpid_t *value1,
- orte_vpid_t *value2,
- orte_data_type_t type);
-
-int orte_ns_base_compare_jobid(orte_jobid_t *value1,
- orte_jobid_t *value2,
- orte_data_type_t type);
-
-int orte_ns_base_compare_cellid(orte_cellid_t *value1,
- orte_cellid_t *value2,
- orte_data_type_t type);
-
-/*
- * size functions
- */
-
-int orte_ns_base_std_size(size_t *size, void *src, orte_data_type_t type);
-
-/*
- * release functions
- */
-
-void orte_ns_base_std_release(orte_data_value_t *value);
-
-/*
- * print functions
- */
-
-int orte_ns_base_std_print(char **output, char *prefix, void *src, orte_data_type_t type);
-
-int orte_ns_base_print_name(char **output, char *prefix, orte_process_name_t *name, orte_data_type_t type);
-
-
/*
* globals that might be needed
*/
diff --git a/orte/mca/ns/base/data_type_support/ns_data_type_compare_fns.c b/orte/mca/ns/base/data_type_support/ns_data_type_compare_fns.c
index 65cceb317d..683f6f1ae7 100755
--- a/orte/mca/ns/base/data_type_support/ns_data_type_compare_fns.c
+++ b/orte/mca/ns/base/data_type_support/ns_data_type_compare_fns.c
@@ -25,7 +25,7 @@
#include "orte/dss/dss.h"
#include "orte/mca/errmgr/errmgr.h"
-#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/base/ns_private.h"
/*
* NUMERIC COMPARE FUNCTIONS
@@ -42,42 +42,13 @@ int orte_ns_base_compare_name(orte_process_name_t *value1,
return ORTE_VALUE1_GREATER;
}
- /** we have to take care of the special case where one of the
- * values is ORTE_NAME_WILDCARD. If any of the fields are wildcard,
- * then we want to just ignore that one field. However, in the case
- * of ORTE_NAME_WILDCARD (where ALL of the fields are wildcard), this
- * would automatically result in ORTE_EQUAL for any name in the other
- * value - a totally useless result.
- *
- * Instead, what we want to know in this case is if the value actually
- * *is* ORTE_NAME_WILDCARD. So, we need to detect if one of the values
- * is ORTE_NAME_WILDCARD, and then specifically check the other one
- * to see if it matches
- */
- if (value2->cellid == ORTE_CELLID_WILDCARD &&
- value2->jobid == ORTE_JOBID_WILDCARD &&
- value2->vpid == ORTE_VPID_WILDCARD) {
- if (value1->cellid == ORTE_CELLID_WILDCARD &&
- value1->jobid == ORTE_JOBID_WILDCARD &&
- value1->vpid == ORTE_VPID_WILDCARD) {
- return ORTE_EQUAL;
- } else {
- return ORTE_VALUE1_GREATER;
- }
- } else if (value1->cellid == ORTE_CELLID_WILDCARD &&
- value1->jobid == ORTE_JOBID_WILDCARD &&
- value1->vpid == ORTE_VPID_WILDCARD) {
- if (value2->cellid == ORTE_CELLID_WILDCARD &&
- value2->jobid == ORTE_JOBID_WILDCARD &&
- value2->vpid == ORTE_VPID_WILDCARD) {
- return ORTE_EQUAL;
- } else {
- return ORTE_VALUE2_GREATER;
- }
- }
-
- /** now that the special cases are done, go through the progression */
-
+ /* If any of the fields are wildcard,
+ * then we want to just ignore that one field. In the case
+ * of ORTE_NAME_WILDCARD (where ALL of the fields are wildcard), this
+ * will automatically result in ORTE_EQUAL for any name in the other
+ * value - a totally useless result, but consistent in behavior.
+ */
+
/** check the cellids - if one of them is WILDCARD, then ignore
* this field since anything is okay
*/
@@ -163,3 +134,18 @@ int orte_ns_base_compare_cellid(orte_cellid_t *value1,
return ORTE_EQUAL;
}
+
+int orte_ns_base_compare_nodeid(orte_nodeid_t *value1,
+ orte_nodeid_t *value2,
+ orte_data_type_t type)
+{
+ /** if either value is WILDCARD, then return equal */
+ if (*value1 == ORTE_NODEID_WILDCARD ||
+ *value2 == ORTE_NODEID_WILDCARD) return ORTE_EQUAL;
+
+ if (*value1 > *value2) return ORTE_VALUE1_GREATER;
+
+ if (*value2 > *value1) return ORTE_VALUE2_GREATER;
+
+ return ORTE_EQUAL;
+}
diff --git a/orte/mca/ns/base/data_type_support/ns_data_type_copy_fns.c b/orte/mca/ns/base/data_type_support/ns_data_type_copy_fns.c
index 01fc378998..9ddfad4a42 100755
--- a/orte/mca/ns/base/data_type_support/ns_data_type_copy_fns.c
+++ b/orte/mca/ns/base/data_type_support/ns_data_type_copy_fns.c
@@ -23,7 +23,7 @@
#include "orte/mca/errmgr/errmgr.h"
-#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/base/ns_private.h"
/*
* VPID
@@ -63,6 +63,25 @@ int orte_ns_base_copy_cellid(orte_cellid_t **dest, orte_cellid_t *src, orte_data
return ORTE_SUCCESS;
}
+/*
+ * NODEID
+ */
+int orte_ns_base_copy_nodeid(orte_nodeid_t **dest, orte_nodeid_t *src, orte_data_type_t type)
+{
+ orte_nodeid_t *val;
+
+ val = (orte_nodeid_t*)malloc(sizeof(orte_nodeid_t));
+ if (NULL == val) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ *val = *src;
+ *dest = val;
+
+ return ORTE_SUCCESS;
+}
+
/*
* JOBID
*/
diff --git a/orte/mca/ns/base/data_type_support/ns_data_type_packing_fns.c b/orte/mca/ns/base/data_type_support/ns_data_type_packing_fns.c
index fe9b00ff65..769d67e04f 100644
--- a/orte/mca/ns/base/data_type_support/ns_data_type_packing_fns.c
+++ b/orte/mca/ns/base/data_type_support/ns_data_type_packing_fns.c
@@ -28,7 +28,7 @@
#include "orte/mca/errmgr/errmgr.h"
#include "orte/dss/dss_internal.h"
-#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/base/ns_private.h"
/*
* NAME
@@ -51,11 +51,7 @@ int orte_ns_base_pack_name(orte_buffer_t *buffer, void *src,
}
proc = (orte_process_name_t*)src;
for (i=0; i < num_vals; i++) {
- if (ORTE_SUCCESS != (rc = orte_ns.get_cellid(&cellid[i], proc))) {
- ORTE_ERROR_LOG(rc);
- free(cellid);
- return rc;
- }
+ cellid[i] = proc->cellid;
proc++;
}
/* now pack them in one shot */
@@ -75,11 +71,7 @@ int orte_ns_base_pack_name(orte_buffer_t *buffer, void *src,
}
proc = (orte_process_name_t*)src;
for (i=0; i < num_vals; i++) {
- if (ORTE_SUCCESS != (rc = orte_ns.get_jobid(&jobid[i], proc))) {
- ORTE_ERROR_LOG(rc);
- free(jobid);
- return rc;
- }
+ jobid[i] = proc->jobid;
proc++;
}
/* now pack them in one shot */
@@ -99,11 +91,7 @@ int orte_ns_base_pack_name(orte_buffer_t *buffer, void *src,
}
proc = (orte_process_name_t*)src;
for (i=0; i < num_vals; i++) {
- if (ORTE_SUCCESS != (rc = orte_ns.get_vpid(&vpid[i], proc))) {
- ORTE_ERROR_LOG(rc);
- free(vpid);
- return rc;
- }
+ vpid[i] = proc->vpid;
proc++;
}
/* now pack them in one shot */
@@ -128,13 +116,30 @@ int orte_ns_base_pack_cellid(orte_buffer_t *buffer, void *src,
/* Turn around and pack the real type */
if (ORTE_SUCCESS != (
- ret = orte_dss_pack_buffer(buffer, src, num_vals, ORTE_STD_CNTR_T))) {
+ ret = orte_dss_pack_buffer(buffer, src, num_vals, ORTE_CELLID_T))) {
ORTE_ERROR_LOG(ret);
}
return ret;
}
+/*
+ * NODEID
+ */
+int orte_ns_base_pack_nodeid(orte_buffer_t *buffer, void *src,
+ orte_std_cntr_t num_vals, orte_data_type_t type)
+{
+ int ret;
+
+ /* Turn around and pack the real type */
+ if (ORTE_SUCCESS != (
+ ret = orte_dss_pack_buffer(buffer, src, num_vals, ORTE_NODEID_T))) {
+ ORTE_ERROR_LOG(ret);
+ }
+
+ return ret;
+}
+
/*
* JOBID
*/
@@ -145,7 +150,7 @@ int orte_ns_base_pack_jobid(orte_buffer_t *buffer, void *src,
/* Turn around and pack the real type */
if (ORTE_SUCCESS != (
- ret = orte_dss_pack_buffer(buffer, src, num_vals, ORTE_STD_CNTR_T))) {
+ ret = orte_dss_pack_buffer(buffer, src, num_vals, ORTE_JOBID_T))) {
ORTE_ERROR_LOG(ret);
}
@@ -162,7 +167,7 @@ int orte_ns_base_pack_vpid(orte_buffer_t *buffer, void *src,
/* Turn around and pack the real type */
if (ORTE_SUCCESS != (
- ret = orte_dss_pack_buffer(buffer, src, num_vals, ORTE_STD_CNTR_T))) {
+ ret = orte_dss_pack_buffer(buffer, src, num_vals, ORTE_VPID_T))) {
ORTE_ERROR_LOG(ret);
}
diff --git a/orte/mca/ns/base/data_type_support/ns_data_type_print_fns.c b/orte/mca/ns/base/data_type_support/ns_data_type_print_fns.c
index 65fcb263cb..552dbe9f2a 100755
--- a/orte/mca/ns/base/data_type_support/ns_data_type_print_fns.c
+++ b/orte/mca/ns/base/data_type_support/ns_data_type_print_fns.c
@@ -24,7 +24,7 @@
#include "orte/dss/dss.h"
#include "orte/mca/errmgr/errmgr.h"
-#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/base/ns_private.h"
static void orte_ns_base_quick_print(char **output, char *type_name, char *pfx, void *src, size_t src_size);
@@ -49,6 +49,10 @@ int orte_ns_base_std_print(char **output, char *prefix, void *src, orte_data_typ
orte_ns_base_quick_print(output, "ORTE_CELLID", prefix, src, sizeof(orte_cellid_t));
break;
+ case ORTE_NODEID:
+ orte_ns_base_quick_print(output, "ORTE_NODEID", prefix, src, sizeof(orte_nodeid_t));
+ break;
+
default:
ORTE_ERROR_LOG(ORTE_ERR_UNKNOWN_DATA_TYPE);
return ORTE_ERR_UNKNOWN_DATA_TYPE;
@@ -69,9 +73,9 @@ int orte_ns_base_print_name(char **output, char *prefix, orte_process_name_t *na
asprintf(output, "%sData type: ORTE_PROCESS_NAME\tData Value: NULL",
(NULL == prefix ? " " : prefix));
} else {
- asprintf(output, "%sData type: ORTE_PROCESS_NAME\tData Value: [%lu,%lu,%lu]",
- (NULL == prefix ? " " : prefix), (unsigned long)name->cellid,
- (unsigned long)name->jobid, (unsigned long)name->vpid);
+ asprintf(output, "%sData type: ORTE_PROCESS_NAME\tData Value: [%ld,%ld,%ld]",
+ (NULL == prefix ? " " : prefix), (long)name->cellid,
+ (long)name->jobid, (long)name->vpid);
}
return ORTE_SUCCESS;
@@ -80,10 +84,10 @@ int orte_ns_base_print_name(char **output, char *prefix, orte_process_name_t *na
static void orte_ns_base_quick_print(char **output, char *type_name, char *prefix, void *src, size_t src_size)
{
- uint8_t *ui8;
- uint16_t *ui16;
- uint32_t *ui32;
- uint64_t *ui64;
+ int8_t *i8;
+ int16_t *i16;
+ int32_t *i32;
+ int64_t *i64;
char *pfx;
/* set default result */
@@ -99,23 +103,23 @@ static void orte_ns_base_quick_print(char **output, char *type_name, char *prefi
switch(src_size) {
case 1:
- ui8 = (uint8_t*)src;
- asprintf(output, "%sData type: %s\tValue: %d", pfx, type_name, (int) *ui8);
+ i8 = (int8_t*)src;
+ asprintf(output, "%sData type: %s\tValue: %d", pfx, type_name, (int) *i8);
break;
case 2:
- ui16 = (uint16_t*)src;
- asprintf(output, "%sData type: %s\tValue: %d", pfx, type_name, (int) *ui16);
+ i16 = (int16_t*)src;
+ asprintf(output, "%sData type: %s\tValue: %d", pfx, type_name, (int) *i16);
break;
case 4:
- ui32 = (uint32_t*)src;
- asprintf(output, "%sData type: %s\tValue: %lu", pfx, type_name, (unsigned long) *ui32);
+ i32 = (int32_t*)src;
+ asprintf(output, "%sData type: %s\tValue: %ld", pfx, type_name, (long) *i32);
break;
case 8:
- ui64 = (uint64_t*)src;
- asprintf(output, "%sData type: %s\tValue: %lu", pfx, type_name, (unsigned long) *ui64);
+ i64 = (int64_t*)src;
+ asprintf(output, "%sData type: %s\tValue: %ld", pfx, type_name, (long) *i64);
break;
default:
diff --git a/orte/mca/ns/base/data_type_support/ns_data_type_release_fns.c b/orte/mca/ns/base/data_type_support/ns_data_type_release_fns.c
index 8fb49b3e4b..16ebe91117 100644
--- a/orte/mca/ns/base/data_type_support/ns_data_type_release_fns.c
+++ b/orte/mca/ns/base/data_type_support/ns_data_type_release_fns.c
@@ -24,7 +24,7 @@
#include "orte/dss/dss.h"
#include "orte/mca/errmgr/errmgr.h"
-#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/base/ns_private.h"
/*
* STANDARD RELEASE FUNCTION - WORKS FOR EVERYTHING NON-STRUCTURED
diff --git a/orte/mca/ns/base/data_type_support/ns_data_type_size_fns.c b/orte/mca/ns/base/data_type_support/ns_data_type_size_fns.c
index 5d298e23c3..2abbe6630e 100755
--- a/orte/mca/ns/base/data_type_support/ns_data_type_size_fns.c
+++ b/orte/mca/ns/base/data_type_support/ns_data_type_size_fns.c
@@ -24,7 +24,7 @@
#include "orte/dss/dss.h"
#include "orte/mca/errmgr/errmgr.h"
-#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/base/ns_private.h"
/*
* STANDARD SIZE FUNCTION - WORKS FOR EVERYTHING NON-STRUCTURED
@@ -44,6 +44,10 @@ int orte_ns_base_std_size(size_t *size, void *src, orte_data_type_t type)
*size = sizeof(orte_cellid_t);
break;
+ case ORTE_NODEID:
+ *size = sizeof(orte_nodeid_t);
+ break;
+
case ORTE_NAME:
*size = sizeof(orte_process_name_t);
break;
diff --git a/orte/mca/ns/base/data_type_support/ns_data_type_unpacking_fns.c b/orte/mca/ns/base/data_type_support/ns_data_type_unpacking_fns.c
index 27df2c64da..9506c04652 100644
--- a/orte/mca/ns/base/data_type_support/ns_data_type_unpacking_fns.c
+++ b/orte/mca/ns/base/data_type_support/ns_data_type_unpacking_fns.c
@@ -26,7 +26,7 @@
#include "orte/mca/errmgr/errmgr.h"
#include "orte/dss/dss_internal.h"
-#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/base/ns_private.h"
/*
* NAME
@@ -123,14 +123,29 @@ int orte_ns_base_unpack_cellid(orte_buffer_t *buffer, void *dest,
int ret;
/* Turn around and unpack the real type */
- if (ORTE_SUCCESS != (
- ret = orte_dss_unpack_buffer(buffer, dest, num_vals, ORTE_STD_CNTR_T))) {
+ if (ORTE_SUCCESS != (ret = orte_dss_unpack_buffer(buffer, dest, num_vals, ORTE_CELLID_T))) {
ORTE_ERROR_LOG(ret);
}
return ret;
}
+/*
+ * NODEID
+ */
+int orte_ns_base_unpack_nodeid(orte_buffer_t *buffer, void *dest,
+ orte_std_cntr_t *num_vals, orte_data_type_t type)
+{
+ int ret;
+
+ /* Turn around and unpack the real type */
+ if (ORTE_SUCCESS != (ret = orte_dss_unpack_buffer(buffer, dest, num_vals, ORTE_NODEID_T))) {
+ ORTE_ERROR_LOG(ret);
+ }
+
+ return ret;
+}
+
/*
* JOBID
*/
@@ -140,8 +155,7 @@ int orte_ns_base_unpack_jobid(orte_buffer_t *buffer, void *dest,
int ret;
/* Turn around and unpack the real type */
- if (ORTE_SUCCESS != (
- ret = orte_dss_unpack_buffer(buffer, dest, num_vals, ORTE_STD_CNTR_T))) {
+ if (ORTE_SUCCESS != (ret = orte_dss_unpack_buffer(buffer, dest, num_vals, ORTE_JOBID_T))) {
ORTE_ERROR_LOG(ret);
}
@@ -157,8 +171,7 @@ int orte_ns_base_unpack_vpid(orte_buffer_t *buffer, void *dest,
int ret;
/* Turn around and unpack the real type */
- if (ORTE_SUCCESS != (
- ret = orte_dss_unpack_buffer(buffer, dest, num_vals, ORTE_STD_CNTR_T))) {
+ if (ORTE_SUCCESS != (ret = orte_dss_unpack_buffer(buffer, dest, num_vals, ORTE_VPID_T))) {
ORTE_ERROR_LOG(ret);
}
diff --git a/orte/mca/ns/base/ns_base_cell_fns.c b/orte/mca/ns/base/ns_base_cell_fns.c
new file mode 100644
index 0000000000..d4f9d8b543
--- /dev/null
+++ b/orte/mca/ns/base/ns_base_cell_fns.c
@@ -0,0 +1,232 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+
+#include "orte_config.h"
+
+#include <stdio.h>
+#include <string.h>
+#include <stddef.h>
+#include <stdlib.h>
+#if HAVE_NETINET_IN_H
+#include <netinet/in.h>
+#endif
+
+#include "orte/orte_constants.h"
+
+#include "opal/util/output.h"
+#include "opal/util/printf.h"
+#include "opal/mca/mca.h"
+
+#include "orte/mca/schema/schema_types.h"
+#include "orte/mca/errmgr/errmgr.h"
+
+#include "orte/mca/ns/base/ns_private.h"
+
+/*
+ * "not available" functions
+ */
+int
+orte_ns_base_create_cellid_not_available(orte_cellid_t *cellid, char *site, char *resource)
+{
+ *cellid = ORTE_CELLID_INVALID;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int
+orte_ns_base_get_cell_info_not_available(orte_cellid_t cellid,
+ char **site, char **resource)
+{
+ *site = NULL;
+ *resource = NULL;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int
+orte_ns_base_create_nodeids_not_available(orte_nodeid_t **nodeids, orte_std_cntr_t *nnodes, orte_cellid_t cellid, char **nodename)
+{
+ *nodeids = NULL;
+ *nnodes = 0;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int
+orte_ns_base_get_node_info_not_available(char ***nodenames, orte_cellid_t cellid,
+ orte_std_cntr_t num_nodeids, orte_nodeid_t *nodeids)
+{
+ *nodenames = NULL;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+
+/**** CELL STRING FUNCTIONS ****/
+int orte_ns_base_get_cellid_string(char **cellid_string, const orte_process_name_t* name)
+{
+ if (NULL == name) { /* got an error */
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ *cellid_string = NULL;
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ /* check for wildcard value - handle appropriately */
+ if (ORTE_CELLID_WILDCARD == name->cellid) {
+ *cellid_string = strdup(ORTE_SCHEMA_WILDCARD_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ /* check for invalid value - handle appropriately */
+ if (ORTE_CELLID_INVALID == name->cellid) {
+ *cellid_string = strdup(ORTE_SCHEMA_INVALID_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ if (0 > asprintf(cellid_string, "%ld", (long) name->cellid)) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_base_convert_cellid_to_string(char **cellid_string, const orte_cellid_t cellid)
+{
+ /* check for wildcard value - handle appropriately */
+ if (ORTE_CELLID_WILDCARD == cellid) {
+ *cellid_string = strdup(ORTE_SCHEMA_WILDCARD_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ /* check for invalid value - handle appropriately */
+ if (ORTE_CELLID_INVALID == cellid) {
+ *cellid_string = strdup(ORTE_SCHEMA_INVALID_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ if (0 > asprintf(cellid_string, "%ld", (long) cellid)) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_base_convert_string_to_cellid(orte_cellid_t *cellid, const char *cellidstring)
+{
+ long int tmpint;
+
+ if (NULL == cellidstring) {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ *cellid = ORTE_CELLID_INVALID;
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ /** check for wildcard string - handle appropriately */
+ if (0 == strcmp(ORTE_SCHEMA_WILDCARD_STRING, cellidstring)) {
+ *cellid = ORTE_CELLID_WILDCARD;
+ return ORTE_SUCCESS;
+ }
+
+ /** check for invalid string - handle appropriately */
+ if (0 == strcmp(ORTE_SCHEMA_INVALID_STRING, cellidstring)) {
+ *cellid = ORTE_CELLID_INVALID;
+ return ORTE_SUCCESS;
+ }
+
+ tmpint = strtol(cellidstring, NULL, 10);
+
+ if (ORTE_CELLID_MAX >= tmpint && ORTE_CELLID_MIN <= tmpint) {
+ *cellid = (orte_cellid_t)tmpint;
+ } else {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ *cellid = ORTE_CELLID_INVALID;
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+/**** NODEID STRING FUNCTIONS ****/
+int orte_ns_base_convert_nodeid_to_string(char **string, const orte_nodeid_t nodeid)
+{
+ *string = NULL;
+
+ /* check for wildcard value - handle appropriately */
+ if (ORTE_NODEID_WILDCARD == nodeid) {
+ *string = strdup(ORTE_SCHEMA_WILDCARD_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ /* check for invalid value - handle appropriately */
+ if (ORTE_NODEID_INVALID == nodeid) {
+ *string = strdup(ORTE_SCHEMA_INVALID_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ if (0 > asprintf(string, "%ld", (long)nodeid)) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_base_convert_string_to_nodeid(orte_nodeid_t *nodeid, const char* string)
+{
+ long int tmpint;
+
+ if (NULL == string) { /* got an error */
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ *nodeid = ORTE_NODEID_INVALID;
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ /** check for wildcard character - handle appropriately */
+ if (0 == strcmp(ORTE_SCHEMA_WILDCARD_STRING, string)) {
+ *nodeid = ORTE_NODEID_WILDCARD;
+ return ORTE_SUCCESS;
+ }
+
+ /* check for invalid value */
+ if (0 == strcmp(ORTE_SCHEMA_INVALID_STRING, string)) {
+ *nodeid = ORTE_NODEID_INVALID;
+ return ORTE_SUCCESS;
+ }
+
+ tmpint = strtol(string, NULL, 10);
+
+ if (ORTE_NODEID_MAX >= tmpint && ORTE_NODEID_MIN <= tmpint) {
+ *nodeid = (orte_nodeid_t)tmpint;
+ } else {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ *nodeid = ORTE_NODEID_INVALID;
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ return ORTE_SUCCESS;
+}
+
diff --git a/orte/mca/ns/base/ns_base_diag_fns.c b/orte/mca/ns/base/ns_base_diag_fns.c
new file mode 100644
index 0000000000..734ed4cfea
--- /dev/null
+++ b/orte/mca/ns/base/ns_base_diag_fns.c
@@ -0,0 +1,98 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+
+#include "orte_config.h"
+
+#include <stdio.h>
+#include <string.h>
+#include <stddef.h>
+#include <stdlib.h>
+#if HAVE_NETINET_IN_H
+#include <netinet/in.h>
+#endif
+
+#include "orte/orte_constants.h"
+
+#include "opal/util/output.h"
+#include "opal/util/printf.h"
+#include "opal/mca/mca.h"
+
+#include "orte/dss/dss.h"
+#include "orte/mca/errmgr/errmgr.h"
+
+#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/base/ns_private.h"
+
+/*
+ * "not available" functions
+ */
+
+int
+orte_ns_base_dump_cells_not_available(void)
+{
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int
+orte_ns_base_dump_jobs_not_available(void)
+{
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int
+orte_ns_base_dump_tags_not_available(void)
+{
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int
+orte_ns_base_dump_datatypes_not_available(void)
+{
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+/**** DIAGNOSTIC FUNCTIONS ****/
+int orte_ns_base_print_dump(orte_buffer_t *buffer)
+{
+ char *line;
+ orte_std_cntr_t n;
+ orte_data_type_t type;
+ int rc;
+
+ n = 1;
+ while (ORTE_SUCCESS == orte_dss.peek(buffer, &type, &n)) {
+ if (ORTE_SUCCESS !=
+ (rc = orte_dss.unpack(buffer, &line, &n, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ opal_output(mca_ns_base_output, "%s", line);
+ free(line);
+ n=1;
+ }
+
+ return ORTE_SUCCESS;
+}
+
diff --git a/orte/mca/ns/base/ns_base_general_fns.c b/orte/mca/ns/base/ns_base_general_fns.c
new file mode 100644
index 0000000000..884fbeaa1b
--- /dev/null
+++ b/orte/mca/ns/base/ns_base_general_fns.c
@@ -0,0 +1,78 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+
+#include "orte_config.h"
+
+#include <stdio.h>
+#include <string.h>
+#include <stddef.h>
+#include <stdlib.h>
+#if HAVE_NETINET_IN_H
+#include <netinet/in.h>
+#endif
+
+#include "orte/orte_constants.h"
+
+#include "opal/util/output.h"
+#include "opal/util/printf.h"
+#include "opal/mca/mca.h"
+
+#include "orte/mca/errmgr/errmgr.h"
+
+#include "orte/mca/ns/base/ns_private.h"
+
+/*
+ * "not available" functions
+ */
+int
+orte_ns_base_module_init_not_available(void)
+{
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int
+orte_ns_base_assign_rml_tag_not_available(orte_rml_tag_t *tag, char *name)
+{
+ *tag = ORTE_RML_TAG_MAX;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int
+orte_ns_base_define_data_type_not_available(
+ const char *name,
+ orte_data_type_t *type)
+{
+ *type = ORTE_DSS_ID_MAX;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+/**** GET PEERS ****/
+int orte_ns_base_get_peers_not_available(orte_process_name_t **procs,
+ orte_std_cntr_t *num_procs, opal_list_t *attributes)
+{
+ *procs = NULL;
+ *num_procs = 0;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
diff --git a/orte/mca/ns/base/ns_base_job_fns.c b/orte/mca/ns/base/ns_base_job_fns.c
new file mode 100644
index 0000000000..f15eb7498f
--- /dev/null
+++ b/orte/mca/ns/base/ns_base_job_fns.c
@@ -0,0 +1,169 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+
+#include "orte_config.h"
+
+#include <stdio.h>
+#include <string.h>
+#include <stddef.h>
+#include <stdlib.h>
+#if HAVE_NETINET_IN_H
+#include <netinet/in.h>
+#endif
+
+#include "orte/orte_constants.h"
+
+#include "opal/util/output.h"
+#include "opal/util/printf.h"
+#include "opal/mca/mca.h"
+
+#include "orte/mca/schema/schema_types.h"
+#include "orte/mca/errmgr/errmgr.h"
+
+#include "orte/mca/ns/base/ns_private.h"
+
+/*
+ * "not available" functions
+ */
+int
+orte_ns_base_create_jobid_not_available(orte_jobid_t *jobid, opal_list_t *attrs)
+{
+ *jobid = ORTE_JOBID_INVALID;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int
+orte_ns_base_get_vpid_range_not_available(orte_jobid_t job,
+ orte_vpid_t range,
+ orte_vpid_t *startvpid)
+{
+ *startvpid = ORTE_VPID_INVALID;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int orte_ns_base_get_job_descendants_not_available(orte_jobid_t** descendants,
+ orte_std_cntr_t *num_desc,
+ orte_jobid_t job)
+{
+ *descendants = NULL;
+ *num_desc = 0;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int orte_ns_base_get_job_children_not_available(orte_jobid_t** children,
+ orte_std_cntr_t *num_childs,
+ orte_jobid_t job)
+{
+ *children = NULL;
+ *num_childs = 0;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int orte_ns_base_get_root_job_not_available(orte_jobid_t *root_job, orte_jobid_t job)
+{
+ *root_job = ORTE_JOBID_INVALID;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+int orte_ns_base_get_parent_job_not_available(orte_jobid_t *parent, orte_jobid_t job)
+{
+ *parent = ORTE_JOBID_INVALID;
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+
+/**** JOB STRING FUNCTIONS ****/
+int orte_ns_base_get_jobid_string(char **jobid_string, const orte_process_name_t* name)
+{
+ if (NULL == name) { /* got an error */
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ *jobid_string = NULL;
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ /* check for wildcard value - handle appropriately */
+ if (ORTE_JOBID_WILDCARD == name->jobid) {
+ *jobid_string = strdup(ORTE_SCHEMA_WILDCARD_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ if (0 > asprintf(jobid_string, "%ld", (long) name->jobid)) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_base_convert_jobid_to_string(char **jobid_string, const orte_jobid_t jobid)
+{
+ /* check for wildcard value - handle appropriately */
+ if (ORTE_JOBID_WILDCARD == jobid) {
+ *jobid_string = strdup(ORTE_SCHEMA_WILDCARD_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ if (0 > asprintf(jobid_string, "%ld", (long) jobid)) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_base_convert_string_to_jobid(orte_jobid_t *jobid, const char* jobidstring)
+{
+ long int tmpint;
+
+ if (NULL == jobidstring) { /* got an error */
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ *jobid = ORTE_JOBID_INVALID;
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ tmpint = strtoul(jobidstring, NULL, 10);
+
+ /* check for invalid value */
+ if (ORTE_JOBID_INVALID == tmpint) {
+ *jobid = ORTE_JOBID_INVALID;
+ return ORTE_SUCCESS;
+ }
+
+ if (ORTE_JOBID_MAX >= tmpint && ORTE_JOBID_MIN <= tmpint) {
+ *jobid = (orte_jobid_t)tmpint;
+ } else {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ *jobid = ORTE_JOBID_INVALID;
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
diff --git a/orte/mca/ns/base/ns_base_local_fns.c b/orte/mca/ns/base/ns_base_local_fns.c
deleted file mode 100644
index 80dbd28b95..0000000000
--- a/orte/mca/ns/base/ns_base_local_fns.c
+++ /dev/null
@@ -1,678 +0,0 @@
-/*
- * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
- * University Research and Technology
- * Corporation. All rights reserved.
- * Copyright (c) 2004-2005 The University of Tennessee and The University
- * of Tennessee Research Foundation. All rights
- * reserved.
- * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
- * University of Stuttgart. All rights reserved.
- * Copyright (c) 2004-2005 The Regents of the University of California.
- * All rights reserved.
- * $COPYRIGHT$
- *
- * Additional copyrights may follow
- *
- * $HEADER$
- */
-/** @file:
- *
- */
-
-#include "orte_config.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <stddef.h>
-#include <stdlib.h>
-#if HAVE_NETINET_IN_H
-#include <netinet/in.h>
-#endif
-
-#include "orte/orte_constants.h"
-
-#include "opal/util/output.h"
-#include "opal/util/printf.h"
-#include "opal/mca/mca.h"
-
-#include "orte/mca/errmgr/errmgr.h"
-
-#include "orte/mca/ns/base/base.h"
-
-/**
- * globals
- */
-
-
-/*
- * "not available" functions
- */
-int
-orte_ns_base_module_init_not_available(void)
-{
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int
-orte_ns_base_create_cellid_not_available(orte_cellid_t *cellid, char *site, char *resource)
-{
- *cellid = ORTE_CELLID_MAX;
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int
-orte_ns_base_get_cell_info_not_available(orte_cellid_t cellid,
- char **site, char **resource)
-{
- *site = NULL;
- *resource = NULL;
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int
-orte_ns_base_create_jobid_not_available(orte_jobid_t *jobid)
-{
- *jobid = ORTE_JOBID_MAX;
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int
-orte_ns_base_get_vpid_range_not_available(orte_jobid_t job,
- orte_vpid_t range,
- orte_vpid_t *startvpid)
-{
- *startvpid = ORTE_VPID_MAX;
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int
-orte_ns_base_assign_rml_tag_not_available(orte_rml_tag_t *tag, char *name)
-{
- *tag = ORTE_RML_TAG_MAX;
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int
-orte_ns_base_define_data_type_not_available(
- const char *name,
- orte_data_type_t *type)
-{
- *type = ORTE_DSS_ID_MAX;
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int
-orte_ns_base_create_my_name_not_available(void)
-{
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int orte_ns_base_get_job_peers_not_available(orte_process_name_t **procs,
- orte_std_cntr_t *num_procs, orte_jobid_t job)
-{
- *procs = NULL;
- *num_procs = 0;
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int
-orte_ns_base_dump_cells_not_available(void)
-{
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int
-orte_ns_base_dump_jobs_not_available(void)
-{
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int
-orte_ns_base_dump_tags_not_available(void)
-{
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-int
-orte_ns_base_dump_datatypes_not_available(void)
-{
- ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
- return ORTE_ERR_UNREACH;
-}
-
-
-
-/*
- * functions
- */
-
-int orte_ns_base_assign_cellid_to_process(orte_process_name_t *name)
-{
- if (NULL == name) {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- name->cellid = 0;
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_create_process_name(orte_process_name_t **name,
- orte_cellid_t cell,
- orte_jobid_t job,
- orte_vpid_t vpid)
-{
- *name = NULL;
-
- if (ORTE_CELLID_MAX < cell ||
- ORTE_JOBID_MAX < job ||
- ORTE_VPID_MAX < vpid) {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- *name = (orte_process_name_t*)malloc(sizeof(orte_process_name_t));
- if (NULL == *name) { /* got an error */
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- (*name)->cellid = cell;
- (*name)->jobid = job;
- (*name)->vpid = vpid;
- return ORTE_SUCCESS;
-}
-
-int orte_ns_base_derive_vpid(orte_vpid_t *vpid, orte_vpid_t base_vpid, int offset)
-{
- *vpid = base_vpid + (orte_vpid_t)offset;
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_copy_process_name(orte_process_name_t **dest,
- orte_process_name_t* src)
-{
- orte_cellid_t cell;
- orte_jobid_t job;
- orte_vpid_t vpid;
- int rc;
-
- *dest = NULL;
-
- if (NULL == src) {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- if (ORTE_SUCCESS != orte_ns_base_get_cellid(&cell, src)) {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
- if (ORTE_SUCCESS != orte_ns_base_get_jobid(&job, src)) {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
- if (ORTE_SUCCESS != orte_ns_base_get_vpid(&vpid, src)) {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- if (ORTE_SUCCESS != (rc = orte_ns_base_create_process_name(dest, cell, job, vpid))) {
- ORTE_ERROR_LOG(rc);
- }
-
- return rc;
-}
-
-int orte_ns_base_get_proc_name_string(char **name_string,
- const orte_process_name_t* name)
-{
- *name_string = NULL;
-
- if (NULL == name) { /* got an error */
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- if (0 > asprintf(name_string, "%lu.%lu.%lu", ORTE_NAME_ARGS(name))) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- return ORTE_SUCCESS;
-}
-
-int orte_ns_base_convert_string_to_process_name(orte_process_name_t **name,
- const char* name_string)
-{
- char *temp, *token;
- orte_cellid_t cell;
- orte_jobid_t job;
- orte_vpid_t vpid;
- unsigned long int tmpint;
- int return_code=ORTE_SUCCESS;
-
- const char delimiters[] = ".";
-
- *name = NULL;
-
- /* check for NULL string - error */
- if (NULL == name_string) {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- temp = strdup(name_string);
- token = strtok(temp, delimiters); /* get first field -> cellid */
-
- /* convert to largest possible unsigned int - unsigned long long is only supported
- * in C99, so we have to use unsigned long for backward compatibility - then
- * check to ensure it is within range of cellid_t before casting */
-
- tmpint = strtoul(token, NULL, 10);
- if (ORTE_CELLID_MAX >= tmpint) {
- cell = (orte_cellid_t)tmpint;
- } else {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return_code = ORTE_ERR_BAD_PARAM;
- goto CLEANUP;
- }
-
- token = strtok(NULL, delimiters); /* get second field -> jobid */
-
- /* convert to largest possible unsigned int - then
- * check to ensure it is within range of jobid_t before casting */
-
- tmpint = strtoul(token, NULL, 10);
- if (ORTE_JOBID_MAX >= tmpint) {
- job = (orte_jobid_t)tmpint;
- } else {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return_code = ORTE_ERR_BAD_PARAM;
- goto CLEANUP;
- }
-
- token = strtok(NULL, delimiters); /* get third field -> vpid */
-
- /* convert to largest possible unsigned int then
- * check to ensure it is within range of vpid_t before casting */
-
- tmpint = strtoul(token, NULL, 10);
- if (ORTE_VPID_MAX >= tmpint) {
- vpid = (orte_vpid_t)tmpint;
- } else {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return_code = ORTE_ERR_BAD_PARAM;
- goto CLEANUP;
- }
-
- if (ORTE_SUCCESS != (return_code = orte_ns_base_create_process_name(name, cell, job, vpid))) {
- ORTE_ERROR_LOG(return_code);
- }
-
- CLEANUP:
- if (temp) {
- free(temp);
- }
-
- return return_code;
-}
-
-
-int orte_ns_base_get_vpid_string(char **vpid_string, const orte_process_name_t* name)
-{
- *vpid_string = NULL;
-
- if (NULL == name) { /* got an error */
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- if (0 > asprintf(vpid_string, "%lu", (unsigned long) name->vpid)) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_convert_vpid_to_string(char **vpid_string, const orte_vpid_t vpid)
-{
- *vpid_string = NULL;
-
- if (0 > asprintf(vpid_string, "%lu", (unsigned long) vpid)) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_convert_string_to_vpid(orte_vpid_t *vpid, const char* vpidstring)
-{
- unsigned long int tmpint;
-
- *vpid = ORTE_VPID_MAX;
-
- if (NULL == vpidstring) { /* got an error */
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- tmpint = strtoul(vpidstring, NULL, 10);
- if (ORTE_VPID_MAX >= tmpint) {
- *vpid = (orte_vpid_t)tmpint;
- } else {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- *vpid = ORTE_VPID_MAX;
- return ORTE_ERR_BAD_PARAM;
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_get_jobid_string(char **jobid_string, const orte_process_name_t* name)
-{
- *jobid_string = NULL;
-
- if (NULL == name) { /* got an error */
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- if (0 > asprintf(jobid_string, "%lu", (unsigned long) name->jobid)) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_convert_jobid_to_string(char **jobid_string, const orte_jobid_t jobid)
-{
- *jobid_string = NULL;
-
- if (0 > asprintf(jobid_string, "%lu", (unsigned long) jobid)) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_convert_string_to_jobid(orte_jobid_t *jobid, const char* jobidstring)
-{
- unsigned long int tmpint;
-
- *jobid = ORTE_JOBID_MAX;
-
- if (NULL == jobidstring) { /* got an error */
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- tmpint = strtoul(jobidstring, NULL, 10);
- if (ORTE_JOBID_MAX >= tmpint) {
- *jobid = (orte_jobid_t)tmpint;
- } else {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- *jobid = ORTE_JOBID_MAX;
- return ORTE_ERR_BAD_PARAM;
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_get_cellid_string(char **cellid_string, const orte_process_name_t* name)
-{
- *cellid_string = NULL;
-
- if (NULL == name) { /* got an error */
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- if (0 > asprintf(cellid_string, "%lu", (unsigned long) name->cellid)) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_convert_cellid_to_string(char **cellid_string, const orte_cellid_t cellid)
-{
- *cellid_string = NULL;
-
- if (0 > asprintf(cellid_string, "%lu", (unsigned long) cellid)) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_convert_string_to_cellid(orte_cellid_t *cellid, const char *cellidstring)
-{
- unsigned long int tmpint;
-
- *cellid = ORTE_CELLID_MAX;
-
- if (NULL == cellidstring) {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- tmpint = strtoul(cellidstring, NULL, 10);
- if (ORTE_CELLID_MAX >= tmpint) {
- *cellid = (orte_cellid_t)tmpint;
- } else {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- *cellid = ORTE_CELLID_MAX;
- return ORTE_ERR_BAD_PARAM;
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_get_vpid(orte_vpid_t *vpid, const orte_process_name_t* name)
-{
- *vpid = ORTE_VPID_MAX;
-
- if (NULL == name) { /* got an error */
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- *vpid = name->vpid;
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_get_jobid(orte_jobid_t *jobid, const orte_process_name_t* name)
-{
- *jobid = ORTE_JOBID_MAX;
-
- if (NULL == name) { /* got an error */
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- *jobid = name->jobid;
-
- return ORTE_SUCCESS;
-}
-
-int orte_ns_base_get_cellid(orte_cellid_t *cellid, const orte_process_name_t* name)
-{
- *cellid = ORTE_CELLID_MAX;
-
- if (NULL == name) { /* got an error */
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- *cellid = name->cellid;
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_base_compare(orte_ns_cmp_bitmask_t fields,
- const orte_process_name_t* name1,
- const orte_process_name_t* name2)
-{
- if (NULL == name1 && NULL == name2) {
- return 0;
- } else if (NULL == name1) {
- return -1;
- } else if (NULL == name2) {
- return 1;
- }
-
- if (ORTE_NS_CMP_CELLID & fields) { /* check cellid field */
- if (name1->cellid < name2->cellid) {
- return -1;
- } else if (name1->cellid > name2->cellid) {
- return 1;
- }
- }
-
- /* get here if cellid's are equal, or cellid not being checked */
- /* now check job id */
-
- if (ORTE_NS_CMP_JOBID & fields) {
- if (name1->jobid < name2->jobid) {
- return -1;
- } else if (name1->jobid > name2->jobid) {
- return 1;
- }
- }
-
- /* get here if cellid's and jobid's are equal, or neither being checked,
- * or cellid not checked and jobid's equal.
- * now check vpid
- */
-
- if (ORTE_NS_CMP_VPID & fields) {
- if (name1->vpid < name2->vpid) {
- return -1;
- } else if (name1->vpid > name2->vpid) {
- return 1;
- }
- }
-
- /* only way to get here is if all fields are being checked and are equal,
- * or cellid not checked, but jobid and vpid equal,
- * or cellid and jobid not checked, but vpid equal,
- * only vpid being checked, and equal
- * return that fact
- */
- return 0;
-}
-
-
-int orte_ns_base_free_name(orte_process_name_t **name)
-{
- if (NULL != name && NULL != *name) {
- free(*name);
- }
-
- *name = NULL;
-
- return ORTE_SUCCESS;
-}
-
-int orte_ns_base_get_peers(orte_process_name_t **procs,
- orte_std_cntr_t *num_procs, orte_std_cntr_t *self)
-{
- orte_std_cntr_t i;
- int rc;
- orte_cellid_t mycellid;
- orte_jobid_t myjobid;
- orte_vpid_t myvpid;
-
- *procs = (orte_process_name_t*)malloc(orte_process_info.num_procs *
- sizeof(orte_process_name_t));
- if (NULL == *procs) {
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (ORTE_SUCCESS != (rc = orte_ns.get_cellid(&mycellid, orte_process_info.my_name))) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
-
- if (ORTE_SUCCESS != orte_ns.get_jobid(&myjobid, orte_process_info.my_name)) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
-
- if (ORTE_SUCCESS != orte_ns.get_vpid(&myvpid, orte_process_info.my_name)) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
-
- for (i=0; i < orte_process_info.num_procs; i++) {
- (*procs)[i].cellid = mycellid;
- (*procs)[i].jobid = myjobid;
- (*procs)[i].vpid = orte_process_info.vpid_start + i;
- }
-
- *num_procs = orte_process_info.num_procs;
- *self = (orte_std_cntr_t)(myvpid - orte_process_info.vpid_start);
-
- return ORTE_SUCCESS;
-}
-
-
-/*
- * DIAGNOSTIC FUNCTIONS
- */
-int orte_ns_base_print_dump(orte_buffer_t *buffer)
-{
- char *line;
- orte_std_cntr_t n;
- orte_data_type_t type;
- int rc;
-
- n = 1;
- while (ORTE_SUCCESS == orte_dss.peek(buffer, &type, &n)) {
- if (ORTE_SUCCESS !=
- (rc = orte_dss.unpack(buffer, &line, &n, ORTE_STRING))) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
- opal_output(mca_ns_base_output, "%s", line);
- free(line);
- n=1;
- }
-
- return ORTE_SUCCESS;
-}
-
diff --git a/orte/mca/ns/base/ns_base_open.c b/orte/mca/ns/base/ns_base_open.c
index e6c49b0975..0a0f110c82 100644
--- a/orte/mca/ns/base/ns_base_open.c
+++ b/orte/mca/ns/base/ns_base_open.c
@@ -29,6 +29,7 @@
#include "orte/dss/dss.h"
#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/base/ns_private.h"
/*
@@ -43,7 +44,9 @@
* globals
*/
-orte_process_name_t orte_name_all = {ORTE_CELLID_MAX, ORTE_JOBID_MAX, ORTE_VPID_MAX};
+orte_process_name_t orte_ns_name_wildcard = {ORTE_CELLID_WILDCARD, ORTE_JOBID_WILDCARD, ORTE_VPID_WILDCARD};
+orte_process_name_t orte_ns_name_invalid = {ORTE_CELLID_INVALID, ORTE_JOBID_INVALID, ORTE_VPID_INVALID};
+orte_process_name_t orte_ns_name_my_hnp = {0, 0, 0};
/*
* Global variables
@@ -54,35 +57,37 @@ mca_ns_base_module_t orte_ns = {
orte_ns_base_module_init_not_available,
/* cell functions */
orte_ns_base_create_cellid_not_available,
- orte_ns_base_get_cellid,
orte_ns_base_get_cell_info_not_available,
- orte_ns_base_assign_cellid_to_process,
orte_ns_base_get_cellid_string,
orte_ns_base_convert_cellid_to_string,
orte_ns_base_convert_string_to_cellid,
+ /* node functions */
+ orte_ns_base_create_nodeids_not_available,
+ orte_ns_base_get_node_info_not_available,
+ orte_ns_base_convert_nodeid_to_string,
+ orte_ns_base_convert_string_to_nodeid,
/* jobid functions */
orte_ns_base_create_jobid_not_available,
- orte_ns_base_get_jobid,
+ orte_ns_base_get_job_descendants_not_available,
+ orte_ns_base_get_job_children_not_available,
+ orte_ns_base_get_root_job_not_available,
+ orte_ns_base_get_parent_job_not_available,
orte_ns_base_get_jobid_string,
orte_ns_base_convert_jobid_to_string,
orte_ns_base_convert_string_to_jobid,
- /* vpid functions */
orte_ns_base_get_vpid_range_not_available,
- orte_ns_base_get_vpid,
+ /* vpid functions */
orte_ns_base_get_vpid_string,
orte_ns_base_convert_vpid_to_string,
orte_ns_base_convert_string_to_vpid,
/* name functions */
orte_ns_base_create_process_name,
orte_ns_base_create_my_name_not_available,
- orte_ns_base_copy_process_name,
orte_ns_base_convert_string_to_process_name,
- orte_ns_base_free_name,
orte_ns_base_get_proc_name_string,
- orte_ns_base_compare,
+ orte_ns_base_compare_fields,
/* peer functions */
- orte_ns_base_get_peers,
- orte_ns_base_get_job_peers_not_available,
+ orte_ns_base_get_peers_not_available,
/* tag server functions */
orte_ns_base_assign_rml_tag_not_available,
/* data type functions */
diff --git a/orte/mca/ns/base/ns_base_vpid_name_fns.c b/orte/mca/ns/base/ns_base_vpid_name_fns.c
new file mode 100644
index 0000000000..4d5465335c
--- /dev/null
+++ b/orte/mca/ns/base/ns_base_vpid_name_fns.c
@@ -0,0 +1,397 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+
+#include "orte_config.h"
+
+#include <stdio.h>
+#include <string.h>
+#include <stddef.h>
+#include <stdlib.h>
+#if HAVE_NETINET_IN_H
+#include <netinet/in.h>
+#endif
+
+#include "orte/orte_constants.h"
+
+#include "opal/util/output.h"
+#include "opal/util/printf.h"
+#include "opal/mca/mca.h"
+
+#include "orte/mca/schema/schema_types.h"
+#include "orte/mca/errmgr/errmgr.h"
+
+#include "orte/mca/ns/base/ns_private.h"
+
+/*
+ * "not available" functions
+ */
+int
+orte_ns_base_create_my_name_not_available(void)
+{
+ ORTE_ERROR_LOG(ORTE_ERR_UNREACH);
+ return ORTE_ERR_UNREACH;
+}
+
+/**** NAME STRING FUNCTIONS ****/
+
+int orte_ns_base_get_proc_name_string(char **name_string,
+ const orte_process_name_t* name)
+{
+ char *tmp, *tmp2;
+
+ if (NULL == name) { /* got an error */
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ /* handle the more typical case where none of the fields
+ * contain WILDCARD or INVALID values
+ */
+ if ((ORTE_CELLID_WILDCARD != name->cellid && ORTE_CELLID_INVALID != name->cellid) &&
+ (ORTE_JOBID_WILDCARD != name->jobid && ORTE_JOBID_INVALID != name->jobid) &&
+ (ORTE_VPID_WILDCARD != name->vpid && ORTE_VPID_INVALID != name->vpid)) {
+ if (0 > asprintf(name_string, "%ld%c%ld%c%ld", (long)name->cellid,
+ ORTE_SCHEMA_DELIMITER_CHAR, (long)name->jobid,
+ ORTE_SCHEMA_DELIMITER_CHAR, (long)name->vpid)) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ return ORTE_SUCCESS;
+ }
+
+ /* okay, now handle the corner cases */
+ if (ORTE_CELLID_WILDCARD == name->cellid) {
+ tmp = strdup(ORTE_SCHEMA_WILDCARD_STRING);
+ } else if (ORTE_CELLID_INVALID == name->cellid) {
+ tmp = strdup(ORTE_SCHEMA_INVALID_STRING);
+ } else {
+ asprintf(&tmp, "%ld", (long)name->cellid);
+ }
+
+ if (ORTE_JOBID_WILDCARD == name->jobid) {
+ asprintf(&tmp2, "%s%c%s", tmp, ORTE_SCHEMA_DELIMITER_CHAR,
+ ORTE_SCHEMA_WILDCARD_STRING, ORTE_SCHEMA_DELIMITER_CHAR);
+ } else if (ORTE_JOBID_INVALID == name->jobid) {
+ asprintf(&tmp2, "%s%c%s", tmp, ORTE_SCHEMA_DELIMITER_CHAR,
+ ORTE_SCHEMA_INVALID_STRING, ORTE_SCHEMA_DELIMITER_CHAR);
+ } else {
+ asprintf(&tmp2, "%s%c%ld", tmp, ORTE_SCHEMA_DELIMITER_CHAR,
+ (long)name->jobid, ORTE_SCHEMA_DELIMITER_CHAR);
+ }
+ free(tmp);
+
+ if (ORTE_VPID_WILDCARD == name->vpid) {
+ asprintf(name_string, "%s%c%s", tmp2, ORTE_SCHEMA_DELIMITER_CHAR,
+ ORTE_SCHEMA_WILDCARD_STRING);
+ } else if (ORTE_VPID_INVALID == name->vpid) {
+ asprintf(name_string, "%s%c%s", tmp2, ORTE_SCHEMA_DELIMITER_CHAR,
+ ORTE_SCHEMA_INVALID_STRING);
+ } else {
+ asprintf(name_string, "%s%c%ld", tmp2, ORTE_SCHEMA_DELIMITER_CHAR,
+ (long)name->vpid);
+ }
+ free(tmp2);
+
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_base_convert_string_to_process_name(orte_process_name_t **name,
+ const char* name_string)
+{
+ char *temp, *token;
+ orte_cellid_t cell;
+ orte_jobid_t job;
+ orte_vpid_t vpid;
+ long int tmpint;
+ int return_code=ORTE_SUCCESS;
+
+ /* check for NULL string - error */
+ if (NULL == name_string) {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ temp = strdup(name_string); /** copy input string as the strtok process is destructive */
+ token = strtok(temp, ORTE_SCHEMA_DELIMITER_STRING); /** get first field -> cellid */
+
+ /* check for error */
+ if (NULL == token) {
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ /* convert to largest possible int - then
+ * check to ensure it is within range of cellid_t before casting
+ */
+
+ /* first, though, check for WILDCARD character - assign
+ * value accordingly, if found
+ */
+ if (0 == strcmp(token, ORTE_SCHEMA_WILDCARD_STRING)) {
+ cell = ORTE_CELLID_WILDCARD;
+ } else if (0 == strcmp(token, ORTE_SCHEMA_INVALID_STRING)) {
+ cell = ORTE_CELLID_INVALID;
+ } else {
+ tmpint = strtol(token, NULL, 10);
+ if (ORTE_CELLID_MAX >= tmpint && ORTE_CELLID_MIN <= tmpint) {
+ cell = (orte_cellid_t)tmpint;
+ } else {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ return_code = ORTE_ERR_BAD_PARAM;
+ goto CLEANUP;
+ }
+ }
+
+ token = strtok(NULL, ORTE_SCHEMA_DELIMITER_STRING); /** get next field -> jobid */
+
+ /** convert to largest possible int - then
+ * check to ensure it is within range of jobid_t before casting */
+
+ /* check for error */
+ if (NULL == token) {
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ /** first, though, check for WILDCARD character - assign
+ * value accordingly, if found
+ */
+ if (0 == strcmp(token, ORTE_SCHEMA_WILDCARD_STRING)) {
+ job = ORTE_JOBID_WILDCARD;
+ } else if (0 == strcmp(token, ORTE_SCHEMA_INVALID_STRING)) {
+ job = ORTE_JOBID_INVALID;
+ } else {
+ tmpint = strtol(token, NULL, 10);
+ if (ORTE_JOBID_MAX >= tmpint && ORTE_JOBID_MIN <= tmpint) {
+ job = (orte_jobid_t)tmpint;
+ } else {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ return_code = ORTE_ERR_BAD_PARAM;
+ goto CLEANUP;
+ }
+ }
+
+ token = strtok(NULL, ORTE_SCHEMA_DELIMITER_STRING); /** get next field -> vpid */
+
+ /* check for error */
+ if (NULL == token) {
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ /** convert to largest possible int then
+ * check to ensure it is within range of vpid_t before casting */
+
+ /** first, though, check for WILDCARD character - assign
+ * value accordingly, if found
+ */
+ if (0 == strcmp(token, ORTE_SCHEMA_WILDCARD_STRING)) {
+ vpid = ORTE_VPID_WILDCARD;
+ } else if (0 == strcmp(token, ORTE_SCHEMA_INVALID_STRING)) {
+ vpid = ORTE_VPID_INVALID;
+ } else {
+ tmpint = strtol(token, NULL, 10);
+ if (ORTE_VPID_MAX >= tmpint && ORTE_VPID_MIN <= tmpint) {
+ vpid = (orte_vpid_t)tmpint;
+ } else {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ return_code = ORTE_ERR_BAD_PARAM;
+ goto CLEANUP;
+ }
+ }
+
+ if (ORTE_SUCCESS != (return_code =
+ orte_ns_base_create_process_name(name, cell, job, vpid))) {
+ ORTE_ERROR_LOG(return_code);
+ }
+
+CLEANUP:
+ free(temp);
+
+ return return_code;
+}
+
+/**** CREATE PROCESS NAME ****/
+int orte_ns_base_create_process_name(orte_process_name_t **name,
+ orte_cellid_t cell,
+ orte_jobid_t job,
+ orte_vpid_t vpid)
+{
+ *name = NULL;
+
+ *name = (orte_process_name_t*)malloc(sizeof(orte_process_name_t));
+ if (NULL == *name) { /* got an error */
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ (*name)->cellid = cell;
+ (*name)->jobid = job;
+ (*name)->vpid = vpid;
+ return ORTE_SUCCESS;
+}
+
+
+/**** VPID STRING FUNCTIONS ****/
+int orte_ns_base_get_vpid_string(char **vpid_string, const orte_process_name_t* name)
+{
+ if (NULL == name) { /* got an error */
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ *vpid_string = NULL;
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ /* check for wildcard value - handle appropriately */
+ if (ORTE_VPID_WILDCARD == name->vpid) {
+ *vpid_string = strdup(ORTE_SCHEMA_WILDCARD_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ /* check for invalid value - handle appropriately */
+ if (ORTE_VPID_INVALID == name->vpid) {
+ *vpid_string = strdup(ORTE_SCHEMA_INVALID_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ if (0 > asprintf(vpid_string, "%ld", (long) name->vpid)) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_base_convert_vpid_to_string(char **vpid_string, const orte_vpid_t vpid)
+{
+ /* check for wildcard value - handle appropriately */
+ if (ORTE_VPID_WILDCARD == vpid) {
+ *vpid_string = strdup(ORTE_SCHEMA_WILDCARD_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ /* check for invalid value - handle appropriately */
+ if (ORTE_VPID_INVALID == vpid) {
+ *vpid_string = strdup(ORTE_SCHEMA_INVALID_STRING);
+ return ORTE_SUCCESS;
+ }
+
+ if (0 > asprintf(vpid_string, "%ld", (long) vpid)) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_base_convert_string_to_vpid(orte_vpid_t *vpid, const char* vpidstring)
+{
+ long int tmpint;
+
+ if (NULL == vpidstring) { /* got an error */
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ *vpid = ORTE_VPID_INVALID;
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ /** check for wildcard character - handle appropriately */
+ if (0 == strcmp(ORTE_SCHEMA_WILDCARD_STRING, vpidstring)) {
+ *vpid = ORTE_VPID_WILDCARD;
+ return ORTE_SUCCESS;
+ }
+
+ /* check for invalid value */
+ if (0 == strcmp(ORTE_SCHEMA_INVALID_STRING, vpidstring)) {
+ *vpid = ORTE_VPID_INVALID;
+ return ORTE_SUCCESS;
+ }
+
+ tmpint = strtol(vpidstring, NULL, 10);
+
+ if (ORTE_VPID_MAX >= tmpint && ORTE_VPID_MIN <= tmpint) {
+ *vpid = (orte_vpid_t)tmpint;
+ } else {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ *vpid = ORTE_VPID_INVALID;
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+/**** COMPARE NAME FIELDS ****/
+int orte_ns_base_compare_fields(orte_ns_cmp_bitmask_t fields,
+ const orte_process_name_t* name1,
+ const orte_process_name_t* name2)
+{
+ /* handle the NULL pointer case */
+ if (NULL == name1 && NULL == name2) {
+ return ORTE_EQUAL;
+ } else if (NULL == name1) {
+ return ORTE_VALUE2_GREATER;
+ } else if (NULL == name2) {
+ return ORTE_VALUE1_GREATER;
+ }
+
+ /* in this comparison function, we check for exact equalities.
+ * In the case of wildcards, we check to ensure that the fields
+ * actually match those values - thus, a "wildcard" in this
+ * function does not actually stand for a wildcard value, but
+ * rather a specific value
+ */
+ if (ORTE_NS_CMP_CELLID & fields) { /* check cellid field */
+ if (name1->cellid < name2->cellid) {
+ return ORTE_VALUE2_GREATER;
+ } else if (name1->cellid > name2->cellid) {
+ return ORTE_VALUE1_GREATER;
+ }
+ }
+
+ /* get here if cellid's are equal, or cellid not being checked */
+ /* now check job id */
+
+ if (ORTE_NS_CMP_JOBID & fields) {
+ if (name1->jobid < name2->jobid) {
+ return ORTE_VALUE2_GREATER;
+ } else if (name1->jobid > name2->jobid) {
+ return ORTE_VALUE1_GREATER;
+ }
+ }
+
+ /* get here if cellid's and jobid's are equal, or neither being checked,
+ * or cellid not checked and jobid's equal.
+ * now check vpid
+ */
+
+ if (ORTE_NS_CMP_VPID & fields) {
+ if (name1->vpid < name2->vpid) {
+ return ORTE_VALUE2_GREATER;
+ } else if (name1->vpid > name2->vpid) {
+ return ORTE_VALUE1_GREATER;
+ }
+ }
+
+ /* only way to get here is if all fields are being checked and are equal,
+ * or cellid not checked, but jobid and vpid equal,
+ * or cellid and jobid not checked, but vpid equal,
+ * only vpid being checked, and equal
+ * return that fact
+ */
+ return ORTE_EQUAL;
+}
diff --git a/orte/mca/ns/base/ns_private.h b/orte/mca/ns/base/ns_private.h
new file mode 100644
index 0000000000..9cfadf206a
--- /dev/null
+++ b/orte/mca/ns/base/ns_private.h
@@ -0,0 +1,285 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ */
+
+#ifndef MCA_NS_PRIVATE_H
+#define MCA_NS_PRIVATE_H
+
+/*
+ * includes
+ */
+#include "orte_config.h"
+#include "orte/orte_constants.h"
+
+#include "opal/class/opal_list.h"
+#include "opal/mca/mca.h"
+
+#include "orte/dss/dss_types.h"
+#include "orte/mca/ns/ns_types.h"
+#include "orte/mca/rml/rml_types.h"
+
+/*
+ * Global functions for MCA overall collective open and close
+ */
+#if defined(c_plusplus) || defined(__cplusplus)
+extern "C" {
+#endif
+
+/* default limits */
+#define ORTE_NS_ARRAY_MAX_SIZE INT_MAX
+#define ORTE_NS_ARRAY_BLOCK_SIZE 512
+/*
+ * Internal definitions
+ */
+typedef uint8_t orte_ns_cmd_bitmask_t;
+typedef uint8_t orte_ns_cmd_flag_t;
+
+/*
+ * packing type definitions
+ */
+/* CAUTION - any changes here must also change corresponding
+ * typedefs above and in ns_types.h
+ */
+#define ORTE_NS_CMD ORTE_INT8
+#define ORTE_CELLID_T ORTE_INT32
+#define ORTE_NODEID_T ORTE_INT32
+#define ORTE_JOBID_T ORTE_INT32
+#define ORTE_VPID_T ORTE_INT32
+
+/*
+ * define flag values for remote commands - only used internally
+ */
+#define ORTE_NS_CREATE_CELLID_CMD (int8_t) 1
+#define ORTE_NS_GET_CELL_INFO_CMD (int8_t) 2
+#define ORTE_NS_CREATE_NODEID_CMD (int8_t) 3
+#define ORTE_NS_GET_NODE_INFO_CMD (int8_t) 4
+#define ORTE_NS_CREATE_JOBID_CMD (int8_t) 5
+#define ORTE_NS_GET_JOB_DESC_CMD (int8_t) 6
+#define ORTE_NS_GET_JOB_CHILD_CMD (int8_t) 7
+#define ORTE_NS_GET_ROOT_JOB_CMD (int8_t) 8
+#define ORTE_NS_GET_PARENT_JOB_CMD (int8_t) 9
+#define ORTE_NS_RESERVE_RANGE_CMD (int8_t) 10
+#define ORTE_NS_ASSIGN_OOB_TAG_CMD (int8_t) 11
+#define ORTE_NS_GET_PEERS_CMD (int8_t) 12
+#define ORTE_NS_DEFINE_DATA_TYPE_CMD (int8_t) 13
+#define ORTE_NS_CREATE_MY_NAME_CMD (int8_t) 14
+#define ORTE_NS_DUMP_CELLS_CMD (int8_t) 15
+#define ORTE_NS_DUMP_JOBIDS_CMD (int8_t) 16
+#define ORTE_NS_DUMP_TAGS_CMD (int8_t) 17
+#define ORTE_NS_DUMP_DATATYPES_CMD (int8_t) 18
+
+
+/*
+ * Base functions that are common to all implementations - can be overridden
+ */
+
+ORTE_DECLSPEC int orte_ns_base_create_process_name(orte_process_name_t **name,
+ orte_cellid_t cell,
+ orte_jobid_t job,
+ orte_vpid_t vpid);
+
+ORTE_DECLSPEC int orte_ns_base_convert_string_to_process_name(orte_process_name_t **name,
+ const char* name_string);
+
+ORTE_DECLSPEC int orte_ns_base_get_proc_name_string(char **name_string,
+ const orte_process_name_t* name);
+
+ORTE_DECLSPEC int orte_ns_base_get_vpid_string(char **vpid_string, const orte_process_name_t* name);
+
+ORTE_DECLSPEC int orte_ns_base_convert_vpid_to_string(char **vpid_string, const orte_vpid_t vpid);
+
+ORTE_DECLSPEC int orte_ns_base_convert_string_to_vpid(orte_vpid_t *vpid, const char* vpidstring);
+
+ORTE_DECLSPEC int orte_ns_base_get_jobid_string(char **jobid_string, const orte_process_name_t* name);
+
+ORTE_DECLSPEC int orte_ns_base_convert_jobid_to_string(char **jobid_string, const orte_jobid_t jobid);
+
+ORTE_DECLSPEC int orte_ns_base_convert_string_to_jobid(orte_jobid_t *jobid, const char* jobidstring);
+
+ORTE_DECLSPEC int orte_ns_base_get_cellid_string(char **cellid_string, const orte_process_name_t* name);
+
+ORTE_DECLSPEC int orte_ns_base_convert_string_to_cellid(orte_cellid_t *cellid, const char *cellidstring);
+
+ORTE_DECLSPEC int orte_ns_base_convert_cellid_to_string(char **cellid_string, const orte_cellid_t cellid);
+
+ORTE_DECLSPEC int orte_ns_base_get_vpid(orte_vpid_t *vpid, const orte_process_name_t* name);
+
+ORTE_DECLSPEC int orte_ns_base_get_jobid(orte_jobid_t *jobid, const orte_process_name_t* name);
+
+ORTE_DECLSPEC int orte_ns_base_get_cellid(orte_cellid_t *cellid, const orte_process_name_t* name);
+
+ORTE_DECLSPEC int orte_ns_base_convert_string_to_nodeid(orte_nodeid_t *cellid, const char *string);
+
+ORTE_DECLSPEC int orte_ns_base_convert_nodeid_to_string(char **nodeid_string, const orte_nodeid_t nodeid);
+
+ORTE_DECLSPEC int orte_ns_base_compare_fields(orte_ns_cmp_bitmask_t fields,
+ const orte_process_name_t* name1,
+ const orte_process_name_t* name2);
+
+ORTE_DECLSPEC int orte_ns_base_print_dump(orte_buffer_t *buffer);
+
+
+/* not available functions */
+ORTE_DECLSPEC int orte_ns_base_module_init_not_available(void);
+
+ORTE_DECLSPEC int orte_ns_base_create_cellid_not_available(orte_cellid_t *cellid,
+ char *site, char *resource);
+
+ORTE_DECLSPEC int orte_ns_base_get_cell_info_not_available(orte_cellid_t cellid,
+ char **site, char **resource);
+
+ORTE_DECLSPEC int orte_ns_base_create_nodeids_not_available(orte_nodeid_t **nodeids, orte_std_cntr_t *nnodes,
+ orte_cellid_t cellid, char **nodename);
+
+ORTE_DECLSPEC int orte_ns_base_get_node_info_not_available(char ***nodename, orte_cellid_t cellid,
+ orte_std_cntr_t num_nodes, orte_nodeid_t *nodeids);
+
+ORTE_DECLSPEC int orte_ns_base_create_jobid_not_available(orte_jobid_t *jobid, opal_list_t *attrs);
+
+ORTE_DECLSPEC int orte_ns_base_get_job_descendants_not_available(orte_jobid_t** descendants,
+ orte_std_cntr_t *num_desc,
+ orte_jobid_t job);
+
+ORTE_DECLSPEC int orte_ns_base_get_job_children_not_available(orte_jobid_t** children,
+ orte_std_cntr_t *num_childs,
+ orte_jobid_t job);
+
+ORTE_DECLSPEC int orte_ns_base_get_root_job_not_available(orte_jobid_t *root_job, orte_jobid_t job);
+
+ORTE_DECLSPEC int orte_ns_base_get_parent_job_not_available(orte_jobid_t *parent, orte_jobid_t job);
+
+ORTE_DECLSPEC int orte_ns_base_get_vpid_range_not_available(orte_jobid_t job,
+ orte_vpid_t range,
+ orte_vpid_t *startvpid);
+
+ORTE_DECLSPEC int orte_ns_base_assign_rml_tag_not_available(orte_rml_tag_t *tag, char *name);
+
+ORTE_DECLSPEC int orte_ns_base_define_data_type_not_available(
+ const char *name,
+ orte_data_type_t *type);
+
+ORTE_DECLSPEC int orte_ns_base_create_my_name_not_available(void);
+
+ORTE_DECLSPEC int orte_ns_base_get_peers_not_available(orte_process_name_t **procs,
+ orte_std_cntr_t *num_procs, opal_list_t *attributes);
+
+ORTE_DECLSPEC int orte_ns_base_dump_cells_not_available(void);
+ORTE_DECLSPEC int orte_ns_base_dump_jobs_not_available(void);
+ORTE_DECLSPEC int orte_ns_base_dump_tags_not_available(void);
+ORTE_DECLSPEC int orte_ns_base_dump_datatypes_not_available(void);
+
+/* Base functions used everywhere */
+ORTE_DECLSPEC int orte_ns_base_pack_name(orte_buffer_t *buffer, void *src,
+ orte_std_cntr_t num_vals, orte_data_type_t type);
+
+ORTE_DECLSPEC int orte_ns_base_pack_cellid(orte_buffer_t *buffer, void *src,
+ orte_std_cntr_t num_vals, orte_data_type_t type);
+
+ORTE_DECLSPEC int orte_ns_base_pack_nodeid(orte_buffer_t *buffer, void *src,
+ orte_std_cntr_t num_vals, orte_data_type_t type);
+
+ORTE_DECLSPEC int orte_ns_base_pack_jobid(orte_buffer_t *buffer, void *src,
+ orte_std_cntr_t num_vals, orte_data_type_t type);
+
+ORTE_DECLSPEC int orte_ns_base_pack_vpid(orte_buffer_t *buffer, void *src,
+ orte_std_cntr_t num_vals, orte_data_type_t type);
+
+ORTE_DECLSPEC int orte_ns_base_unpack_name(orte_buffer_t *buffer, void *dest,
+ orte_std_cntr_t *num_vals, orte_data_type_t type);
+
+ORTE_DECLSPEC int orte_ns_base_unpack_cellid(orte_buffer_t *buffer, void *dest,
+ orte_std_cntr_t *num_vals, orte_data_type_t type);
+
+ORTE_DECLSPEC int orte_ns_base_unpack_nodeid(orte_buffer_t *buffer, void *dest,
+ orte_std_cntr_t *num_vals, orte_data_type_t type);
+
+ORTE_DECLSPEC int orte_ns_base_unpack_jobid(orte_buffer_t *buffer, void *dest,
+ orte_std_cntr_t *num_vals, orte_data_type_t type);
+
+ORTE_DECLSPEC int orte_ns_base_unpack_vpid(orte_buffer_t *buffer, void *dest,
+ orte_std_cntr_t *num_vals, orte_data_type_t type);
+
+/*
+ * copy functions
+ */
+
+int orte_ns_base_copy_name(orte_process_name_t **dest, orte_process_name_t *src, orte_data_type_t type);
+
+int orte_ns_base_copy_vpid(orte_vpid_t **dest, orte_vpid_t *src, orte_data_type_t type);
+
+int orte_ns_base_copy_cellid(orte_cellid_t **dest, orte_cellid_t *src, orte_data_type_t type);
+
+int orte_ns_base_copy_nodeid(orte_nodeid_t **dest, orte_nodeid_t *src, orte_data_type_t type);
+
+int orte_ns_base_copy_jobid(orte_jobid_t **dest, orte_jobid_t *src, orte_data_type_t type);
+
+/*
+ * compare functions
+ */
+
+int orte_ns_base_compare_name(orte_process_name_t *value1,
+ orte_process_name_t *value2,
+ orte_data_type_t type);
+
+
+int orte_ns_base_compare_vpid(orte_vpid_t *value1,
+ orte_vpid_t *value2,
+ orte_data_type_t type);
+
+int orte_ns_base_compare_jobid(orte_jobid_t *value1,
+ orte_jobid_t *value2,
+ orte_data_type_t type);
+
+int orte_ns_base_compare_cellid(orte_cellid_t *value1,
+ orte_cellid_t *value2,
+ orte_data_type_t type);
+
+int orte_ns_base_compare_nodeid(orte_nodeid_t *value1,
+ orte_nodeid_t *value2,
+ orte_data_type_t type);
+
+/*
+ * size functions
+ */
+
+int orte_ns_base_std_size(size_t *size, void *src, orte_data_type_t type);
+
+/*
+ * release functions
+ */
+
+void orte_ns_base_std_release(orte_data_value_t *value);
+
+/*
+ * print functions
+ */
+
+int orte_ns_base_std_print(char **output, char *prefix, void *src, orte_data_type_t type);
+
+int orte_ns_base_print_name(char **output, char *prefix, orte_process_name_t *name, orte_data_type_t type);
+
+
+/*
+ * external API functions will be documented in the mca/ns/ns.h file
+ */
+
+#if defined(c_plusplus) || defined(__cplusplus)
+}
+#endif
+#endif
diff --git a/orte/mca/ns/ns.h b/orte/mca/ns/ns.h
index db893efb27..4086f61e6a 100644
--- a/orte/mca/ns/ns.h
+++ b/orte/mca/ns/ns.h
@@ -39,7 +39,7 @@
#include "orte/dss/dss.h"
#include "opal/mca/mca.h"
-#include "orte/mca/oob/oob_types.h"
+#include "orte/mca/rml/rml_types.h"
#include "ns_types.h"
@@ -59,26 +59,18 @@ typedef int (*orte_ns_base_module_init_fn_t)(void);
/**** CELL FUNCTIONS ****/
/**
* Create a new cell id.
- * The create_cellid() function allocates a new cell id for use by the caller.
- * The function checks to find the next available cell id, reserves it, and returns that
- * number. No memory for names is allocated by this process. The range of answers is from
- * 1 to MCA_NS_BASE_CELLID_MAX-1 (zero is reserved for the seed name and cannot therefore be
- * allocated).
+ * Allocates a new cell id for use by the caller. The function returns an
+ * existing cellid if the specified site/resource already has been assigned
+ * one.
*
* @param site The name of the site where the cell is located.
* @param resource The name of the resource associated with this cell (e.g., the name
* of the cluster).
- * @param cellid The numerical value of the allocated cell id. A value of
- * MCA_NS_BASE_CELLID_MAX indicates
- * that an error occurred - this represents a very unlikely
- * event meaning that the system ran out of cell id's. This probably indicates
- * an error in the calling program as the number of available cell id's is extremely large.
+ * @param cellid The location where the cellid is to be stored.
*
* @retval ORTE_SUCCESS A cellid was created and returned.
* @retval ORTE_ERROR_VALUE An error code indicative of the problem.
*
- * @code
- * new_cellid = ompi_name_server.create_cellid()
* @endcode
*/
typedef int (*orte_ns_base_module_create_cellid_fn_t)(orte_cellid_t *cellid,
@@ -97,34 +89,6 @@ typedef int (*orte_ns_base_module_create_cellid_fn_t)(orte_cellid_t *cellid,
typedef int (*orte_ns_base_module_get_cell_info_fn_t)(orte_cellid_t cellid,
char **site, char **resource);
-/**
- * Get the cell id for a process.
- * The cellid designator represents the physical location of the process - it is associated with
- * the hardware/system where the process is executing. Each process name contains this identifier
- * so that the system can issue commands (e.g., "die") to a collection of processes that are
- * executing on a common platform.
- *
- * Given that usage, it is necessary that the system have a way of telling a process its cellid.
- * The create_cellid() function is used by the system to associate a "cellid" identifier with
- * each platform. This function - assign_cellid_to_process() - is used to inform the process
- * of its cellid.
- *
- * Given a process name, this function will lookup its current platform and update the name with the
- * cellid.
- *
- * @param name Pointer to an ompi_process_name structure. The function will update the cellid
- * entry in the structure.
- *
- * @retval ORTE_SUCCESS Update was successful.
- * @retval OMPI_ERROR Update failed, most likely due to either a NULL process name pointer or the
- * inability to locate the process name in the lookup table.
- *
- * @code
- * return_value = ompi_name_server.assign_cellid_to_process(ompi_process_name_t* name);
- * @endcode
- */
-typedef int (*orte_ns_base_module_assign_cellid_to_process_fn_t)(orte_process_name_t* name);
-
/**
* Get the cell id as a character string.
* The get_cellid_string() function returns the cell id in a character string
@@ -181,50 +145,105 @@ typedef int (*orte_ns_base_module_get_cellid_string_fn_t)(char **cellid_string,
typedef int (*orte_ns_base_module_convert_string_to_cellid_fn_t)(orte_cellid_t *cellid, const char *cellidstring);
-/**
- * Get the cell id as a numberic value.
- * The get_cellid() function returns the cell id in a numeric representation -
- * i.e., in an integer form.
- *
- * @param *name A pointer to the name structure containing the name.
- *
- * @retval cellid The cell id field of the provided name.
- * @retval MCA_NS_BASE_CELLID_MAX Indicates that an error occurred - in this case, that
- * the name variable provided was NULL.
- *
- * @code
- * cellid = ompi_name_server.get_cellid(&name)
- * @endcode
+/**** NODE FUNCTIONS ****/
+/*
+ * Get an array of node id's
+ * Given the cell and a NULL-terminated array of names of nodes within it, this function assigns an id to represent
+ * each node within the cell.
*/
-typedef int (*orte_ns_base_module_get_cellid_fn_t)(orte_cellid_t *cellid, const orte_process_name_t* name);
+typedef int (*orte_ns_base_module_create_nodeids_fn_t)(orte_nodeid_t **nodes, orte_std_cntr_t *nnodes,
+ orte_cellid_t cellid, char **nodename);
+
+/*
+ * Get node info
+ * Retrieve the names of an array of nodes given their cellid and nodeids. The cellid
+ * is required as the nodeids are only unique within a given cell.
+ *
+ * @param cellid The id of the cell of the node.
+ * @param nodeids The ids of the node.
+ * @param nodenames Returns a pointer to a NULL-terminated array of strdup'd strings containing the node names.
+ * @retval ORTE_SUCCESS The nodename was created and returned.
+ * @retval ORTE_ERROR_VALUE An error code indicative of the problem.
+ */
+typedef int (*orte_ns_base_module_get_node_info_fn_t)(char ***nodename, orte_cellid_t cellid,
+ orte_std_cntr_t num_nodes, orte_nodeid_t *nodeids);
+
+/*
+ * Convert nodeid to character string
+ * Returns the nodeid in a character string representation. The string is created
+ * by expressing the provided nodeid in decimal. Memory for the string is
+ * allocated by the function - releasing that allocation is the responsibility of
+ * the calling program.
+ *
+ * @param nodeid The nodeid to be converted.
+ *
+ * @param *nodeid_string A pointer to a character string representation of the nodeid.
+ * @retval ORTE_SUCCESS The string was created and returned.
+ * @retval ORTE_ERROR_VALUE An error code indicative of the problem.
+ */
+typedef int (*orte_ns_base_module_convert_nodeid_to_string_fn_t)(char **nodeid_string, const orte_nodeid_t nodeid);
+
+/*
+ * Convert a string to a nodeid.
+ * Converts a characters string into a nodeid. The character string must be a
+ * decimal representation of a valid nodeid.
+ *
+ * @param nodeidstring The string to be converted.
+ *
+ * @param nodeid A pointer to a location where the resulting nodeid is to be stored.
+ * @retval ORTE_SUCCESS The string was created and returned.
+ * @retval ORTE_ERROR_VALUE An error code indicative of the problem.
+ */
+typedef int (*orte_ns_base_module_convert_string_to_nodeid_fn_t)(orte_nodeid_t *nodeid, const char *nodeidstring);
/**** JOB ID FUNCTIONS ****/
/**
* Create a new job id.
- * The create_jobid() function allocates a new job id for use by the caller.
- * The function checks to find the next available job id, reserves it, and returns that
- * number. No memory for names is allocated by this process. The range of answers is from
- * 1 to MCA_NS_BASE_JOBID_MAX-1 (zero is reserved for the seed name and cannot therefore be
- * allocated).
-
- *
+ * Allocate a new job id for use by the caller.
+ *
* The 0 job id is reserved for daemons within the system and will not be allocated.
* Developers should therefore assume that the daemon job id is automatically allocated
* and proceed to request names against it.
*
* @param None
- * @retval jobid The numerical value of the allocated job id. A value of
- * MCA_NS_BASE_JOBID_MAX indicates
- * that an error occurred - this represents a very unlikely
- * event meaning that the system ran out of job id's. This probably indicates
- * an error in the calling program as the number of available job id's is extremely large.
- *
- * @code
- * new_jobid = ompi_name_server.create_jobid()
- * @endcode
+ * @param jobid A pointer to the location where the jobid is to be returned.
+ * @param attrs A list of attributes that describe any conditions to be placed on
+ * the assigned jobid. For example, specifying USE_PARENT indicates that the specified
+ * jobid is to be identified as the parent of the new jobid. USE_ROOT indicates that
+ * the root of the job family of the specified jobid is to be identified as the parent.
*/
-typedef int (*orte_ns_base_module_create_jobid_fn_t)(orte_jobid_t *jobid);
+typedef int (*orte_ns_base_module_create_jobid_fn_t)(orte_jobid_t *jobid, opal_list_t *attrs);
+
+/*
+ * Get job descendants
+ * Given a jobid, return the array of jobids that descend from this one.
+ */
+typedef int (*orte_ns_base_module_get_job_descendants_fn_t)(orte_jobid_t** descendants,
+ orte_std_cntr_t *num_desc,
+ orte_jobid_t job);
+
+/*
+ * Get job children
+ * Given a jobid, return the array of jobids that are direct children of that job
+ */
+typedef int (*orte_ns_base_module_get_job_children_fn_t)(orte_jobid_t** children,
+ orte_std_cntr_t *num_childs,
+ orte_jobid_t job);
+
+/*
+ * Get root job from job family
+ * Given a jobid, return the jobid at the head of this job's family. If the jobid provided is the
+ * root for that family, that value will be returned.
+ */
+typedef int (*orte_ns_base_module_get_root_job_fn_t)(orte_jobid_t *root_job, orte_jobid_t job);
+
+/*
+ * Get parent jobid
+ * Given a jobid, return the parent job from which it descended. If the provided jobid is the
+ * root (i.e., has no parent), this function will return that same value.
+ */
+typedef int (*orte_ns_base_module_get_parent_job_fn_t)(orte_jobid_t *parent, orte_jobid_t job);
/**
* Reserve a range of process id's.
@@ -305,23 +324,6 @@ typedef int (*orte_ns_base_module_convert_jobid_to_string_fn_t)(char **jobid_str
*/
typedef int (*orte_ns_base_module_convert_string_to_jobid_fn_t)(orte_jobid_t *jobid, const char* jobidstring);
-/**
- * Get the job id as a numeric value.
- * The get_jobid() function returns the job id in a numeric representation -
- * i.e., in an integer form.
- *
- * @param *name A pointer to the name structure containing the name.
- *
- * @retval jobid The job id field of the provided name.
- * @retval MCA_NS_BASE_JOBID_MAX Indicates that an error occurred - in this case, that
- * the name variable provided was NULL.
- *
- * @code
- * jobid = ompi_name_server.get_jobid(&name)
- * @endcode
- */
-typedef int (*orte_ns_base_module_get_jobid_fn_t)(orte_jobid_t *jobid, const orte_process_name_t* name);
-
/**** NAME FUNCTIONS ****/
@@ -365,20 +367,6 @@ typedef int (*orte_ns_base_module_create_proc_name_fn_t)(orte_process_name_t **n
*/
typedef int (*orte_ns_base_module_create_my_name_fn_t)(void);
-/**
- * Make a copy of a process name.
- * Given a process name, this function creates a copy of it and returns a pointer
- * to the duplicate structure.
- *
- * @param *name Pointer to an existing process name structure.
- *
- * @retval *newname Pointer to the duplicate structure, with all fields transferred.
- * @retval NULL Indicates an error - most likely due to a NULL process name
- * pointer being supplied as input.
- */
-typedef int (*orte_ns_base_module_copy_proc_name_fn_t)(orte_process_name_t **dest,
- orte_process_name_t* src);
-
/**
* Convert a string representation to a process name.
* The convert_string_to_process_name() function converts a string representation of a process
@@ -399,30 +387,6 @@ typedef int (*orte_ns_base_module_convert_string_to_process_name_fn_t)(orte_proc
const char* name_string);
-/**
- * Free (release) a process name.
- * The free_name() function releases the process name from the "used" list
- * maintained within the name server for the jobid contained in the specified
- * name. The memory for the name is also released at that time.
- *
- * Name values are currently \em not re-used. Hence, free-ing a name
- * does not provide any noticeable benefit other than releasing the memory. In
- * the future, names may be re-used if this becomes desirable.
- *
- * @param *name A pointer to the name structure containing the name being released.
- *
- * @retval ORTE_SUCCESS Indicates the release was succesfully accomplished.
- * @retval OMPI_ERROR Indicates the release failed - most likely due to an
- * error when free-ing the memory allocation.
- *
- * @code
- * if (OMPI_ERROR == ompi_name_server.free_name(&name) {
- * report error
- * }
- * @endcode
- */
-typedef int (*orte_ns_base_module_free_name_fn_t)(orte_process_name_t **name);
-
/**
* Get the process name as a character string.
* The get_proc_name_string() function returns the entire process name in a
@@ -478,9 +442,9 @@ typedef int (*orte_ns_base_module_get_proc_name_string_fn_t)(char **name_string,
* result = ompi_name_server.compare(bit_mask, &name1, &name2)
* @endcode
*/
-typedef int (*orte_ns_base_module_compare_fn_t)(orte_ns_cmp_bitmask_t fields,
- const orte_process_name_t* name1,
- const orte_process_name_t* name2);
+typedef int (*orte_ns_base_module_compare_fields_fn_t)(orte_ns_cmp_bitmask_t fields,
+ const orte_process_name_t* name1,
+ const orte_process_name_t* name2);
/**** VPID FUNCTIONS ****/
@@ -539,22 +503,7 @@ typedef int (*orte_ns_base_module_get_vpid_string_fn_t)(char **vpid_string, cons
*/
typedef int (*orte_ns_base_module_convert_string_to_vpid_fn_t)(orte_vpid_t *vpid, const char* vpidstring);
-/**
- * Get the virtual process id as a numeric value.
- * The get_vpid() function returns the vpid in a numeric representation -
- * i.e., in an integer form.
- *
- * @param *name A pointer to the name structure containing the name.
- *
- * @retval vpid The vpid field of the provided name.
- * @retval MCA_NS_BASE_VPID_MAX Indicates that an error occurred - in this case, that
- * the name variable provided was NULL.
- *
- * @code
- * vpid = ompi_name_server.get_vpid(&name)
- * @endcode
- */
-typedef int (*orte_ns_base_module_get_vpid_fn_t)(orte_vpid_t *vpid, const orte_process_name_t *name);
+
/**** TAG SERVER ****/
/*
@@ -576,24 +525,33 @@ typedef int (*orte_ns_base_module_define_data_type_fn_t)(
/**** PEER RETRIEVAL ****/
-/*
- * Get my peers
+/**
+ * Get the process names of all processes in the specified conditions. It is
+ * sometimes necessary for a process to communicate to all processes of a
+ * given job, all processes in a given cell or on a given node, etc. The RML
+ * communication system utilizes the process name as its "pointer" for
+ * sending messages to another process. This function returns an array of
+ * process name pointers that contains the names of all processes that
+ * meet the specified combination of attributes.
+ *
+ * @param procs The location where the address of the array of pointers
+ * is to be stored. The function will dynamically allocate space for the
+ * array - the caller is responsible for releasing this space.
+ * @param num_procs The location where the number of entries in the
+ * returned array is to be stored.
+ * @param attributes A list of conditions to be used in defining the
+ * peers to be included in the returned array. This can include a
+ * request that all peers for the parent job be returned, for example.
+ * More common options would be to specify a cell or job.
+ *
+ * NOTE The combination of ORTE_CELLID_WILDCARD and ORTE_JOBID_WILDCARD
+ * in the attribute list will cause the function to return the names of *all*
+ * processes currently active in the universe.
*
- * THIS FUNCTION MAY BE ELIMINATED IN FUTURE VERSIONS TO REMOVE MULTIPLE STORAGE
- * OF O(N) ARRAYS IN THE SYSTEM
*/
typedef int (*orte_ns_base_module_get_peers_fn_t)(orte_process_name_t **procs,
- orte_std_cntr_t *num_procs, orte_std_cntr_t *self);
-
-/*
- * Get the list of peers from a specified job
- *
- * THIS FUNCTION MAY BE ELIMINATED IN FUTURE VERSIONS TO REMOVE MULTIPLE STORAGE
- * OF O(N) ARRAYS IN THE SYSTEM
- */
-typedef int (*orte_ns_base_module_get_job_peers_fn_t)(orte_process_name_t **procs,
- orte_std_cntr_t *num_procs, orte_jobid_t job);
-
+ orte_std_cntr_t *num_procs,
+ opal_list_t *attributes);
/*
@@ -609,55 +567,57 @@ typedef int (*orte_ns_base_module_dump_datatypes_fn_t)(void);
/*
- * Ver 1.0.0
+ * Ver 2.0
*/
-struct mca_ns_base_module_1_0_0_t {
+struct mca_ns_base_module_2_0_0_t {
/* init */
- orte_ns_base_module_init_fn_t init;
+ orte_ns_base_module_init_fn_t init;
/* cell functions */
- orte_ns_base_module_create_cellid_fn_t create_cellid;
- orte_ns_base_module_get_cellid_fn_t get_cellid;
- orte_ns_base_module_get_cell_info_fn_t get_cell_info;
- orte_ns_base_module_assign_cellid_to_process_fn_t assign_cellid_to_process;
- orte_ns_base_module_get_cellid_string_fn_t get_cellid_string;
- orte_ns_base_module_convert_cellid_to_string_fn_t convert_cellid_to_string;
- orte_ns_base_module_convert_string_to_cellid_fn_t convert_string_to_cellid;
+ orte_ns_base_module_create_cellid_fn_t create_cellid;
+ orte_ns_base_module_get_cell_info_fn_t get_cell_info;
+ orte_ns_base_module_get_cellid_string_fn_t get_cellid_string;
+ orte_ns_base_module_convert_cellid_to_string_fn_t convert_cellid_to_string;
+ orte_ns_base_module_convert_string_to_cellid_fn_t convert_string_to_cellid;
+ /** node functions */
+ orte_ns_base_module_create_nodeids_fn_t create_nodeids;
+ orte_ns_base_module_get_node_info_fn_t get_node_info;
+ orte_ns_base_module_convert_nodeid_to_string_fn_t convert_nodeid_to_string;
+ orte_ns_base_module_convert_string_to_nodeid_fn_t convert_string_to_nodeid;
/* jobid functions */
- orte_ns_base_module_create_jobid_fn_t create_jobid;
- orte_ns_base_module_get_jobid_fn_t get_jobid;
- orte_ns_base_module_get_jobid_string_fn_t get_jobid_string;
- orte_ns_base_module_convert_jobid_to_string_fn_t convert_jobid_to_string;
- orte_ns_base_module_convert_string_to_jobid_fn_t convert_string_to_jobid;
+ orte_ns_base_module_create_jobid_fn_t create_jobid;
+ orte_ns_base_module_get_job_descendants_fn_t get_job_descendants;
+ orte_ns_base_module_get_job_children_fn_t get_job_children;
+ orte_ns_base_module_get_root_job_fn_t get_root_job;
+ orte_ns_base_module_get_parent_job_fn_t get_parent_job;
+ orte_ns_base_module_get_jobid_string_fn_t get_jobid_string;
+ orte_ns_base_module_convert_jobid_to_string_fn_t convert_jobid_to_string;
+ orte_ns_base_module_convert_string_to_jobid_fn_t convert_string_to_jobid;
+ orte_ns_base_module_reserve_range_fn_t reserve_range;
/* vpid functions */
- orte_ns_base_module_reserve_range_fn_t reserve_range;
- orte_ns_base_module_get_vpid_fn_t get_vpid;
- orte_ns_base_module_get_vpid_string_fn_t get_vpid_string;
- orte_ns_base_module_convert_vpid_to_string_fn_t convert_vpid_to_string;
- orte_ns_base_module_convert_string_to_vpid_fn_t convert_string_to_vpid;
+ orte_ns_base_module_get_vpid_string_fn_t get_vpid_string;
+ orte_ns_base_module_convert_vpid_to_string_fn_t convert_vpid_to_string;
+ orte_ns_base_module_convert_string_to_vpid_fn_t convert_string_to_vpid;
/* name functions */
- orte_ns_base_module_create_proc_name_fn_t create_process_name;
- orte_ns_base_module_create_my_name_fn_t create_my_name;
- orte_ns_base_module_copy_proc_name_fn_t copy_process_name;
+ orte_ns_base_module_create_proc_name_fn_t create_process_name;
+ orte_ns_base_module_create_my_name_fn_t create_my_name;
orte_ns_base_module_convert_string_to_process_name_fn_t convert_string_to_process_name;
- orte_ns_base_module_free_name_fn_t free_name;
- orte_ns_base_module_get_proc_name_string_fn_t get_proc_name_string;
- orte_ns_base_module_compare_fn_t compare;
+ orte_ns_base_module_get_proc_name_string_fn_t get_proc_name_string;
+ orte_ns_base_module_compare_fields_fn_t compare_fields;
/* peer functions */
- orte_ns_base_module_get_peers_fn_t get_peers;
- orte_ns_base_module_get_job_peers_fn_t get_job_peers;
+ orte_ns_base_module_get_peers_fn_t get_peers;
/* tag server functions */
- orte_ns_base_module_assign_rml_tag_fn_t assign_rml_tag;
+ orte_ns_base_module_assign_rml_tag_fn_t assign_rml_tag;
/* data type functions */
- orte_ns_base_module_define_data_type_fn_t define_data_type;
+ orte_ns_base_module_define_data_type_fn_t define_data_type;
/* diagnostic functions */
- orte_ns_base_module_dump_cells_fn_t dump_cells;
- orte_ns_base_module_dump_jobs_fn_t dump_jobs;
- orte_ns_base_module_dump_tags_fn_t dump_tags;
- orte_ns_base_module_dump_datatypes_fn_t dump_datatypes;
+ orte_ns_base_module_dump_cells_fn_t dump_cells;
+ orte_ns_base_module_dump_jobs_fn_t dump_jobs;
+ orte_ns_base_module_dump_tags_fn_t dump_tags;
+ orte_ns_base_module_dump_datatypes_fn_t dump_datatypes;
};
-typedef struct mca_ns_base_module_1_0_0_t mca_ns_base_module_1_0_0_t;
-typedef mca_ns_base_module_1_0_0_t mca_ns_base_module_t;
+typedef struct mca_ns_base_module_2_0_0_t mca_ns_base_module_2_0_0_t;
+typedef mca_ns_base_module_2_0_0_t mca_ns_base_module_t;
/*
* NS Component
@@ -677,26 +637,26 @@ typedef int (*mca_ns_base_component_finalize_fn_t)(void);
* the standard component data structure
*/
-struct mca_ns_base_component_1_0_0_t {
+struct mca_ns_base_component_2_0_0_t {
mca_base_component_t ns_version;
mca_base_component_data_1_0_0_t ns_data;
mca_ns_base_component_init_fn_t ns_init;
mca_ns_base_component_finalize_fn_t ns_finalize;
};
-typedef struct mca_ns_base_component_1_0_0_t mca_ns_base_component_1_0_0_t;
-typedef mca_ns_base_component_1_0_0_t mca_ns_base_component_t;
+typedef struct mca_ns_base_component_2_0_0_t mca_ns_base_component_2_0_0_t;
+typedef mca_ns_base_component_2_0_0_t mca_ns_base_component_t;
/*
- * Macro for use in components that are of type ns v1.0.0
+ * Macro for use in components that are of type ns v2.0.0
*/
-#define MCA_NS_BASE_VERSION_1_0_0 \
- /* ns v1.0 is chained to MCA v1.0 */ \
+#define MCA_NS_BASE_VERSION_2_0_0 \
+ /* ns v2.0 is chained to MCA v1.0 */ \
MCA_BASE_VERSION_1_0_0, \
- /* ns v1.0 */ \
- "ns", 1, 0, 0
+ /* ns v2.0 */ \
+ "ns", 2, 0, 0
/* Global structure for accessing name server functions
*/
diff --git a/orte/mca/ns/ns_types.h b/orte/mca/ns/ns_types.h
index 0042e0ffbf..72a9650ef3 100644
--- a/orte/mca/ns/ns_types.h
+++ b/orte/mca/ns/ns_types.h
@@ -47,10 +47,20 @@
extern "C" {
#endif
+/**** NS ATTRIBUTES ****/
+#define ORTE_NS_USE_PARENT "orte-ns-use-parent"
+#define ORTE_NS_USE_ROOT "orte-ns-use-root"
+#define ORTE_NS_USE_CELL "orte-ns-use-cell"
+#define ORTE_NS_USE_JOBID "orte-ns-use-job"
+#define ORTE_NS_USE_NODE "orte-ns-use-node"
+#define ORTE_NS_INCLUDE_DESCENDANTS "orte-ns-include-desc"
+#define ORTE_NS_INCLUDE_CHILDREN "orte-ns-include-child"
+
+
#define ORTE_NAME_ARGS(n) \
- (unsigned long) ((NULL == n) ? -1 : (int32_t)(n)->cellid), \
- (unsigned long) ((NULL == n) ? -1 : (int32_t)(n)->jobid), \
- (unsigned long) ((NULL == n) ? -1 : (int32_t)(n)->vpid)
+ (long) ((NULL == n) ? (long)-1 : (long)(n)->cellid), \
+ (long) ((NULL == n) ? (long)-1 : (long)(n)->jobid), \
+ (long) ((NULL == n) ? (long)-1 : (long)(n)->vpid)
/*
@@ -69,18 +79,19 @@ extern "C" {
/** Set the allowed range for ids in each space
*
* NOTE: Be sure to update the ORTE_NAME_ARGS #define (above) and all
- * uses of it if these types change to be larger than (unsigned long)!
+ * uses of it if these types change to be larger than (long)!
*/
typedef orte_std_cntr_t orte_jobid_t;
typedef orte_std_cntr_t orte_cellid_t;
+typedef orte_std_cntr_t orte_nodeid_t;
typedef orte_std_cntr_t orte_vpid_t;
+
typedef uint8_t orte_ns_cmp_bitmask_t; /**< Bit mask for comparing process names */
-typedef uint8_t orte_ns_cmd_flag_t;
struct orte_process_name_t {
- orte_cellid_t cellid; /**< Cell number */
- orte_jobid_t jobid; /**< Job number */
- orte_vpid_t vpid; /**< Process number */
+ orte_cellid_t cellid; /**< Cell number */
+ orte_jobid_t jobid; /**< Job number */
+ orte_vpid_t vpid; /**< Process number */
};
typedef struct orte_process_name_t orte_process_name_t;
@@ -90,6 +101,15 @@ typedef struct orte_process_name_t orte_process_name_t;
#define ORTE_CELLID_MAX ORTE_STD_CNTR_MAX
#define ORTE_JOBID_MAX ORTE_STD_CNTR_MAX
#define ORTE_VPID_MAX ORTE_STD_CNTR_MAX
+#define ORTE_NODEID_MAX ORTE_STD_CNTR_MAX
+
+/*
+ * define minimum value for id's in any field
+ */
+#define ORTE_CELLID_MIN ORTE_STD_CNTR_MIN
+#define ORTE_JOBID_MIN ORTE_STD_CNTR_MIN
+#define ORTE_VPID_MIN ORTE_STD_CNTR_MIN
+#define ORTE_NODEID_MIN ORTE_STD_CNTR_MIN
/*
* define invalid values
@@ -97,19 +117,31 @@ typedef struct orte_process_name_t orte_process_name_t;
#define ORTE_CELLID_INVALID -999
#define ORTE_JOBID_INVALID -999
#define ORTE_VPID_INVALID -999
+#define ORTE_NODEID_INVALID -999
/*
- * define wildcard values
+ * define wildcard values (should be -1)
*/
#define ORTE_CELLID_WILDCARD -1
#define ORTE_JOBID_WILDCARD -1
#define ORTE_VPID_WILDCARD -1
+#define ORTE_NODEID_WILDCARD -1
-ORTE_DECLSPEC extern orte_process_name_t orte_name_all;
-#define ORTE_NAME_ALL &orte_name_all
+/*
+ * Shortcut for some commonly used names
+ */
+
+#define ORTE_NAME_WILDCARD &orte_ns_name_wildcard
+ORTE_DECLSPEC extern orte_process_name_t orte_ns_name_wildcard; /** instantiated in orte/mca/ns/base/ns_base_open.c */
+
+#define ORTE_NAME_INVALID &orte_ns_name_invalid
+ORTE_DECLSPEC extern orte_process_name_t orte_ns_name_invalid; /** instantiated in orte/mca/ns/base/ns_base_open.c */
#define ORTE_PROC_MY_NAME orte_process_info.my_name
+#define ORTE_PROC_MY_HNP &orte_ns_name_my_hnp
+ORTE_DECLSPEC extern orte_process_name_t orte_ns_name_my_hnp; /** instantiated in orte/mca/ns/base/ns_base_open.c */
+
/**
* Convert process name from host to network byte order.
*
diff --git a/orte/mca/ns/proxy/Makefile.am b/orte/mca/ns/proxy/Makefile.am
index 3f1498cf81..143c49770b 100644
--- a/orte/mca/ns/proxy/Makefile.am
+++ b/orte/mca/ns/proxy/Makefile.am
@@ -16,37 +16,34 @@
# $HEADER$
#
-# Use the top-level Makefile.options
-
-
-
-sources =
-include src/Makefile.extra
+sources = \
+ ns_proxy.h \
+ ns_proxy_cell_fns.c \
+ ns_proxy_diag_fns.c \
+ ns_proxy_general_fns.c \
+ ns_proxy_job_fns.c \
+ ns_proxy_component.c
# Make the output library in this directory, and name it either
# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
# (for static builds).
if OMPI_BUILD_ns_proxy_DSO
-lib =
-lib_sources =
-component = mca_ns_proxy.la
-component_sources = $(sources)
+component_noinst =
+component_install = mca_ns_proxy.la
else
-lib = libmca_ns_proxy.la
-lib_sources = $(sources)
-component =
-component_sources =
+component_noinst = libmca_ns_proxy.la
+component_install =
endif
mcacomponentdir = $(libdir)/openmpi
-mcacomponent_LTLIBRARIES = $(component)
-mca_ns_proxy_la_SOURCES = $(component_sources)
+mcacomponent_LTLIBRARIES = $(component_install)
+mca_ns_proxy_la_SOURCES = $(sources)
mca_ns_proxy_la_LDFLAGS = -module -avoid-version
mca_ns_proxy_la_LIBADD = \
$(top_ompi_builddir)/orte/liborte.la \
$(top_ompi_builddir)/opal/libopal.la
-noinst_LTLIBRARIES = $(lib)
-libmca_ns_proxy_la_SOURCES = $(lib_sources)
+noinst_LTLIBRARIES = $(component_noinst)
+libmca_ns_proxy_la_SOURCES =$(sources)
libmca_ns_proxy_la_LDFLAGS = -module -avoid-version
diff --git a/orte/mca/ns/proxy/configure.params b/orte/mca/ns/proxy/configure.params
index 4686b02ce4..a0ccbf55aa 100644
--- a/orte/mca/ns/proxy/configure.params
+++ b/orte/mca/ns/proxy/configure.params
@@ -19,5 +19,6 @@
# Specific to this module
-PARAM_INIT_FILE=src/ns_proxy.c
+PARAM_INIT_FILE=ns_proxy.c
+PARAM_CONFIG_HEADER_FILE="ns_proxy.h"
PARAM_CONFIG_FILES="Makefile"
diff --git a/orte/mca/ns/proxy/src/ns_proxy.h b/orte/mca/ns/proxy/ns_proxy.h
similarity index 75%
rename from orte/mca/ns/proxy/src/ns_proxy.h
rename to orte/mca/ns/proxy/ns_proxy.h
index f7c4375ac7..10fea146f1 100644
--- a/orte/mca/ns/proxy/src/ns_proxy.h
+++ b/orte/mca/ns/proxy/ns_proxy.h
@@ -22,27 +22,20 @@
#include "orte_config.h"
-#include "opal/types.h"
#include "orte/orte_constants.h"
+
+#include "opal/types.h"
#include "opal/class/opal_list.h"
+
#include "orte/dss/dss.h"
-#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/ns.h"
+#include "orte/mca/ns/base/ns_private.h"
#if defined(c_plusplus) || defined(__cplusplus)
extern "C" {
#endif
-struct orte_ns_proxy_cell_info_t {
- opal_object_t super;
- orte_cellid_t cellid;
- char *site;
- char *resource;
-};
-typedef struct orte_ns_proxy_cell_info_t orte_ns_proxy_cell_info_t;
-
-OBJ_CLASS_DECLARATION(orte_ns_proxy_cell_info_t);
-
struct orte_ns_proxy_tagitem_t {
opal_object_t super;
orte_rml_tag_t tag; /**< OOB tag */
@@ -81,7 +74,6 @@ int orte_ns_proxy_finalize(void);
*/
typedef struct {
size_t max_size, block_size;
- orte_process_name_t *my_replica;
int debug;
orte_cellid_t num_cells;
orte_pointer_array_t *cells;
@@ -94,6 +86,12 @@ typedef struct {
extern orte_ns_proxy_globals_t orte_ns_proxy;
+/*
+ * simplifying define
+ */
+#define ORTE_NS_MY_REPLICA orte_process_info.ns_replica
+
+
/*
* proxy function prototypes
*/
@@ -101,13 +99,26 @@ int orte_ns_proxy_create_cellid(orte_cellid_t *cellid, char *site, char *resourc
int orte_ns_proxy_get_cell_info(orte_cellid_t cellid, char **site, char **resource);
-int orte_ns_proxy_create_jobid(orte_jobid_t *jobid);
+int orte_ns_proxy_create_nodeids(orte_nodeid_t **nodeids, orte_std_cntr_t *nnodes,
+ orte_cellid_t cellid, char **nodenames);
+
+int orte_ns_proxy_get_node_info(char ***nodename, orte_cellid_t cellid, orte_std_cntr_t num_nodes, orte_nodeid_t *nodeids);
+
+int orte_ns_proxy_create_jobid(orte_jobid_t *jobid, opal_list_t *attrs);
+
+int orte_ns_proxy_get_job_descendants(orte_jobid_t** descendants, orte_std_cntr_t *ndesc, orte_jobid_t job);
+
+int orte_ns_proxy_get_job_children(orte_jobid_t** descendants, orte_std_cntr_t *ndesc, orte_jobid_t job);
+
+int orte_ns_proxy_get_root_job(orte_jobid_t *root_job, orte_jobid_t job);
+
+int orte_ns_proxy_get_parent_job(orte_jobid_t *parent, orte_jobid_t job);
int orte_ns_proxy_reserve_range(orte_jobid_t job, orte_vpid_t range,
orte_vpid_t *startvpid);
-int orte_ns_proxy_get_job_peers(orte_process_name_t **procs,
- orte_std_cntr_t *num_procs, orte_jobid_t job);
+int orte_ns_proxy_get_peers(orte_process_name_t **procs,
+ orte_std_cntr_t *num_procs, opal_list_t *attrs);
int orte_ns_proxy_assign_rml_tag(orte_rml_tag_t *tag, char *name);
diff --git a/orte/mca/ns/proxy/ns_proxy_cell_fns.c b/orte/mca/ns/proxy/ns_proxy_cell_fns.c
new file mode 100644
index 0000000000..f6738b9e18
--- /dev/null
+++ b/orte/mca/ns/proxy/ns_proxy_cell_fns.c
@@ -0,0 +1,440 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2006 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+
+#include "orte_config.h"
+
+#include <string.h>
+
+#include "orte/orte_constants.h"
+#include "orte/orte_types.h"
+
+#include "opal/mca/mca.h"
+#include "opal/util/argv.h"
+#include "opal/util/output.h"
+#include "opal/util/trace.h"
+
+#include "orte/dss/dss.h"
+#include "orte/mca/errmgr/errmgr.h"
+#include "orte/mca/rml/rml.h"
+
+#include "ns_proxy.h"
+
+/**
+ * globals
+ */
+
+/*
+ * functions
+ */
+
+int orte_ns_proxy_create_cellid(orte_cellid_t *cellid, char *site, char *resource)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count;
+ int rc;
+
+ OPAL_TRACE(1);
+
+ /* set the default value of error */
+ *cellid = ORTE_CELLID_INVALID;
+
+ command = ORTE_NS_CREATE_CELLID_CMD;
+
+ cmd = OBJ_NEW(orte_buffer_t);
+ if (cmd == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &site, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &resource, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ answer = OBJ_NEW(orte_buffer_t);
+ if(answer == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ if (ORTE_NS_CREATE_CELLID_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, cellid, &count, ORTE_CELLID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+ OBJ_RELEASE(answer);
+
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_proxy_get_cell_info(orte_cellid_t cellid,
+ char **site, char **resource)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count;
+ int rc, ret=ORTE_SUCCESS;
+
+ OPAL_TRACE(1);
+
+ command = ORTE_NS_GET_CELL_INFO_CMD;
+
+ cmd = OBJ_NEW(orte_buffer_t);
+ if (cmd == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &cellid, 1, ORTE_CELLID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ answer = OBJ_NEW(orte_buffer_t);
+ if(answer == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (ORTE_NS_GET_CELL_INFO_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, site, &count, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, resource, &count, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_proxy_create_nodeids(orte_nodeid_t **nodeids, orte_std_cntr_t *nnodes,
+ orte_cellid_t cellid, char **nodenames)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count, index;
+ int rc;
+
+ OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
+
+ command = ORTE_NS_CREATE_NODEID_CMD;
+
+ cmd = OBJ_NEW(orte_buffer_t);
+ if (cmd == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &cellid, 1, ORTE_CELLID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ count = opal_argv_count(nodenames);
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &count, 1, ORTE_STD_CNTR))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, nodenames, count, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ answer = OBJ_NEW(orte_buffer_t);
+ if(answer == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ if (ORTE_NS_CREATE_NODEID_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &index, &count, ORTE_STD_CNTR))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ /** allocate the space for the nodeids */
+ *nodeids = (orte_nodeid_t*)malloc(index * sizeof(orte_nodeid_t));
+ if (NULL == *nodeids) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, nodeids, &index, ORTE_NODEID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+ OBJ_RELEASE(answer);
+
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_proxy_get_node_info(char ***nodenames, orte_cellid_t cellid,
+ orte_std_cntr_t num_nodes, orte_nodeid_t *nodeids)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count, index;
+ int rc, ret=ORTE_SUCCESS;
+
+ OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
+
+ command = ORTE_NS_GET_NODE_INFO_CMD;
+
+ cmd = OBJ_NEW(orte_buffer_t);
+ if (cmd == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &cellid, 1, ORTE_CELLID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &num_nodes, 1, ORTE_STD_CNTR))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, nodeids, num_nodes, ORTE_NODEID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ answer = OBJ_NEW(orte_buffer_t);
+ if(answer == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (ORTE_NS_GET_NODE_INFO_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &index, &count, ORTE_STD_CNTR))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ /** create the space for the nodenames */
+ *nodenames = (char**)malloc(index * sizeof(char*));
+ if (NULL == *nodenames) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, *nodenames, &index, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &ret, &count, ORTE_INT))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ret;
+}
diff --git a/orte/mca/ns/proxy/src/ns_proxy_component.c b/orte/mca/ns/proxy/ns_proxy_component.c
similarity index 84%
rename from orte/mca/ns/proxy/src/ns_proxy_component.c
rename to orte/mca/ns/proxy/ns_proxy_component.c
index 2ce677cf56..fc438b636b 100644
--- a/orte/mca/ns/proxy/src/ns_proxy_component.c
+++ b/orte/mca/ns/proxy/ns_proxy_component.c
@@ -47,7 +47,7 @@
*/
mca_ns_base_component_t mca_ns_proxy_component = {
{
- MCA_NS_BASE_VERSION_1_0_0,
+ MCA_NS_BASE_VERSION_2_0_0,
"proxy", /* MCA module name */
ORTE_MAJOR_VERSION, /* MCA module major version */
@@ -71,35 +71,37 @@ static mca_ns_base_module_t orte_ns_proxy_module = {
orte_ns_proxy_module_init,
/* cell functions */
orte_ns_proxy_create_cellid,
- orte_ns_base_get_cellid,
orte_ns_proxy_get_cell_info,
- orte_ns_base_assign_cellid_to_process,
orte_ns_base_get_cellid_string,
orte_ns_base_convert_cellid_to_string,
orte_ns_base_convert_string_to_cellid,
+ /** node functions */
+ orte_ns_proxy_create_nodeids,
+ orte_ns_proxy_get_node_info,
+ orte_ns_base_convert_nodeid_to_string,
+ orte_ns_base_convert_string_to_nodeid,
/* jobid functions */
orte_ns_proxy_create_jobid,
- orte_ns_base_get_jobid,
+ orte_ns_proxy_get_job_descendants,
+ orte_ns_proxy_get_job_children,
+ orte_ns_proxy_get_root_job,
+ orte_ns_proxy_get_parent_job,
orte_ns_base_get_jobid_string,
orte_ns_base_convert_jobid_to_string,
orte_ns_base_convert_string_to_jobid,
- /* vpid functions */
orte_ns_proxy_reserve_range,
- orte_ns_base_get_vpid,
+ /* vpid functions */
orte_ns_base_get_vpid_string,
orte_ns_base_convert_vpid_to_string,
orte_ns_base_convert_string_to_vpid,
/* name functions */
orte_ns_base_create_process_name,
orte_ns_proxy_create_my_name,
- orte_ns_base_copy_process_name,
orte_ns_base_convert_string_to_process_name,
- orte_ns_base_free_name,
orte_ns_base_get_proc_name_string,
- orte_ns_base_compare,
+ orte_ns_base_compare_fields,
/* peer functions */
- orte_ns_base_get_peers,
- orte_ns_proxy_get_job_peers,
+ orte_ns_proxy_get_peers,
/* tag server functions */
orte_ns_proxy_assign_rml_tag,
/* data type functions */
@@ -116,31 +118,6 @@ static mca_ns_base_module_t orte_ns_proxy_module = {
*/
static bool initialized = false;
-/* constructor - used to initialize state of cell info list instance */
-static void orte_ns_proxy_cell_info_construct(orte_ns_proxy_cell_info_t* ptr)
-{
- ptr->resource = NULL;
- ptr->site = NULL;
-}
-
-/* destructor - used to free any resources held by instance */
-static void orte_ns_proxy_cell_info_destructor(orte_ns_proxy_cell_info_t* ptr)
-{
- if (NULL != ptr->resource) {
- free(ptr->resource);
- }
- if (NULL != ptr->site) {
- free(ptr->site);
- }
-}
-
-/* define instance of opal_class_t */
-OBJ_CLASS_INSTANCE(
- orte_ns_proxy_cell_info_t, /* type name */
- opal_object_t, /* parent "class" name */
- orte_ns_proxy_cell_info_construct, /* constructor */
- orte_ns_proxy_cell_info_destructor); /* destructor */
-
/* constructor - used to initialize state of taglist instance */
static void orte_ns_proxy_tagitem_construct(orte_ns_proxy_tagitem_t* tagitem)
{
@@ -245,14 +222,10 @@ mca_ns_base_module_t* orte_ns_proxy_init(int *priority)
ORTE_ERROR_LOG(ret);
return NULL;
}
- if(ORTE_SUCCESS != (ret = orte_ns.copy_process_name(&orte_process_info.ns_replica, &name))) {
+ if(ORTE_SUCCESS != (ret = orte_dss.copy((void**)&orte_process_info.ns_replica, &name, ORTE_NAME))) {
ORTE_ERROR_LOG(ret);
return NULL;
}
- if (ORTE_SUCCESS != orte_ns_base_copy_process_name(&orte_ns_proxy.my_replica,
- orte_process_info.ns_replica)) { /* can't operate */
- return NULL;
- }
/* initialize the cell info tracker */
if (ORTE_SUCCESS != (rc = orte_pointer_array_init(&(orte_ns_proxy.cells),
@@ -315,7 +288,6 @@ int orte_ns_proxy_module_init(void)
*/
int orte_ns_proxy_finalize(void)
{
- orte_ns_proxy_cell_info_t **cptr;
orte_ns_proxy_tagitem_t **tag;
orte_ns_proxy_dti_t **dti;
orte_std_cntr_t i;
@@ -323,14 +295,6 @@ int orte_ns_proxy_finalize(void)
/* free all tracking storage, but only if this component was initialized */
if (initialized) {
- cptr = (orte_ns_proxy_cell_info_t**)(orte_ns_proxy.cells)->addr;
- for (i=0; i < (orte_ns_proxy.cells)->size; i++) {
- if (NULL != cptr[i]) {
- OBJ_RELEASE(cptr[i]);
- }
- }
- OBJ_RELEASE(orte_ns_proxy.cells);
-
tag = (orte_ns_proxy_tagitem_t**)(orte_ns_proxy.tags)->addr;
for (i=0; i < (orte_ns_proxy.tags)->size; i++) {
if (NULL != tag[i]) OBJ_RELEASE(tag[i]);
diff --git a/orte/mca/ns/proxy/ns_proxy_diag_fns.c b/orte/mca/ns/proxy/ns_proxy_diag_fns.c
new file mode 100644
index 0000000000..5cfce73c65
--- /dev/null
+++ b/orte/mca/ns/proxy/ns_proxy_diag_fns.c
@@ -0,0 +1,313 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2006 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+
+#include "orte_config.h"
+
+#include <string.h>
+
+#include "orte/orte_constants.h"
+#include "orte/orte_types.h"
+
+#include "opal/mca/mca.h"
+#include "opal/util/output.h"
+#include "opal/util/trace.h"
+
+#include "orte/dss/dss.h"
+#include "orte/mca/errmgr/errmgr.h"
+#include "orte/mca/rml/rml.h"
+
+#include "orte/mca/ns/base/base.h"
+#include "ns_proxy.h"
+
+/*
+ * DIAGNOSTIC functions
+ */
+int orte_ns_proxy_dump_cells(void)
+{
+ orte_buffer_t cmd;
+ orte_buffer_t answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count;
+ int rc;
+
+ command = ORTE_NS_DUMP_CELLS_CMD;
+
+ OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
+
+ /* dump name service replica cell tracker */
+ OBJ_CONSTRUCT(&cmd, orte_buffer_t);
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, &command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ OBJ_DESTRUCT(&cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, &cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_DESTRUCT(&cmd);
+
+ OBJ_CONSTRUCT(&answer, orte_buffer_t);
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, &answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(&answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&answer);
+ return rc;
+ }
+
+ if (ORTE_NS_DUMP_CELLS_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&answer))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&answer);
+ return rc;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_proxy_dump_jobs(void)
+{
+ orte_buffer_t cmd;
+ orte_buffer_t answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count;
+ int rc;
+
+ command = ORTE_NS_DUMP_JOBIDS_CMD;
+
+ OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
+
+ /* dump name service replica jobid tracker */
+ OBJ_CONSTRUCT(&cmd, orte_buffer_t);
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, &command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ OBJ_DESTRUCT(&cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, &cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_DESTRUCT(&cmd);
+
+ OBJ_CONSTRUCT(&answer, orte_buffer_t);
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, &answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(&answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&answer);
+ return rc;
+ }
+
+ if (ORTE_NS_DUMP_JOBIDS_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&answer))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&answer);
+ return rc;
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_proxy_dump_tags(void)
+{
+ orte_buffer_t cmd;
+ orte_buffer_t answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t i;
+ orte_std_cntr_t count;
+ orte_rml_tag_t j;
+ orte_ns_proxy_tagitem_t **ptr;
+ int rc;
+
+ command = ORTE_NS_DUMP_TAGS_CMD;
+
+ OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
+
+ /* dump name service replica tag tracker */
+ OBJ_CONSTRUCT(&cmd, orte_buffer_t);
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, &command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ OBJ_DESTRUCT(&cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, &cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_DESTRUCT(&cmd);
+
+ OBJ_CONSTRUCT(&answer, orte_buffer_t);
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, &answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(&answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&answer);
+ return rc;
+ }
+
+ if (ORTE_NS_DUMP_TAGS_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&answer))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&answer);
+ return rc;
+ }
+
+ /* dump local tag tracker */
+ opal_output(mca_ns_base_output, "\n\n[%lu,%lu,%lu] Dump of Local Tag Tracker\n",
+ ORTE_NAME_ARGS(orte_process_info.my_name));
+ ptr = (orte_ns_proxy_tagitem_t**)(orte_ns_proxy.tags)->addr;
+ for (i=0, j=0; j < orte_ns_proxy.num_tags &&
+ i < (orte_ns_proxy.tags)->size; i++) {
+ if (NULL != ptr[i]) {
+ j++;
+ opal_output(mca_ns_base_output, "Num: %lu\tTag: %lu\tTag name: %s\n",
+ (unsigned long)j, (unsigned long)ptr[i]->tag, ptr[i]->name);
+ }
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_proxy_dump_datatypes(void)
+{
+ orte_buffer_t cmd;
+ orte_buffer_t answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t i, j;
+ orte_std_cntr_t count;
+ orte_ns_proxy_dti_t **ptr;
+ int rc;
+
+ command = ORTE_NS_DUMP_DATATYPES_CMD;
+
+ OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
+
+ /* dump name service replica datatype tracker */
+ OBJ_CONSTRUCT(&cmd, orte_buffer_t);
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, &command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ OBJ_DESTRUCT(&cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, &cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_DESTRUCT(&cmd);
+
+ OBJ_CONSTRUCT(&answer, orte_buffer_t);
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, &answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(&answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&answer);
+ return rc;
+ }
+
+ if (ORTE_NS_DUMP_DATATYPES_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_DESTRUCT(&answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&answer))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&answer);
+ return rc;
+ }
+
+ /* dump local datatype tracker */
+ opal_output(mca_ns_base_output, "\n\n[%lu,%lu,%lu] Dump of Local Datatype Tracker\n",
+ ORTE_NAME_ARGS(orte_process_info.my_name));
+ ptr = (orte_ns_proxy_dti_t**)(orte_ns_proxy.dts)->addr;
+ for (i=0, j=0; j < orte_ns_proxy.num_dts &&
+ i < (orte_ns_proxy.dts)->size; i++) {
+ if (NULL != ptr[i]) {
+ j++;
+ opal_output(mca_ns_base_output, "Num: %lu\tDatatype id: %lu\tDatatype name: %s\n",
+ (unsigned long)j, (unsigned long)ptr[i]->id, ptr[i]->name);
+ }
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
diff --git a/orte/mca/ns/proxy/ns_proxy_general_fns.c b/orte/mca/ns/proxy/ns_proxy_general_fns.c
new file mode 100644
index 0000000000..c2250eee6a
--- /dev/null
+++ b/orte/mca/ns/proxy/ns_proxy_general_fns.c
@@ -0,0 +1,495 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2006 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+
+#include "orte_config.h"
+
+#include <string.h>
+
+#include "orte/orte_constants.h"
+#include "orte/orte_types.h"
+
+#include "opal/mca/mca.h"
+#include "opal/util/output.h"
+#include "opal/util/trace.h"
+
+#include "orte/dss/dss.h"
+#include "orte/mca/errmgr/errmgr.h"
+#include "orte/mca/rmgr/rmgr.h"
+#include "orte/mca/rml/rml.h"
+
+#include "ns_proxy.h"
+
+/*
+ * PEER functions
+ */
+int orte_ns_proxy_get_peers(orte_process_name_t **procs,
+ orte_std_cntr_t *num_procs, opal_list_t *attrs)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count, nprocs, i;
+ orte_cellid_t *cptr;
+ orte_attribute_t *attr;
+ int rc;
+
+ OPAL_TRACE(1);
+
+ /* set default value */
+ *procs = NULL;
+ *num_procs = 0;
+
+ /* check the attributes to see if USE_JOB or USE_CELL has been set. If not, then this is
+ * a request for my own job peers - process that one locally
+ */
+
+ /* if the cell is given AND it matches my own, then we can process this
+ * quickly. Otherwise, we have to do some more work.
+ *
+ * RHC: when we go multi-cell, we need a way to find all the cells upon
+ * which a job is executing so we can make this work!
+ */
+ if (NULL != (attr = orte_rmgr.find_attribute(attrs, ORTE_NS_USE_CELL))) {
+ if (ORTE_SUCCESS != (rc = orte_dss.get((void**)&cptr, attr->value, ORTE_CELLID))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+ if (*cptr != ORTE_PROC_MY_NAME->cellid && *cptr != ORTE_CELLID_WILDCARD) {
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_IMPLEMENTED);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_NOT_IMPLEMENTED;
+ }
+ }
+
+ if (NULL == (attr = orte_rmgr.find_attribute(attrs, ORTE_NS_USE_JOBID))) {
+ /* get my own job peers, assuming all are on this cell - process here
+ *
+ * RHC: This is a bad assumption. When we go multi-cell, we are going to have to process
+ * get peer requests solely on the HNP since we won't know the cellid otherwise
+ */
+ *procs = (orte_process_name_t*)malloc(orte_process_info.num_procs * sizeof(orte_process_name_t));
+ if (NULL == *procs) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ for (i=0; i < orte_process_info.num_procs; i++) {
+ (*procs)[i].cellid = ORTE_PROC_MY_NAME->cellid;
+ (*procs)[i].jobid = ORTE_PROC_MY_NAME->jobid;
+ (*procs)[i].vpid = orte_process_info.vpid_start + i;
+ }
+
+ *num_procs = orte_process_info.num_procs;
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_SUCCESS;
+ }
+
+ /* non-local request for peers in another job - send to replica for processing */
+ if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ command = ORTE_NS_GET_PEERS_CMD;
+ /* pack the command */
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) { /* got a problem */
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ /* pack the attributes */
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, attrs, 1, ORTE_ATTR_LIST))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (ORTE_NS_GET_PEERS_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &nprocs, &count, ORTE_STD_CNTR))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ /* allocate space for array of proc names */
+ if (0 < nprocs) {
+ *procs = (orte_process_name_t*)malloc((nprocs) * sizeof(orte_process_name_t));
+ if (NULL == *procs) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, *procs, &nprocs, ORTE_NAME))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+ }
+ *num_procs = nprocs;
+
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_proxy_assign_rml_tag(orte_rml_tag_t *tag,
+ char *name)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_ns_proxy_tagitem_t* tagitem, **tags;
+ orte_std_cntr_t count, i;
+ orte_rml_tag_t j;
+ int rc;
+
+ OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
+
+ if (NULL != name) {
+ /* see if this name is already in list - if so, return tag */
+ tags = (orte_ns_proxy_tagitem_t**)orte_ns_proxy.tags->addr;
+ for (i=0, j=0; j < orte_ns_proxy.num_tags &&
+ i < (orte_ns_proxy.tags)->size; i++) {
+ if (NULL != tags[i]) {
+ j++;
+ if (tags[i]->name != NULL &&
+ 0 == strcmp(name, tags[i]->name)) { /* found name on list */
+ *tag = tags[i]->tag;
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_SUCCESS;
+ }
+ }
+ }
+ }
+
+ /* okay, not on local list - so go get one from tag server */
+ command = ORTE_NS_ASSIGN_OOB_TAG_CMD;
+ *tag = ORTE_RML_TAG_MAX; /* set the default error value */
+
+ if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (NULL == name) {
+ name = "NULL";
+ }
+
+ if (0 > (rc = orte_dss.pack(cmd, &name, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (ORTE_NS_ASSIGN_OOB_TAG_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, tag, &count, ORTE_RML_TAG))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ OBJ_RELEASE(answer);
+
+ /* add the new tag to the local list so we don't have to get it again */
+ tagitem = OBJ_NEW(orte_ns_proxy_tagitem_t);
+ if (NULL == tagitem) { /* out of memory */
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ if (ORTE_SUCCESS != (rc = orte_pointer_array_add(&i,
+ orte_ns_proxy.tags, tagitem))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+ tagitem->tag = *tag;
+ (orte_ns_proxy.num_tags)++;
+ if (NULL != name) { /* provided - can look it up later */
+ tagitem->name = strdup(name);
+ } else {
+ tagitem->name = NULL;
+ }
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+
+ /* all done */
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_proxy_define_data_type(const char *name,
+ orte_data_type_t *type)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_ns_proxy_dti_t **dti, *dtip;
+ orte_std_cntr_t count, i, j;
+ int rc=ORTE_SUCCESS;
+
+ if (NULL == name || 0 < *type) {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
+
+ /* first, check to see if name is already on local list
+ * if so, return id, ensure registered with dss
+ */
+ dti = (orte_ns_proxy_dti_t**)orte_ns_proxy.dts->addr;
+ for (i=0, j=0; j < orte_ns_proxy.num_dts &&
+ i < orte_ns_proxy.dts->size; i++) {
+ if (NULL != dti[i]) {
+ j++;
+ if (dti[i]->name != NULL &&
+ 0 == strcmp(name, dti[i]->name)) { /* found name on list */
+ *type = dti[i]->id;
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_SUCCESS;
+ }
+ }
+ }
+
+
+ /* okay, not on local list - so go get one from tag server */
+ command = ORTE_NS_DEFINE_DATA_TYPE_CMD;
+ *type = ORTE_DSS_ID_MAX; /* set the default error value */
+
+ if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&name, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+
+ if (ORTE_NS_ASSIGN_OOB_TAG_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, type, &count, ORTE_DATA_TYPE))) {
+ ORTE_ERROR_LOG(ORTE_ERR_UNPACK_FAILURE);
+ OBJ_RELEASE(answer);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_UNPACK_FAILURE;
+ }
+ OBJ_RELEASE(answer);
+
+ /* add the new id to the local list so we don't have to get it again */
+ dtip = OBJ_NEW(orte_ns_proxy_dti_t);
+ if (NULL == dtip) { /* out of memory */
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ dtip->name = strdup(name);
+ if (ORTE_SUCCESS != (rc = orte_pointer_array_add(&i,
+ orte_ns_proxy.dts, dtip))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+ return rc;
+ }
+ dtip->id = *type;
+ (orte_ns_proxy.num_dts)++;
+
+ OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
+
+ /* all done */
+ return rc;
+}
+
+/*
+ * Take advantage of the way the RML uses the process name as its index into
+ * the RML communicator table. Because the RML needs a name right away, it will
+ * automatically assign us one when it receives a message - and it communicates
+ * that assignment back to us automatically. Thus, to get a name for ourselves,
+ * all we have to do is send a message! No response from the replica is required.
+ */
+int orte_ns_proxy_create_my_name(void)
+{
+ orte_buffer_t* cmd;
+ orte_ns_cmd_flag_t command;
+ int rc;
+
+ command = ORTE_NS_CREATE_MY_NAME_CMD;
+
+ cmd = OBJ_NEW(orte_buffer_t);
+ if (cmd == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ return ORTE_SUCCESS;
+}
+
diff --git a/orte/mca/ns/proxy/ns_proxy_job_fns.c b/orte/mca/ns/proxy/ns_proxy_job_fns.c
new file mode 100644
index 0000000000..181d94ccc9
--- /dev/null
+++ b/orte/mca/ns/proxy/ns_proxy_job_fns.c
@@ -0,0 +1,526 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2006 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+
+#include "orte_config.h"
+
+#include <string.h>
+
+#include "orte/orte_constants.h"
+#include "orte/orte_types.h"
+
+#include "opal/mca/mca.h"
+#include "opal/util/output.h"
+#include "opal/util/trace.h"
+
+#include "orte/dss/dss.h"
+#include "orte/mca/errmgr/errmgr.h"
+#include "orte/mca/rml/rml.h"
+
+#include "ns_proxy.h"
+
+/**** CREATE JOBID ****/
+int orte_ns_proxy_create_jobid(orte_jobid_t *job, opal_list_t *attrs)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count;
+ int rc;
+
+ OPAL_TRACE(1);
+
+ /* set default value */
+ *job = ORTE_JOBID_INVALID;
+
+ if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ command = ORTE_NS_CREATE_JOBID_CMD;
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, attrs, 1, ORTE_ATTR_LIST))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ if (ORTE_NS_CREATE_JOBID_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, job, &count, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ OBJ_RELEASE(answer);
+ return ORTE_SUCCESS;
+}
+
+
+/**** GET JOB DESCENDANTS ****/
+int orte_ns_proxy_get_job_descendants(orte_jobid_t **descendants, orte_std_cntr_t *num_desc, orte_jobid_t job)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count, ndesc=0;
+ orte_jobid_t *jobs=NULL;
+ int rc;
+
+ OPAL_TRACE(1);
+
+ /* set default response */
+ *descendants = NULL;
+ *num_desc = 0;
+
+ if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ command = ORTE_NS_GET_JOB_DESC_CMD;
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) { /* got a problem */
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&job, 1, ORTE_JOBID))) { /* got a problem */
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ if (ORTE_NS_GET_JOB_DESC_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &ndesc, &count, ORTE_STD_CNTR))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ /* if there are any descendants, allocate space for them and unpack */
+ if (0 < ndesc) {
+ jobs = (orte_jobid_t*)malloc(ndesc * sizeof(orte_jobid_t));
+ if (NULL == jobs) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ count = ndesc;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, jobs, &count, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+ }
+
+ OBJ_RELEASE(answer);
+
+ *descendants = jobs;
+ *num_desc = count;
+
+ return ORTE_SUCCESS;
+}
+
+/**** GET JOB CHILDREN ****/
+int orte_ns_proxy_get_job_children(orte_jobid_t **descendants, orte_std_cntr_t *num_desc, orte_jobid_t job)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count, ndesc=0;
+ orte_jobid_t *jobs=NULL;
+ int rc;
+
+ OPAL_TRACE(1);
+
+ /* set default response */
+ *descendants = NULL;
+ *num_desc = 0;
+
+ if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ command = ORTE_NS_GET_JOB_CHILD_CMD;
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) { /* got a problem */
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&job, 1, ORTE_JOBID))) { /* got a problem */
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ if (ORTE_NS_GET_JOB_DESC_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &ndesc, &count, ORTE_STD_CNTR))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ /* if there are any descendants, allocate space for them and unpack */
+ if (0 < ndesc) {
+ jobs = (orte_jobid_t*)malloc(ndesc * sizeof(orte_jobid_t));
+ if (NULL == jobs) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ count = ndesc;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, jobs, &count, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+ }
+
+ OBJ_RELEASE(answer);
+
+ *descendants = jobs;
+ *num_desc = count;
+
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_proxy_get_root_job(orte_jobid_t *root_job, orte_jobid_t job)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count;
+ int rc;
+
+ OPAL_TRACE(1);
+
+ /* set default value */
+ *root_job = ORTE_JOBID_INVALID;
+
+ if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ command = ORTE_NS_GET_ROOT_JOB_CMD;
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &job, 1, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ if (ORTE_NS_GET_ROOT_JOB_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, root_job, &count, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ OBJ_RELEASE(answer);
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_proxy_get_parent_job(orte_jobid_t *parent, orte_jobid_t job)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count;
+ int rc;
+
+ OPAL_TRACE(1);
+
+ /* set default value */
+ *parent = ORTE_JOBID_INVALID;
+
+ if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ command = ORTE_NS_GET_PARENT_JOB_CMD;
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &job, 1, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+ if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ if (ORTE_NS_GET_PARENT_JOB_CMD != command) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, parent, &count, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ OBJ_RELEASE(answer);
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_proxy_reserve_range(orte_jobid_t job, orte_vpid_t range, orte_vpid_t *starting_vpid)
+{
+ orte_buffer_t* cmd;
+ orte_buffer_t* answer;
+ orte_ns_cmd_flag_t command;
+ orte_std_cntr_t count;
+ int rc;
+
+ OPAL_TRACE(1);
+
+ /* set default return value */
+ *starting_vpid = ORTE_VPID_INVALID;
+
+ if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ command = ORTE_NS_RESERVE_RANGE_CMD;
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) { /* got a problem */
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&job, 1, ORTE_JOBID))) { /* got a problem */
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&range, 1, ORTE_VPID))) { /* got a problem */
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
+ if (0 > orte_rml.send_buffer(ORTE_NS_MY_REPLICA, cmd, ORTE_RML_TAG_NS, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(cmd);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+ OBJ_RELEASE(cmd);
+
+
+ if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ if (0 > orte_rml.recv_buffer(ORTE_NS_MY_REPLICA, answer, ORTE_RML_TAG_NS)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ OBJ_RELEASE(answer);
+ return ORTE_ERR_COMM_FAILURE;
+ }
+
+ count = 1;
+ if ((ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD)))
+ || (ORTE_NS_RESERVE_RANGE_CMD != command)) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, starting_vpid, &count, ORTE_VPID))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(answer);
+ return rc;
+ }
+ OBJ_RELEASE(answer);
+ return ORTE_SUCCESS;
+}
+
diff --git a/orte/mca/ns/proxy/src/Makefile.extra b/orte/mca/ns/proxy/src/Makefile.extra
deleted file mode 100644
index 8e963be1ce..0000000000
--- a/orte/mca/ns/proxy/src/Makefile.extra
+++ /dev/null
@@ -1,23 +0,0 @@
-# -*- makefile -*-
-#
-# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
-# University Research and Technology
-# Corporation. All rights reserved.
-# Copyright (c) 2004-2005 The University of Tennessee and The University
-# of Tennessee Research Foundation. All rights
-# reserved.
-# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
-# University of Stuttgart. All rights reserved.
-# Copyright (c) 2004-2005 The Regents of the University of California.
-# All rights reserved.
-# $COPYRIGHT$
-#
-# Additional copyrights may follow
-#
-# $HEADER$
-#
-
-sources += \
- src/ns_proxy.h \
- src/ns_proxy.c \
- src/ns_proxy_component.c
diff --git a/orte/mca/ns/proxy/src/ns_proxy.c b/orte/mca/ns/proxy/src/ns_proxy.c
deleted file mode 100644
index 2cdbabfd2e..0000000000
--- a/orte/mca/ns/proxy/src/ns_proxy.c
+++ /dev/null
@@ -1,1071 +0,0 @@
-/*
- * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
- * University Research and Technology
- * Corporation. All rights reserved.
- * Copyright (c) 2004-2006 The University of Tennessee and The University
- * of Tennessee Research Foundation. All rights
- * reserved.
- * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
- * University of Stuttgart. All rights reserved.
- * Copyright (c) 2004-2005 The Regents of the University of California.
- * All rights reserved.
- * $COPYRIGHT$
- *
- * Additional copyrights may follow
- *
- * $HEADER$
- */
-/** @file:
- *
- */
-
-#include "orte_config.h"
-
-#include <string.h>
-
-#include "orte/orte_constants.h"
-#include "orte/orte_types.h"
-
-#include "opal/mca/mca.h"
-#include "opal/util/output.h"
-#include "opal/util/trace.h"
-
-#include "orte/dss/dss.h"
-#include "orte/mca/errmgr/errmgr.h"
-#include "orte/mca/rml/rml.h"
-
-#include "ns_proxy.h"
-
-/**
- * globals
- */
-
-/*
- * functions
- */
-
-int orte_ns_proxy_create_cellid(orte_cellid_t *cellid, char *site, char *resource)
-{
- orte_buffer_t* cmd;
- orte_buffer_t* answer;
- orte_ns_cmd_flag_t command;
- orte_std_cntr_t count, index;
- int rc;
- orte_ns_proxy_cell_info_t *new_cell;
-
- OPAL_TRACE(1);
-
- /* set the default value of error */
- *cellid = ORTE_CELLID_MAX;
-
- command = ORTE_NS_CREATE_CELLID_CMD;
-
- cmd = OBJ_NEW(orte_buffer_t);
- if (cmd == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- return rc;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &site, 1, ORTE_STRING))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- return rc;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &resource, 1, ORTE_STRING))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(cmd);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_RELEASE(cmd);
-
- answer = OBJ_NEW(orte_buffer_t);
- if(answer == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (0 > orte_rml.recv_buffer(orte_ns_proxy.my_replica, answer, ORTE_RML_TAG_NS)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- return rc;
- }
-
- if (ORTE_NS_CREATE_CELLID_CMD != command) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, cellid, &count, ORTE_CELLID))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- return rc;
- }
- OBJ_RELEASE(answer);
-
- /* store the info locally for later retrieval */
- OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
- new_cell = OBJ_NEW(orte_ns_proxy_cell_info_t);
- if (NULL == new_cell) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
- if (ORTE_SUCCESS != (rc = orte_pointer_array_add(&index,
- orte_ns_proxy.cells, new_cell))) {
- ORTE_ERROR_LOG(rc);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
- if (NULL != site) {
- new_cell->site = strdup(site);
- }
- if (NULL != resource) {
- new_cell->resource = strdup(resource);
- }
-
- new_cell->cellid = orte_ns_proxy.num_cells;
- *cellid = new_cell->cellid;
- (orte_ns_proxy.num_cells)++;
-
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_proxy_get_cell_info(orte_cellid_t cellid,
- char **site, char **resource)
-{
- orte_buffer_t* cmd;
- orte_buffer_t* answer;
- orte_ns_cmd_flag_t command;
- orte_cellid_t j;
- orte_std_cntr_t i, count, index;
- orte_ns_proxy_cell_info_t **cell, *new_cell;
- int rc, ret=ORTE_SUCCESS;
-
- OPAL_TRACE(1);
-
- /* see if we already have the info locally */
- OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
-
- cell = (orte_ns_proxy_cell_info_t**)(orte_ns_proxy.cells)->addr;
- for (i=0, j=0; j < orte_ns_proxy.num_cells &&
- i < (orte_ns_proxy.cells)->size; i++) {
- if (NULL != cell[i]) {
- j++;
- if (cellid == cell[i]->cellid) {
- *site = strdup(cell[i]->site);
- *resource = strdup(cell[i]->resource);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_SUCCESS;
- }
- }
- }
-
- /* okay, don't have it locally - go ask for it */
-
- command = ORTE_NS_GET_CELL_INFO_CMD;
-
- cmd = OBJ_NEW(orte_buffer_t);
- if (cmd == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &cellid, 1, ORTE_CELLID))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_RELEASE(cmd);
-
- answer = OBJ_NEW(orte_buffer_t);
- if(answer == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (0 > orte_rml.recv_buffer(orte_ns_proxy.my_replica, answer, ORTE_RML_TAG_NS)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- if (ORTE_NS_GET_CELL_INFO_CMD != command) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, site, &count, ORTE_STRING))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, resource, &count, ORTE_STRING))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &ret, &count, ORTE_STD_CNTR))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- if (ORTE_SUCCESS == ret) {
- /* remote operation worked - store the info locally for any subsequent requests */
- new_cell = OBJ_NEW(orte_ns_proxy_cell_info_t);
- if (NULL == new_cell) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
- if (ORTE_SUCCESS != (rc = orte_pointer_array_add(&index,
- orte_ns_proxy.cells, new_cell))) {
- ORTE_ERROR_LOG(rc);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
- if (NULL != site) {
- new_cell->site = strdup(*site);
- }
- if (NULL != resource) {
- new_cell->resource = strdup(*resource);
- }
-
- new_cell->cellid = cellid;
- }
-
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ret;
-}
-
-int orte_ns_proxy_create_jobid(orte_jobid_t *job)
-{
- orte_buffer_t* cmd;
- orte_buffer_t* answer;
- orte_ns_cmd_flag_t command;
- orte_std_cntr_t count;
- int rc;
-
- OPAL_TRACE(1);
-
- /* set default value */
- *job = ORTE_JOBID_MAX;
-
- if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- command = ORTE_NS_CREATE_JOBID_CMD;
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) { /* got a problem */
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(cmd);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_RELEASE(cmd);
-
- if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OBJ_RELEASE(answer);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
- if (0 > orte_rml.recv_buffer(orte_ns_proxy.my_replica, answer, ORTE_RML_TAG_NS)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- return rc;
- }
-
- if (ORTE_NS_CREATE_JOBID_CMD != command) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, job, &count, ORTE_JOBID))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- return rc;
- }
-
- OBJ_RELEASE(answer);
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_proxy_reserve_range(orte_jobid_t job, orte_vpid_t range, orte_vpid_t *starting_vpid)
-{
- orte_buffer_t* cmd;
- orte_buffer_t* answer;
- orte_ns_cmd_flag_t command;
- orte_std_cntr_t count;
- int rc;
-
- OPAL_TRACE(1);
-
- /* set default return value */
- *starting_vpid = ORTE_VPID_MAX;
-
- if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- command = ORTE_NS_RESERVE_RANGE_CMD;
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) { /* got a problem */
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- return rc;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&job, 1, ORTE_JOBID))) { /* got a problem */
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- return rc;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&range, 1, ORTE_VPID))) { /* got a problem */
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(cmd);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_RELEASE(cmd);
-
-
- if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (0 > orte_rml.recv_buffer(orte_ns_proxy.my_replica, answer, ORTE_RML_TAG_NS)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if ((ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD)))
- || (ORTE_NS_RESERVE_RANGE_CMD != command)) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- return rc;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, starting_vpid, &count, ORTE_VPID))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- return rc;
- }
- OBJ_RELEASE(answer);
- return ORTE_SUCCESS;
-}
-
-
-/*
- * PEER functions
- */
-int orte_ns_proxy_get_job_peers(orte_process_name_t **procs,
- orte_std_cntr_t *num_procs, orte_jobid_t job)
-{
- orte_buffer_t* cmd;
- orte_buffer_t* answer;
- orte_ns_cmd_flag_t command;
- orte_std_cntr_t count, nprocs;
- int rc;
-
- OPAL_TRACE_ARG1(1, job);
-
- /* set default value */
- *procs = NULL;
- *num_procs = 0;
-
- if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- command = ORTE_NS_GET_JOB_PEERS_CMD;
- /* pack the command */
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) { /* got a problem */
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- return rc;
- }
-
- /* pack the jobid */
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &job, 1, ORTE_JOBID))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(cmd);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_RELEASE(cmd);
-
- if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OBJ_RELEASE(answer);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (0 > orte_rml.recv_buffer(orte_ns_proxy.my_replica, answer, ORTE_RML_TAG_NS)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- return rc;
- }
-
- if (ORTE_NS_GET_JOB_PEERS_CMD != command) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &nprocs, &count, ORTE_STD_CNTR))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- return rc;
- }
-
- /* allocate space for array of proc names */
- if (0 < nprocs) {
- *procs = (orte_process_name_t*)malloc((nprocs) * sizeof(orte_process_name_t));
- if (NULL == *procs) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OBJ_RELEASE(answer);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, procs, &nprocs, ORTE_NAME))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- return rc;
- }
- }
- *num_procs = nprocs;
-
- OBJ_RELEASE(answer);
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_proxy_assign_rml_tag(orte_rml_tag_t *tag,
- char *name)
-{
- orte_buffer_t* cmd;
- orte_buffer_t* answer;
- orte_ns_cmd_flag_t command;
- orte_ns_proxy_tagitem_t* tagitem, **tags;
- orte_std_cntr_t count, i;
- orte_rml_tag_t j;
- int rc;
-
- OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
-
- if (NULL != name) {
- /* see if this name is already in list - if so, return tag */
- tags = (orte_ns_proxy_tagitem_t**)orte_ns_proxy.tags->addr;
- for (i=0, j=0; j < orte_ns_proxy.num_tags &&
- i < (orte_ns_proxy.tags)->size; i++) {
- if (NULL != tags[i]) {
- j++;
- if (tags[i]->name != NULL &&
- 0 == strcmp(name, tags[i]->name)) { /* found name on list */
- *tag = tags[i]->tag;
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_SUCCESS;
- }
- }
- }
- }
-
- /* okay, not on local list - so go get one from tag server */
- command = ORTE_NS_ASSIGN_OOB_TAG_CMD;
- *tag = ORTE_RML_TAG_MAX; /* set the default error value */
-
- if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- if (NULL == name) {
- name = "NULL";
- }
-
- if (0 > (rc = orte_dss.pack(cmd, &name, 1, ORTE_STRING))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_RELEASE(cmd);
-
- if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (0 > orte_rml.recv_buffer(orte_ns_proxy.my_replica, answer, ORTE_RML_TAG_NS)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- if (ORTE_NS_ASSIGN_OOB_TAG_CMD != command) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, tag, &count, ORTE_RML_TAG))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- OBJ_RELEASE(answer);
-
- /* add the new tag to the local list so we don't have to get it again */
- tagitem = OBJ_NEW(orte_ns_proxy_tagitem_t);
- if (NULL == tagitem) { /* out of memory */
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
- if (ORTE_SUCCESS != (rc = orte_pointer_array_add(&i,
- orte_ns_proxy.tags, tagitem))) {
- ORTE_ERROR_LOG(rc);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
- tagitem->tag = *tag;
- (orte_ns_proxy.num_tags)++;
- if (NULL != name) { /* provided - can look it up later */
- tagitem->name = strdup(name);
- } else {
- tagitem->name = NULL;
- }
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
-
- /* all done */
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_proxy_define_data_type(const char *name,
- orte_data_type_t *type)
-{
- orte_buffer_t* cmd;
- orte_buffer_t* answer;
- orte_ns_cmd_flag_t command;
- orte_ns_proxy_dti_t **dti, *dtip;
- orte_std_cntr_t count, i, j;
- int rc=ORTE_SUCCESS;
-
- if (NULL == name || 0 < *type) {
- ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
- return ORTE_ERR_BAD_PARAM;
- }
-
- OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
-
- /* first, check to see if name is already on local list
- * if so, return id, ensure registered with dss
- */
- dti = (orte_ns_proxy_dti_t**)orte_ns_proxy.dts->addr;
- for (i=0, j=0; j < orte_ns_proxy.num_dts &&
- i < orte_ns_proxy.dts->size; i++) {
- if (NULL != dti[i]) {
- j++;
- if (dti[i]->name != NULL &&
- 0 == strcmp(name, dti[i]->name)) { /* found name on list */
- *type = dti[i]->id;
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_SUCCESS;
- }
- }
- }
-
-
- /* okay, not on local list - so go get one from tag server */
- command = ORTE_NS_DEFINE_DATA_TYPE_CMD;
- *type = ORTE_DSS_ID_MAX; /* set the default error value */
-
- if ((cmd = OBJ_NEW(orte_buffer_t)) == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, (void*)&name, 1, ORTE_STRING))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_RELEASE(cmd);
-
- if ((answer = OBJ_NEW(orte_buffer_t)) == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (0 > orte_rml.recv_buffer(orte_ns_proxy.my_replica, answer, ORTE_RML_TAG_NS)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, &command, &count, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
-
- if (ORTE_NS_ASSIGN_OOB_TAG_CMD != command) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(answer, type, &count, ORTE_DATA_TYPE))) {
- ORTE_ERROR_LOG(ORTE_ERR_UNPACK_FAILURE);
- OBJ_RELEASE(answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_UNPACK_FAILURE;
- }
- OBJ_RELEASE(answer);
-
- /* add the new id to the local list so we don't have to get it again */
- dtip = OBJ_NEW(orte_ns_proxy_dti_t);
- if (NULL == dtip) { /* out of memory */
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
- dtip->name = strdup(name);
- if (ORTE_SUCCESS != (rc = orte_pointer_array_add(&i,
- orte_ns_proxy.dts, dtip))) {
- ORTE_ERROR_LOG(rc);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return rc;
- }
- dtip->id = *type;
- (orte_ns_proxy.num_dts)++;
-
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
-
- /* all done */
- return rc;
-}
-
-/*
- * Take advantage of the way the RML uses the process name as its index into
- * the RML communicator table. Because the RML needs a name right away, it will
- * automatically assign us one when it receives a message - and it communicates
- * that assignment back to us automatically. Thus, to get a name for ourselves,
- * all we have to do is send a message! No response from the replica is required.
- */
-int orte_ns_proxy_create_my_name(void)
-{
- orte_buffer_t* cmd;
- orte_ns_cmd_flag_t command;
- int rc;
-
- command = ORTE_NS_CREATE_MY_NAME_CMD;
-
- cmd = OBJ_NEW(orte_buffer_t);
- if (cmd == NULL) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(cmd);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_RELEASE(cmd);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_RELEASE(cmd);
-
- return ORTE_SUCCESS;
-}
-
-/*
- * DIAGNOSTIC functions
- */
-int orte_ns_proxy_dump_cells(void)
-{
- orte_buffer_t cmd;
- orte_buffer_t answer;
- orte_ns_cmd_flag_t command;
- orte_std_cntr_t i;
- orte_cellid_t j;
- orte_ns_proxy_cell_info_t **ptr;
- int rc;
-
- command = ORTE_NS_DUMP_CELLS_CMD;
-
- OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
-
- /* dump name service replica cell tracker */
- OBJ_CONSTRUCT(&cmd, orte_buffer_t);
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, &command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- OBJ_DESTRUCT(&cmd);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, &cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_DESTRUCT(&cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_DESTRUCT(&cmd);
-
- OBJ_CONSTRUCT(&answer, orte_buffer_t);
- if (0 > orte_rml.recv_buffer(orte_ns_proxy.my_replica, &answer, ORTE_RML_TAG_NS)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_DESTRUCT(&answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&answer))) {
- ORTE_ERROR_LOG(rc);
- OBJ_DESTRUCT(&answer);
- return rc;
- }
-
- /* dump local cell tracker */
- opal_output(mca_ns_base_output, "\n\n[%lu,%lu,%lu] Dump of Local Cell Tracker\n",
- ORTE_NAME_ARGS(orte_process_info.my_name));
- ptr = (orte_ns_proxy_cell_info_t**)(orte_ns_proxy.cells)->addr;
- for (i=0, j=0; j < orte_ns_proxy.num_cells &&
- i < (orte_ns_proxy.cells)->size; i++) {
- if (NULL != ptr[i]) {
- j++;
- opal_output(mca_ns_base_output, "Num: %lu\tCell: %lu\n",
- (unsigned long)j, (unsigned long)ptr[i]->cellid);
- }
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_proxy_dump_jobs(void)
-{
- orte_buffer_t cmd;
- orte_buffer_t answer;
- orte_ns_cmd_flag_t command;
- int rc;
-
- command = ORTE_NS_DUMP_JOBIDS_CMD;
-
- OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
-
- /* dump name service replica jobid tracker */
- OBJ_CONSTRUCT(&cmd, orte_buffer_t);
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, &command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- OBJ_DESTRUCT(&cmd);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, &cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_DESTRUCT(&cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_DESTRUCT(&cmd);
-
- OBJ_CONSTRUCT(&answer, orte_buffer_t);
- if (0 > orte_rml.recv_buffer(orte_ns_proxy.my_replica, &answer, ORTE_RML_TAG_NS)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_DESTRUCT(&answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&answer))) {
- ORTE_ERROR_LOG(rc);
- OBJ_DESTRUCT(&answer);
- return rc;
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_proxy_dump_tags(void)
-{
- orte_buffer_t cmd;
- orte_buffer_t answer;
- orte_ns_cmd_flag_t command;
- orte_std_cntr_t i;
- orte_rml_tag_t j;
- orte_ns_proxy_tagitem_t **ptr;
- int rc;
-
- command = ORTE_NS_DUMP_TAGS_CMD;
-
- OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
-
- /* dump name service replica tag tracker */
- OBJ_CONSTRUCT(&cmd, orte_buffer_t);
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, &command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- OBJ_DESTRUCT(&cmd);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, &cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_DESTRUCT(&cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_DESTRUCT(&cmd);
-
- OBJ_CONSTRUCT(&answer, orte_buffer_t);
- if (0 > orte_rml.recv_buffer(orte_ns_proxy.my_replica, &answer, ORTE_RML_TAG_NS)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_DESTRUCT(&answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&answer))) {
- ORTE_ERROR_LOG(rc);
- OBJ_DESTRUCT(&answer);
- return rc;
- }
-
- /* dump local tag tracker */
- opal_output(mca_ns_base_output, "\n\n[%lu,%lu,%lu] Dump of Local Tag Tracker\n",
- ORTE_NAME_ARGS(orte_process_info.my_name));
- ptr = (orte_ns_proxy_tagitem_t**)(orte_ns_proxy.tags)->addr;
- for (i=0, j=0; j < orte_ns_proxy.num_tags &&
- i < (orte_ns_proxy.tags)->size; i++) {
- if (NULL != ptr[i]) {
- j++;
- opal_output(mca_ns_base_output, "Num: %lu\tTag: %lu\tTag name: %s\n",
- (unsigned long)j, (unsigned long)ptr[i]->tag, ptr[i]->name);
- }
- }
-
- return ORTE_SUCCESS;
-}
-
-
-int orte_ns_proxy_dump_datatypes(void)
-{
- orte_buffer_t cmd;
- orte_buffer_t answer;
- orte_ns_cmd_flag_t command;
- orte_std_cntr_t i, j;
- orte_ns_proxy_dti_t **ptr;
- int rc;
-
- command = ORTE_NS_DUMP_DATATYPES_CMD;
-
- OPAL_THREAD_LOCK(&orte_ns_proxy.mutex);
-
- /* dump name service replica datatype tracker */
- OBJ_CONSTRUCT(&cmd, orte_buffer_t);
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, &command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- OBJ_DESTRUCT(&cmd);
- return rc;
- }
-
- if (0 > orte_rml.send_buffer(orte_ns_proxy.my_replica, &cmd, ORTE_RML_TAG_NS, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_DESTRUCT(&cmd);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
- OBJ_DESTRUCT(&cmd);
-
- OBJ_CONSTRUCT(&answer, orte_buffer_t);
- if (0 > orte_rml.recv_buffer(orte_ns_proxy.my_replica, &answer, ORTE_RML_TAG_NS)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- OBJ_DESTRUCT(&answer);
- OPAL_THREAD_UNLOCK(&orte_ns_proxy.mutex);
- return ORTE_ERR_COMM_FAILURE;
- }
-
- if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&answer))) {
- ORTE_ERROR_LOG(rc);
- OBJ_DESTRUCT(&answer);
- return rc;
- }
-
- /* dump local datatype tracker */
- opal_output(mca_ns_base_output, "\n\n[%lu,%lu,%lu] Dump of Local Datatype Tracker\n",
- ORTE_NAME_ARGS(orte_process_info.my_name));
- ptr = (orte_ns_proxy_dti_t**)(orte_ns_proxy.dts)->addr;
- for (i=0, j=0; j < orte_ns_proxy.num_dts &&
- i < (orte_ns_proxy.dts)->size; i++) {
- if (NULL != ptr[i]) {
- j++;
- opal_output(mca_ns_base_output, "Num: %lu\tDatatype id: %lu\tDatatype name: %s\n",
- (unsigned long)j, (unsigned long)ptr[i]->id, ptr[i]->name);
- }
- }
-
- return ORTE_SUCCESS;
-}
-
-
diff --git a/orte/mca/ns/replica/Makefile.am b/orte/mca/ns/replica/Makefile.am
index 0f19235d50..9adba2abb1 100644
--- a/orte/mca/ns/replica/Makefile.am
+++ b/orte/mca/ns/replica/Makefile.am
@@ -16,37 +16,37 @@
# $HEADER$
#
-# Use the top-level Makefile.options
-
-
-
-sources =
-include src/Makefile.extra
+sources = \
+ ns_replica.h \
+ ns_replica_class_instances.h \
+ ns_replica_cell_fns.c \
+ ns_replica_diag_fns.c \
+ ns_replica_general_fns.c \
+ ns_replica_job_fns.c \
+ ns_replica_recv.c \
+ ns_replica_support_fns.c \
+ ns_replica_component.c
# Make the output library in this directory, and name it either
# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
# (for static builds).
if OMPI_BUILD_ns_replica_DSO
-lib =
-lib_sources =
-component = mca_ns_replica.la
-component_sources = $(sources)
+component_noinst =
+component_install = mca_ns_replica.la
else
-lib = libmca_ns_replica.la
-lib_sources = $(sources)
-component =
-component_sources =
+component_noinst = libmca_ns_replica.la
+component_install =
endif
mcacomponentdir = $(libdir)/openmpi
-mcacomponent_LTLIBRARIES = $(component)
-mca_ns_replica_la_SOURCES = $(component_sources)
+mcacomponent_LTLIBRARIES = $(component_install)
+mca_ns_replica_la_SOURCES = $(sources)
mca_ns_replica_la_LDFLAGS = -module -avoid-version
mca_ns_replica_la_LIBADD = \
$(top_ompi_builddir)/orte/liborte.la \
$(top_ompi_builddir)/opal/libopal.la
-noinst_LTLIBRARIES = $(lib)
-libmca_ns_replica_la_SOURCES = $(lib_sources)
+noinst_LTLIBRARIES = $(component_noinst)
+libmca_ns_replica_la_SOURCES =$(sources)
libmca_ns_replica_la_LDFLAGS = -module -avoid-version
diff --git a/orte/mca/ns/replica/configure.params b/orte/mca/ns/replica/configure.params
index 33baded7ae..3d04f54b19 100644
--- a/orte/mca/ns/replica/configure.params
+++ b/orte/mca/ns/replica/configure.params
@@ -19,5 +19,6 @@
# Specific to this module
-PARAM_INIT_FILE=src/ns_replica.c
+PARAM_INIT_FILE=ns_replica.c
+PARAM_CONFIG_HEADER_FILE="ns_replica.h"
PARAM_CONFIG_FILES="Makefile"
diff --git a/orte/mca/ns/replica/src/ns_replica.c b/orte/mca/ns/replica/ns_replica.c
similarity index 67%
rename from orte/mca/ns/replica/src/ns_replica.c
rename to orte/mca/ns/replica/ns_replica.c
index a5845c58a8..a1151af690 100644
--- a/orte/mca/ns/replica/src/ns_replica.c
+++ b/orte/mca/ns/replica/ns_replica.c
@@ -38,227 +38,6 @@
*/
#define NS_REPLICA_MAX_STRING_SIZE 256
-/*
- * functions
- */
-
-int orte_ns_replica_create_cellid(orte_cellid_t *cellid, char *site, char *resource)
-{
- orte_ns_replica_cell_tracker_t *new_cell;
- int rc;
- orte_std_cntr_t index;
-
- OPAL_TRACE(1);
-
- OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
-
- *cellid = ORTE_CELLID_MAX;
-
- /* check if cellid is available. NOTE: need to reserve
- * ORTE_CELLID_MAX as an invalid value, so can't allow
- * num_cells to get there
- */
- if (ORTE_CELLID_MAX-2 < orte_ns_replica.num_cells) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- new_cell = OBJ_NEW(orte_ns_replica_cell_tracker_t);
- if (NULL == new_cell) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
- if (ORTE_SUCCESS != (rc = orte_pointer_array_add(&index,
- orte_ns_replica.cells, new_cell))) {
- ORTE_ERROR_LOG(rc);
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return rc;
- }
- new_cell->site = strdup(site);
- new_cell->resource = strdup(resource);
-
- new_cell->cell = orte_ns_replica.num_cells;
- *cellid = new_cell->cell;
- (orte_ns_replica.num_cells)++;
-
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_SUCCESS;
-}
-
-int orte_ns_replica_get_cell_info(orte_cellid_t cellid,
- char **site, char **resource)
-{
- orte_std_cntr_t i;
- orte_cellid_t j;
- orte_ns_replica_cell_tracker_t **cell;
-
- OPAL_TRACE(1);
-
- OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
-
- cell = (orte_ns_replica_cell_tracker_t**)(orte_ns_replica.cells)->addr;
- for (i=0, j=0; j < orte_ns_replica.num_cells &&
- i < (orte_ns_replica.cells)->size; i++) {
- if (NULL != cell[i]) {
- j++;
- if (cellid == cell[i]->cell) {
- *site = strdup(cell[i]->site);
- *resource = strdup(cell[i]->resource);
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_SUCCESS;
- }
- }
- }
-
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_ERR_NOT_FOUND;
-}
-
-
-/*
- * JOBID functions
- */
-int orte_ns_replica_create_jobid(orte_jobid_t *jobid)
-{
- orte_ns_replica_jobid_tracker_t *ptr;
- int rc;
- orte_std_cntr_t index;
-
- OPAL_TRACE(1);
-
- OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
-
- *jobid = ORTE_JOBID_MAX;
- /* check if a jobid is available. NOTE: need to
- * reserve ORTE_JOBID_MAX as an invalid value, so can't let
- * num_jobids get there
- */
- if (ORTE_JOBID_MAX-2 < orte_ns_replica.num_jobids) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
-
- ptr = OBJ_NEW(orte_ns_replica_jobid_tracker_t);
- if (NULL == ptr) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
- if (ORTE_SUCCESS != (rc = orte_pointer_array_add(&index,
- orte_ns_replica.jobids, ptr))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(ptr);
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return rc;
- }
-
- ptr->jobid = orte_ns_replica.num_jobids;
- *jobid = ptr->jobid;
- (orte_ns_replica.num_jobids)++;
-
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_SUCCESS;
-}
-
-
-
-int orte_ns_replica_reserve_range(orte_jobid_t job, orte_vpid_t range,
- orte_vpid_t *start)
-{
- orte_ns_replica_jobid_tracker_t **ptr;
- orte_std_cntr_t j;
- orte_jobid_t k;
-
- OPAL_TRACE(1);
-
- OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
-
- /* find the jobid */
- ptr = (orte_ns_replica_jobid_tracker_t**)(orte_ns_replica.jobids)->addr;
- for (j=0, k=0; k < orte_ns_replica.num_jobids &&
- j < (orte_ns_replica.jobids)->size; j++) {
- if (NULL != ptr[j]) {
- k++;
- if (job == ptr[j]->jobid) {
- goto PROCESS;
- }
- }
- }
- /* didn't find the specified jobid - error */
- ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_ERR_NOT_FOUND;
-
-PROCESS:
- if ((ORTE_VPID_MAX-range-(ptr[j]->next_vpid)) > 0) {
- *start = ptr[j]->next_vpid;
- ptr[j]->next_vpid += range;
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_SUCCESS;
- }
-
- /* get here if the range isn't available */
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
-}
-
-int orte_ns_replica_get_job_peers(orte_process_name_t **procs,
- orte_std_cntr_t *num_procs, orte_jobid_t job)
-{
- orte_ns_replica_jobid_tracker_t **ptr;
- orte_process_name_t *nptr;
- orte_std_cntr_t j;
- orte_jobid_t k;
-
- OPAL_TRACE_ARG1(1, job);
-
- OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
-
- /* find the jobid */
- ptr = (orte_ns_replica_jobid_tracker_t**)(orte_ns_replica.jobids)->addr;
- for (j=0, k=0; k < orte_ns_replica.num_jobids &&
- j < (orte_ns_replica.jobids)->size; j++) {
- if (NULL != ptr[j]) {
- k++;
- if (job == ptr[j]->jobid) {
- goto PROCESS;
- }
- }
- }
- /* didn't find the specified jobid - error */
- ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_ERR_NOT_FOUND;
-
-PROCESS:
- /* the field next_vpid contains the value of the next unassigned
- * vpid, so the job extends from vpid=0 to that value. create
- * an array of process names containing those values
- */
- *procs = (orte_process_name_t*)malloc(ptr[j]->next_vpid * sizeof(orte_process_name_t));
- if (NULL == *procs) {
- ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_ERR_OUT_OF_RESOURCE;
- }
- nptr = *procs;
- for (k=0; k < ptr[j]->next_vpid; k++) {
- nptr->cellid = 0;
- nptr->jobid = job;
- nptr->vpid = (orte_vpid_t)k;
- nptr++;
- }
- *num_procs = (orte_std_cntr_t)ptr[j]->next_vpid;
-
- OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
- return ORTE_SUCCESS;
-}
-
-
/*
* DIAGNOSTIC functions
*/
diff --git a/orte/mca/ns/replica/src/ns_replica.h b/orte/mca/ns/replica/ns_replica.h
similarity index 50%
rename from orte/mca/ns/replica/src/ns_replica.h
rename to orte/mca/ns/replica/ns_replica.h
index 212ca9f618..a8c02d6be0 100644
--- a/orte/mca/ns/replica/src/ns_replica.h
+++ b/orte/mca/ns/replica/ns_replica.h
@@ -34,31 +34,54 @@
extern "C" {
#endif
-/* list class for tracking cellid's
+/*
+ * globals
*/
+#define NS_REPLICA_MAX_STRING_SIZE 256
+
+
+/* class for tracking cellid's */
struct orte_ns_replica_cell_tracker_t {
opal_object_t super;
orte_cellid_t cell;
char *site;
char *resource;
+ orte_nodeid_t next_nodeid;
+ orte_pointer_array_t *nodeids;
};
typedef struct orte_ns_replica_cell_tracker_t orte_ns_replica_cell_tracker_t;
OBJ_CLASS_DECLARATION(orte_ns_replica_cell_tracker_t);
-/*
- * object for tracking vpids/jobids
- * This structure is used to track jobid-max vpid pairs. Basically, we
- * are tracking the max used vpid for each jobid that has been created.
- */
-struct orte_ns_replica_jobid_tracker_t {
+/* object for tracking nodeid's */
+struct orte_ns_replica_nodeid_tracker_t {
opal_object_t super;
- orte_jobid_t jobid; /**< Job id */
- orte_vpid_t next_vpid;
+ char *nodename;
+ orte_nodeid_t nodeid;
};
-typedef struct orte_ns_replica_jobid_tracker_t orte_ns_replica_jobid_tracker_t;
+typedef struct orte_ns_replica_nodeid_tracker_t orte_ns_replica_nodeid_tracker_t;
+
+OBJ_CLASS_DECLARATION(orte_ns_replica_nodeid_tracker_t);
+
+
+/*
+ * object for tracking vpids and jobids for job families
+ * This structure is used to track the parent-child relationship between
+ * jobs. The "root" of the family is the initial parent - each child has
+ * a record under that parent. Any child that subsequently spawns its own
+ * children will form a list of jobids beneath them.
+ *
+ * each object records the jobid of the job it represents, and the next vpid
+ * that will be assigned when a range is requested.
+ */
+typedef struct {
+ opal_list_item_t super;
+ orte_jobid_t jobid;
+ orte_vpid_t next_vpid;
+ opal_list_t children;
+} orte_ns_replica_jobitem_t;
+OBJ_CLASS_DECLARATION(orte_ns_replica_jobitem_t);
-OBJ_CLASS_DECLARATION(orte_ns_replica_jobid_tracker_t);
struct orte_ns_replica_tagitem_t {
opal_object_t super;
@@ -85,12 +108,8 @@ typedef struct {
size_t max_size, block_size;
orte_cellid_t num_cells;
orte_pointer_array_t *cells;
-#if 0
- orte_jobgrp_t num_jobgrps;
- orte_pointer_array_t *jobgrps;
-#endif
orte_jobid_t num_jobids;
- orte_pointer_array_t *jobids;
+ opal_list_t jobs;
orte_pointer_array_t *tags;
orte_rml_tag_t num_tags;
orte_pointer_array_t *dts;
@@ -124,38 +143,53 @@ void orte_ns_replica_recv(int status, orte_process_name_t* sender,
orte_buffer_t* buffer, orte_rml_tag_t tag, void* cbdata);
/*
- * Implementation of create_cellid().
+ * CELL FUNCTIONS
*/
int orte_ns_replica_create_cellid(orte_cellid_t *cellid, char *site, char *resource);
-/*
- * Implementation of get_cell_info()
- */
int orte_ns_replica_get_cell_info(orte_cellid_t cellid,
char **site, char **resource);
-/*
- * Implementation of create_jobid().
- */
-int orte_ns_replica_create_jobid(orte_jobid_t *jobid);
+int orte_ns_replica_create_nodeids(orte_nodeid_t **nodeids, orte_std_cntr_t *nnodes,
+ orte_cellid_t cellid, char **nodenames);
+int orte_ns_replica_get_node_info(char ***nodenames, orte_cellid_t cellid, orte_std_cntr_t num_nodes, orte_nodeid_t *nodeids);
/*
- * Implementation of reserve_range()
+ * JOB FUNCTIONS
*/
+int orte_ns_replica_create_jobid(orte_jobid_t *jobid, opal_list_t *attrs);
+
+int orte_ns_replica_get_job_descendants(orte_jobid_t **descendants, orte_std_cntr_t *num_desc, orte_jobid_t job);
+
+int orte_ns_replica_get_job_children(orte_jobid_t **descendants, orte_std_cntr_t *num_desc, orte_jobid_t job);
+
+int orte_ns_replica_get_root_job(orte_jobid_t *root_job, orte_jobid_t job);
+
+int orte_ns_replica_get_parent_job(orte_jobid_t *parent, orte_jobid_t job);
+
int orte_ns_replica_reserve_range(orte_jobid_t job,
orte_vpid_t range,
orte_vpid_t *startvpid);
/*
- * Peer functions
+ * GENERAL FUNCTIONS
*/
-int orte_ns_replica_get_job_peers(orte_process_name_t **procs,
- orte_std_cntr_t *num_procs, orte_jobid_t job);
+int orte_ns_replica_get_peers(orte_process_name_t **procs,
+ orte_std_cntr_t *num_procs, opal_list_t *attrs);
+
+int orte_ns_replica_assign_rml_tag(orte_rml_tag_t *tag,
+ char *name);
+
+
+int orte_ns_replica_define_data_type(const char *name,
+ orte_data_type_t *type);
+
+int orte_ns_replica_create_my_name(void);
/*
- * Diagnostic functions
+ * DIAGNOSTIC FUNCTIONS
*/
int orte_ns_replica_dump_cells(void);
int orte_ns_replica_dump_cells_fn(orte_buffer_t *buffer);
@@ -171,20 +205,46 @@ int orte_ns_replica_dump_datatypes_fn(orte_buffer_t *buffer);
/*
- * Implementation of assign rml tag
+ * INTERNAL SUPPORT FUNCTIONS
*/
-int orte_ns_replica_assign_rml_tag(orte_rml_tag_t *tag,
- char *name);
-
-int orte_ns_replica_define_data_type(const char *name,
- orte_data_type_t *type);
-
-int orte_ns_replica_create_my_name(void);
-
-/*
- *
+/* find a job's record, wherever it may be located on the list of job families.
+ * this function searches the entire list of job families, traversing the list
+ * of all jobs in each family, until it finds the specified job. It then returns
+ * a pointer to the that job's info structure. It returns
+ * NULL (without error_logging an error) if no record is found
*/
+orte_ns_replica_jobitem_t* orte_ns_replica_find_job(orte_jobid_t job);
+
+/* find the root job for the specified job.
+ * this function searches the entire list of job families, traversing the list
+ * of all jobs in each family, until it finds the specified job. It then returns
+ * a pointer to the root job's info structure for that job family. It returns
+ * NULL (without error_logging an error) if no record is found
+ */
+orte_ns_replica_jobitem_t* orte_ns_replica_find_root_job(orte_jobid_t job);
+
+/* find a job's record on a specified root's family tree.
+ * this function finds the family record for the specified root job. It then
+ * traverses the children of that root until it finds the specified job, and then
+ * returns a pointer to that job's info structure. If root=jobid, then it will
+ * return a pointer to the root job's info structure. It returns
+ * NULL (without error_logging an error) if no record is found
+ */
+orte_ns_replica_jobitem_t* orte_ns_replica_search_job_family_tree(orte_jobid_t root, orte_jobid_t jobid);
+
+/* given a job's record, create a flattened list of descendants below it */
+void orte_ns_replica_construct_flattened_tree(opal_list_t *tree, orte_ns_replica_jobitem_t *ptr);
+
+/* search down a tree, following all the children's branches, to find the specified
+ * job. Return a pointer to that object, and a pointer to the parent object
+ * This function is called recursively, so it passes into it the ptr to the
+ * current object being looked at
+ */
+orte_ns_replica_jobitem_t *down_search(orte_ns_replica_jobitem_t *ptr,
+ orte_ns_replica_jobitem_t **parent_ptr,
+ orte_jobid_t job);
+
ORTE_MODULE_DECLSPEC extern mca_ns_base_component_t mca_ns_replica_component;
#if defined(c_plusplus) || defined(__cplusplus)
diff --git a/orte/mca/ns/replica/ns_replica_cell_fns.c b/orte/mca/ns/replica/ns_replica_cell_fns.c
new file mode 100644
index 0000000000..86758af1ee
--- /dev/null
+++ b/orte/mca/ns/replica/ns_replica_cell_fns.c
@@ -0,0 +1,297 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+#include "orte_config.h"
+
+#include <stdio.h>
+#include <string.h>
+
+#include "opal/threads/mutex.h"
+#include "opal/util/argv.h"
+#include "opal/util/output.h"
+#include "opal/util/trace.h"
+
+#include "orte/dss/dss.h"
+#include "orte/mca/errmgr/errmgr.h"
+
+#include "orte/mca/ns/base/base.h"
+#include "ns_replica.h"
+
+/*
+ * functions
+ */
+
+int orte_ns_replica_create_cellid(orte_cellid_t *cellid, char *site, char *resource)
+{
+ orte_ns_replica_cell_tracker_t *new_cell, **cell;
+ int rc;
+ orte_std_cntr_t i, j, index;
+
+ OPAL_TRACE(1);
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ *cellid = ORTE_CELLID_INVALID;
+
+ /* check for error */
+ if (NULL == site || NULL == resource) {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ /* is this a known cellid? */
+ cell = (orte_ns_replica_cell_tracker_t**)(orte_ns_replica.cells)->addr;
+ for (i=0, j=0; j < orte_ns_replica.num_cells &&
+ i < (orte_ns_replica.cells)->size; i++) {
+ if (NULL != cell[i]) {
+ j++;
+ if (0 == strcmp(site, cell[i]->site) &&
+ 0 == strcmp(resource, cell[i]->resource)) {
+ *cellid = cell[i]->cell;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+ }
+ }
+ }
+
+ /* new cell - check if cellid is available */
+ if (ORTE_CELLID_MAX-1 < orte_ns_replica.num_cells) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ new_cell = OBJ_NEW(orte_ns_replica_cell_tracker_t);
+ if (NULL == new_cell) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ if (ORTE_SUCCESS != (rc = orte_pointer_array_add(&index,
+ orte_ns_replica.cells, new_cell))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+ new_cell->site = strdup(site);
+ new_cell->resource = strdup(resource);
+
+ new_cell->cell = orte_ns_replica.num_cells;
+ *cellid = new_cell->cell;
+ (orte_ns_replica.num_cells)++;
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_replica_get_cell_info(orte_cellid_t cellid,
+ char **site, char **resource)
+{
+ orte_std_cntr_t i;
+ orte_cellid_t j;
+ orte_ns_replica_cell_tracker_t **cell;
+
+ OPAL_TRACE(1);
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ cell = (orte_ns_replica_cell_tracker_t**)(orte_ns_replica.cells)->addr;
+ for (i=0, j=0; j < orte_ns_replica.num_cells &&
+ i < (orte_ns_replica.cells)->size; i++) {
+ if (NULL != cell[i]) {
+ j++;
+ if (cellid == cell[i]->cell) {
+ *site = strdup(cell[i]->site);
+ *resource = strdup(cell[i]->resource);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+ }
+ }
+ }
+
+ /* it isn't an error to not find the cell - so do NOT
+ * report it via ORTE_ERROR_LOG
+ */
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_NOT_FOUND;
+}
+
+/*
+ * NODEID
+ */
+int orte_ns_replica_create_nodeids(orte_nodeid_t **nodeids, orte_std_cntr_t *nnodes,
+ orte_cellid_t cellid, char **nodenames)
+{
+ orte_ns_replica_cell_tracker_t **cell, *cptr;
+ orte_ns_replica_nodeid_tracker_t **nodes, *node;
+ orte_nodeid_t *nds, nid;
+ orte_std_cntr_t i, j, k, m, n, num_nodes;
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ num_nodes = opal_argv_count(nodenames);
+ if (0 == num_nodes) { /** no nodenames provided - just return */
+ *nodeids = NULL;
+ *nnodes = 0;
+ return ORTE_SUCCESS;
+ }
+
+ nds = (orte_nodeid_t*)malloc(num_nodes * sizeof(orte_nodeid_t));
+ if (NULL == nds) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ /** find the cell */
+ cell = (orte_ns_replica_cell_tracker_t**)(orte_ns_replica.cells)->addr;
+ for (i=0, j=0; j < orte_ns_replica.num_cells &&
+ i < (orte_ns_replica.cells)->size; i++) {
+ if (NULL != cell[i]) {
+ j++;
+ if (cellid == cell[i]->cell) {
+ /** found the specified cell - check to see if nodename has already been
+ * defined. if so, just return the nodeid. if not, create a new one
+ */
+ cptr = cell[i];
+ goto PROCESS;
+ }
+ }
+ }
+ /** get here if we didn't find the cell */
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+ free(nds);
+ *nodeids = NULL;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_NOT_FOUND;
+
+PROCESS:
+ nodes = (orte_ns_replica_nodeid_tracker_t**)(cptr->nodeids->addr);
+ for (n=0; n < num_nodes; n++) {
+ for (k=0, m=0; m < cptr->next_nodeid &&
+ k < (cptr->nodeids)->size; k++) {
+ if (NULL != nodes[k]) {
+ m++;
+ if (strcmp(nodenames[n], nodes[k]->nodename) == 0) { /** found same name */
+ nid = nodes[k]->nodeid;
+ goto ASSIGN;
+ }
+ }
+ }
+ /** get here if we don't find this nodename - add one */
+ node = OBJ_NEW(orte_ns_replica_nodeid_tracker_t);
+ if (NULL == node) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ free(nds);
+ *nodeids = NULL;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ node->nodename = strdup(nodenames[n]);
+ node->nodeid = cptr->next_nodeid;
+ cptr->next_nodeid++;
+ nid = node->nodeid;
+
+ASSIGN:
+ nds[n] = nid;
+ } /** for n */
+
+ *nodeids = nds;
+ *nnodes = num_nodes;
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_replica_get_node_info(char ***nodenames, orte_cellid_t cellid,
+ orte_std_cntr_t num_nodes, orte_nodeid_t *nodeids)
+{
+ char **names, *nm;
+ orte_ns_replica_cell_tracker_t **cell, *cptr;
+ orte_ns_replica_nodeid_tracker_t **nodes;
+ orte_std_cntr_t i, j, k, m, n;
+ char *err_name = "NODE_NOT_FOUND"
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ if (0 == num_nodes) {
+ *nodenames = NULL;
+ return ORTE_SUCCESS;
+ }
+
+ /** allocate an extra space for the NULL termination */
+ names = (char**)malloc((num_nodes+1) * sizeof(char*));
+ if (NULL == names) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ names[num_nodes] = NULL; /** NULL-terminate the list */
+
+ /** find the cell */
+ cell = (orte_ns_replica_cell_tracker_t**)(orte_ns_replica.cells)->addr;
+ for (i=0, j=0; j < orte_ns_replica.num_cells &&
+ i < (orte_ns_replica.cells)->size; i++) {
+ if (NULL != cell[i]) {
+ j++;
+ if (cellid == cell[i]->cell) {
+ /** found the specified cell - check to see if nodename has already been
+ * defined. if so, just return the nodeid. if not, create a new one
+ */
+ cptr = cell[i];
+ goto PROCESS;
+ }
+ }
+ }
+ /** get here if we didn't find the cell */
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+ free(names);
+ *nodenames = NULL;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_NOT_FOUND;
+
+PROCESS:
+ nodes = (orte_ns_replica_nodeid_tracker_t**)(cell[i]->nodeids->addr);
+ for (n=0; n < num_nodes; n++) {
+ for (k=0, m=0; m < cell[i]->next_nodeid &&
+ k < (cell[i]->nodeids)->size; k++) {
+ if (NULL != nodes[k]) {
+ m++;
+ if (nodeids[n] == nodes[k]->nodeid) { /** found it */
+ nm = nodes[k]->nodename;
+ goto ASSIGN;
+ }
+ }
+ }
+ /** node not found - set name to error name. Can't set it to NULL since
+ * the list is a NULL-terminated one
+ */
+ nm = err_name;
+
+ASSIGN:
+ names[n] = strdup(nm);
+ }
+
+ *nodenames = names;
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
diff --git a/orte/mca/ns/replica/ns_replica_class_instances.h b/orte/mca/ns/replica/ns_replica_class_instances.h
new file mode 100644
index 0000000000..b2a5c49a5c
--- /dev/null
+++ b/orte/mca/ns/replica/ns_replica_class_instances.h
@@ -0,0 +1,172 @@
+/* -*- C -*-
+*
+* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+* University Research and Technology
+* Corporation. All rights reserved.
+* Copyright (c) 2004-2006 The University of Tennessee and The University
+* of Tennessee Research Foundation. All rights
+* reserved.
+* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+* University of Stuttgart. All rights reserved.
+* Copyright (c) 2004-2005 The Regents of the University of California.
+* All rights reserved.
+* $COPYRIGHT$
+*
+* Additional copyrights may follow
+*
+* $HEADER$
+*
+*/
+#ifndef NS_REPLICA_CLASS_INSTANCES_H
+#define NS_REPLICA_CLASS_INSTANCES_H
+
+#include "orte_config.h"
+#include "orte/orte_types.h"
+#include "orte/orte_constants.h"
+#include "opal/threads/mutex.h"
+#include "opal/class/opal_object.h"
+#include "orte/class/orte_pointer_array.h"
+#include "orte/dss/dss.h"
+#include "orte/mca/oob/oob_types.h"
+#include "orte/mca/ns/base/base.h"
+
+#if defined(c_plusplus) || defined(__cplusplus)
+extern "C" {
+#endif
+
+/*** CELLID ***/
+/* constructor - used to initialize state of cell_tracker instance */
+static void orte_ns_replica_cell_tracker_construct(orte_ns_replica_cell_tracker_t* cell_tracker)
+{
+ cell_tracker->cell = ORTE_CELLID_INVALID;
+ cell_tracker->site = NULL;
+ cell_tracker->resource = NULL;
+
+ cell_tracker->next_nodeid = 0;
+ orte_pointer_array_init(&(cell_tracker->nodeids),
+ orte_ns_replica.block_size,
+ orte_ns_replica.max_size,
+ orte_ns_replica.block_size);
+}
+
+/* destructor - used to free any resources held by instance */
+static void orte_ns_replica_cell_tracker_destructor(orte_ns_replica_cell_tracker_t* cell_tracker)
+{
+ orte_std_cntr_t i, j;
+ orte_ns_replica_nodeid_tracker_t **nodeid;
+
+ if (NULL != cell_tracker->site) free(cell_tracker->site);
+ if (NULL != cell_tracker->resource) free(cell_tracker->resource);
+
+ nodeid = (orte_ns_replica_nodeid_tracker_t**)(cell_tracker->nodeids)->addr;
+
+ for (i=0, j=0; j < cell_tracker->next_nodeid &&
+ i < (cell_tracker->nodeids)->size; i++) {
+ if (NULL != nodeid[i]) {
+ j++;
+ OBJ_RELEASE(nodeid[i]);
+ }
+ }
+ OBJ_RELEASE(cell_tracker->nodeids);
+}
+
+/* define instance of opal_class_t */
+OBJ_CLASS_INSTANCE(orte_ns_replica_cell_tracker_t, /* type name */
+ opal_object_t, /* parent "class" name */
+ orte_ns_replica_cell_tracker_construct, /* constructor */
+ orte_ns_replica_cell_tracker_destructor); /* destructor */
+
+
+/** NODEID */
+static void orte_ns_replica_nodeid_tracker_construct(orte_ns_replica_nodeid_tracker_t *ptr)
+{
+ ptr->nodeid = ORTE_NODEID_INVALID;
+ ptr->nodename = NULL;
+}
+
+static void orte_ns_replica_nodeid_tracker_destructor(orte_ns_replica_nodeid_tracker_t *ptr)
+{
+ if (NULL != ptr->nodename) free(ptr->nodename);
+}
+
+OBJ_CLASS_INSTANCE(orte_ns_replica_nodeid_tracker_t, /* type name */
+ opal_object_t, /* parent "class" name */
+ orte_ns_replica_nodeid_tracker_construct, /* constructor */
+ orte_ns_replica_nodeid_tracker_destructor); /* destructor */
+
+
+/*** JOBITEM ***/
+/* constructor - used to initialize state of jobitem instance */
+static void orte_ns_replica_jobitem_construct(orte_ns_replica_jobitem_t *ptr)
+{
+ ptr->jobid = ORTE_JOBID_INVALID;
+ ptr->next_vpid = 0;
+ OBJ_CONSTRUCT(&ptr->children, opal_list_t);
+}
+
+/* destructor - used to free any resources held by instance */
+static void orte_ns_replica_jobitem_destructor(orte_ns_replica_jobitem_t *ptr){
+ opal_list_item_t *item;
+
+ while (NULL != (item = opal_list_remove_first(&ptr->children))) {
+ OBJ_RELEASE(item);
+ }
+ OBJ_DESTRUCT(&ptr->children);
+}
+
+/* define instance of opal_class_t */
+OBJ_CLASS_INSTANCE(orte_ns_replica_jobitem_t, /* type name */
+ opal_list_item_t, /* parent "class" name */
+ orte_ns_replica_jobitem_construct, /* constructor */
+ orte_ns_replica_jobitem_destructor); /* destructor */
+
+
+/*** RML TAG ***/
+/* constructor - used to initialize state of taglist instance */
+static void orte_ns_replica_tagitem_construct(orte_ns_replica_tagitem_t* tagitem)
+{
+ tagitem->tag = ORTE_RML_TAG_MAX;
+ tagitem->name = NULL;
+}
+
+/* destructor - used to free any resources held by instance */
+static void orte_ns_replica_tagitem_destructor(orte_ns_replica_tagitem_t* tagitem)
+{
+ if (NULL != tagitem->name) {
+ free(tagitem->name);
+ }
+}
+
+/* define instance of opal_class_t */
+OBJ_CLASS_INSTANCE(orte_ns_replica_tagitem_t, /* type name */
+ opal_object_t, /* parent "class" name */
+ orte_ns_replica_tagitem_construct, /* constructor */
+ orte_ns_replica_tagitem_destructor); /* destructor */
+
+
+/*** DATA TYPE ***/
+/* constructor - used to initialize state of dtilist instance */
+static void orte_ns_replica_dti_construct(orte_ns_replica_dti_t* dti)
+{
+ dti->id = ORTE_DSS_ID_MAX;
+ dti->name = NULL;
+}
+
+/* destructor - used to free any resources held by instance */
+static void orte_ns_replica_dti_destructor(orte_ns_replica_dti_t* dti)
+{
+ if (NULL != dti->name) {
+ free(dti->name);
+ }
+}
+
+/* define instance of opal_class_t */
+OBJ_CLASS_INSTANCE(orte_ns_replica_dti_t, /* type name */
+ opal_object_t, /* parent "class" name */
+ orte_ns_replica_dti_construct, /* constructor */
+ orte_ns_replica_dti_destructor); /* destructor */
+
+#if defined(c_plusplus) || defined(__cplusplus)
+}
+#endif
+#endif
diff --git a/orte/mca/ns/replica/ns_replica_component.c b/orte/mca/ns/replica/ns_replica_component.c
new file mode 100644
index 0000000000..eb2359b1e4
--- /dev/null
+++ b/orte/mca/ns/replica/ns_replica_component.c
@@ -0,0 +1,309 @@
+/* -*- C -*-
+*
+* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+* University Research and Technology
+* Corporation. All rights reserved.
+* Copyright (c) 2004-2005 The University of Tennessee and The University
+* of Tennessee Research Foundation. All rights
+* reserved.
+* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+* University of Stuttgart. All rights reserved.
+* Copyright (c) 2004-2005 The Regents of the University of California.
+* All rights reserved.
+* $COPYRIGHT$
+*
+* Additional copyrights may follow
+*
+* $HEADER$
+*/
+/** @file:
+*
+* The Open MPI Name Server
+*
+* The Open MPI Name Server provides unique name ranges for processes
+* within the universe. Each universe will have one name server
+* running within the seed daemon. This is done to prevent the
+* inadvertent duplication of names.
+*/
+
+/*
+ * includes
+ */
+#include "orte_config.h"
+
+#include "orte/orte_constants.h"
+#include "orte/orte_types.h"
+
+#include "opal/threads/mutex.h"
+#include "opal/class/opal_list.h"
+#include "opal/util/output.h"
+
+#include "opal/mca/mca.h"
+#include "opal/mca/base/mca_base_param.h"
+
+#include "orte/util/proc_info.h"
+#include "orte/mca/errmgr/errmgr.h"
+#include "orte/mca/rml/rml.h"
+
+#include "orte/mca/ns/base/ns_private.h"
+#include "ns_replica.h"
+
+
+/*
+ * Struct of function pointers that need to be initialized
+ */
+mca_ns_base_component_t mca_ns_replica_component = {
+{
+ MCA_NS_BASE_VERSION_2_0_0,
+
+ "replica", /* MCA module name */
+ ORTE_MAJOR_VERSION, /* MCA module major version */
+ ORTE_MINOR_VERSION, /* MCA module minor version */
+ ORTE_RELEASE_VERSION, /* MCA module release version */
+ orte_ns_replica_open, /* module open */
+ orte_ns_replica_close /* module close */
+},
+{
+ false /* checkpoint / restart */
+},
+orte_ns_replica_init, /* module init */
+orte_ns_replica_finalize /* module shutdown */
+};
+
+/*
+ * setup the function pointers for the module
+ */
+static mca_ns_base_module_t orte_ns_replica_module = {
+ /* init */
+ orte_ns_replica_module_init,
+ /* cell functions */
+ orte_ns_replica_create_cellid,
+ orte_ns_replica_get_cell_info,
+ orte_ns_base_get_cellid_string,
+ orte_ns_base_convert_cellid_to_string,
+ orte_ns_base_convert_string_to_cellid,
+ /** node functions */
+ orte_ns_replica_create_nodeids,
+ orte_ns_replica_get_node_info,
+ orte_ns_base_convert_nodeid_to_string,
+ orte_ns_base_convert_string_to_nodeid,
+ /* jobid functions */
+ orte_ns_replica_create_jobid,
+ orte_ns_replica_get_job_descendants,
+ orte_ns_replica_get_job_children,
+ orte_ns_replica_get_root_job,
+ orte_ns_replica_get_parent_job,
+ orte_ns_base_get_jobid_string,
+ orte_ns_base_convert_jobid_to_string,
+ orte_ns_base_convert_string_to_jobid,
+ orte_ns_replica_reserve_range,
+ /* vpid functions */
+ orte_ns_base_get_vpid_string,
+ orte_ns_base_convert_vpid_to_string,
+ orte_ns_base_convert_string_to_vpid,
+ /* name functions */
+ orte_ns_base_create_process_name,
+ orte_ns_replica_create_my_name,
+ orte_ns_base_convert_string_to_process_name,
+ orte_ns_base_get_proc_name_string,
+ orte_ns_base_compare_fields,
+ /* peer functions */
+ orte_ns_replica_get_peers,
+ /* tag server functions */
+ orte_ns_replica_assign_rml_tag,
+ /* data type functions */
+ orte_ns_replica_define_data_type,
+ /* diagnostic functions */
+ orte_ns_replica_dump_cells,
+ orte_ns_replica_dump_jobs,
+ orte_ns_replica_dump_tags,
+ orte_ns_replica_dump_datatypes
+};
+
+/*
+ * Whether or not we allowed this component to be selected
+ */
+static bool initialized = false;
+
+/*
+ * class instantiations
+ */
+#include "ns_replica_class_instances.h"
+
+/*
+ * globals needed within replica component
+ */
+orte_ns_replica_globals_t orte_ns_replica;
+
+/*
+ * don't really need this function - could just put NULL in the above structure
+ * Just holding the place in case we decide there is something we need to do
+ */
+int orte_ns_replica_open(void)
+{
+ int id, param;
+
+ id = mca_base_param_register_int("ns", "replica", "debug", NULL, (int)false);
+ mca_base_param_lookup_int(id, &orte_ns_replica.debug);
+
+ id = mca_base_param_register_int("ns", "replica", "isolate", NULL, (int)false);
+ mca_base_param_lookup_int(id, ¶m);
+ if (param) {
+ orte_ns_replica.isolate = true;
+ } else {
+ orte_ns_replica.isolate = false;
+ }
+
+ id = mca_base_param_register_int("ns", "replica", "maxsize", NULL,
+ ORTE_NS_ARRAY_MAX_SIZE);
+ mca_base_param_lookup_int(id, ¶m);
+ orte_ns_replica.max_size = (size_t)param;
+
+ id = mca_base_param_register_int("ns", "replica", "blocksize", NULL,
+ ORTE_NS_ARRAY_BLOCK_SIZE);
+ mca_base_param_lookup_int(id, ¶m);
+ orte_ns_replica.block_size = (size_t)param;
+
+ return ORTE_SUCCESS;
+}
+
+/*
+ * ditto for this one
+ */
+int orte_ns_replica_close(void)
+{
+ return ORTE_SUCCESS;
+}
+
+mca_ns_base_module_t* orte_ns_replica_init(int *priority)
+{
+ int rc;
+
+ /* If we are to host a replica, then we want to be selected, so do all the
+ setup and return the module */
+
+ if (NULL == orte_process_info.ns_replica_uri) {
+
+ /* Return a module (choose an arbitrary, positive priority --
+ it's only relevant compared to other ns components). If
+ we're not the seed, then we don't want to be selected, so
+ return NULL. */
+
+ *priority = 50;
+
+ /* initialize the cell info tracker */
+ if (ORTE_SUCCESS != (rc = orte_pointer_array_init(&(orte_ns_replica.cells),
+ (orte_std_cntr_t)orte_ns_replica.block_size,
+ (orte_std_cntr_t)orte_ns_replica.max_size,
+ (orte_std_cntr_t)orte_ns_replica.block_size))) {
+ ORTE_ERROR_LOG(rc);
+ return NULL;
+ }
+ orte_ns_replica.num_cells = 0;
+
+ /* initialize the job tracking system */
+ OBJ_CONSTRUCT(&orte_ns_replica.jobs, opal_list_t);
+ orte_ns_replica.num_jobids = 0;
+
+ /* initialize the taglist */
+
+ if (ORTE_SUCCESS != (rc = orte_pointer_array_init(&(orte_ns_replica.tags),
+ (orte_std_cntr_t)orte_ns_replica.block_size,
+ (orte_std_cntr_t)orte_ns_replica.max_size,
+ (orte_std_cntr_t)orte_ns_replica.block_size))) {
+ ORTE_ERROR_LOG(rc);
+ return NULL;
+ }
+ orte_ns_replica.num_tags = 0;
+
+ /* initialize the dtlist */
+
+ if (ORTE_SUCCESS != (rc = orte_pointer_array_init(&(orte_ns_replica.dts),
+ (orte_std_cntr_t)orte_ns_replica.block_size,
+ (orte_std_cntr_t)orte_ns_replica.max_size,
+ (orte_std_cntr_t)orte_ns_replica.block_size))) {
+ ORTE_ERROR_LOG(rc);
+ return NULL;
+ }
+ orte_ns_replica.num_dts = 0;
+
+ /* setup the thread lock */
+ OBJ_CONSTRUCT(&orte_ns_replica.mutex, opal_mutex_t);
+
+ /* Return the module */
+
+ initialized = true;
+ return &orte_ns_replica_module;
+ } else {
+ return NULL;
+ }
+}
+
+int orte_ns_replica_module_init(void)
+{
+ int rc;
+ if (orte_ns_replica.isolate) {
+ return ORTE_SUCCESS;
+ }
+
+ /* issue non-blocking receive for call_back function */
+ rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_NS, ORTE_RML_PERSISTENT, orte_ns_replica_recv, NULL);
+ if(rc < 0) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ return ORTE_SUCCESS;
+}
+
+
+/*
+ * finalize routine
+ */
+int orte_ns_replica_finalize(void)
+{
+ orte_ns_replica_cell_tracker_t **cptr;
+ opal_list_item_t *item;
+ orte_ns_replica_tagitem_t **tag;
+ orte_ns_replica_dti_t **dti;
+ orte_std_cntr_t i;
+
+ /* free all tracking storage, but only if this component was initialized */
+
+ if (initialized) {
+ cptr = (orte_ns_replica_cell_tracker_t**)(orte_ns_replica.cells)->addr;
+ for (i=0; i < (orte_ns_replica.cells)->size; i++) {
+ if (NULL != cptr[i]) {
+ OBJ_RELEASE(cptr[i]);
+ }
+ }
+ OBJ_RELEASE(orte_ns_replica.cells);
+
+ while (NULL != (item = opal_list_remove_first(&orte_ns_replica.jobs))) {
+ OBJ_RELEASE(item);
+ }
+ OBJ_DESTRUCT(&orte_ns_replica.jobs);
+
+ tag = (orte_ns_replica_tagitem_t**)(orte_ns_replica.tags)->addr;
+ for (i=0; i < (orte_ns_replica.tags)->size; i++) {
+ if (NULL != tag[i]) OBJ_RELEASE(tag[i]);
+ }
+ OBJ_RELEASE(orte_ns_replica.tags);
+
+ dti = (orte_ns_replica_dti_t**)(orte_ns_replica.dts)->addr;
+ for (i=0; i < (orte_ns_replica.dts)->size; i++) {
+ if (NULL != dti[i]) OBJ_RELEASE(dti[i]);
+ }
+ OBJ_RELEASE(orte_ns_replica.dts);
+
+ initialized = false;
+ }
+
+ /* All done */
+ if (orte_ns_replica.isolate) {
+ return ORTE_SUCCESS;
+ }
+
+ orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_NS);
+ return ORTE_SUCCESS;
+}
+
diff --git a/orte/mca/ns/replica/ns_replica_diag_fns.c b/orte/mca/ns/replica/ns_replica_diag_fns.c
new file mode 100644
index 0000000000..94bd9ff103
--- /dev/null
+++ b/orte/mca/ns/replica/ns_replica_diag_fns.c
@@ -0,0 +1,322 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+#include "orte_config.h"
+
+#include <stdio.h>
+#include <string.h>
+
+#include "opal/threads/mutex.h"
+#include "opal/util/output.h"
+#include "opal/util/trace.h"
+
+#include "orte/dss/dss.h"
+#include "orte/mca/errmgr/errmgr.h"
+
+#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/base/ns_private.h"
+#include "ns_replica.h"
+
+/*
+ * DIAGNOSTIC functions
+ */
+int orte_ns_replica_dump_cells(void)
+{
+ orte_buffer_t buffer;
+ int rc;
+
+ OBJ_CONSTRUCT(&buffer, orte_buffer_t);
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_cells_fn(&buffer))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&buffer))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&buffer);
+ return rc;
+ }
+
+ OBJ_DESTRUCT(&buffer);
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_replica_dump_cells_fn(orte_buffer_t *buffer)
+{
+ orte_std_cntr_t i;
+ orte_cellid_t j;
+ orte_ns_replica_cell_tracker_t **cell;
+ char tmp_out[NS_REPLICA_MAX_STRING_SIZE], *tmp;
+ int rc;
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ tmp = tmp_out;
+ snprintf(tmp, NS_REPLICA_MAX_STRING_SIZE, "Dump of Name Service Cell Tracker\n");
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(buffer, &tmp, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+ cell = (orte_ns_replica_cell_tracker_t**)(orte_ns_replica.cells)->addr;
+ for (i=0, j=0; j < orte_ns_replica.num_cells &&
+ i < (orte_ns_replica.cells)->size; i++) {
+ if (NULL != cell[i]) {
+ j++;
+ snprintf(tmp, NS_REPLICA_MAX_STRING_SIZE, "Num: %lu\tCell: %lu\n",
+ (unsigned long)j, (unsigned long)cell[i]->cell);
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(buffer, &tmp, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+ snprintf(tmp, NS_REPLICA_MAX_STRING_SIZE, "\tSite: %s\n\tResource: %s\n",
+ cell[i]->site, cell[i]->resource);
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(buffer, &tmp, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+ }
+ }
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_replica_dump_jobs(void)
+{
+ orte_buffer_t buffer;
+ int rc;
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ OBJ_CONSTRUCT(&buffer, orte_buffer_t);
+
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_jobs_fn(&buffer))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&buffer))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&buffer);
+ return rc;
+ }
+
+ OBJ_DESTRUCT(&buffer);
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
+static int dump_child_jobs(orte_ns_replica_jobitem_t *ptr, char *prefix, orte_buffer_t *buffer)
+{
+ opal_list_item_t *item;
+ orte_ns_replica_jobitem_t *child;
+ char *tmp;
+ int rc;
+ char *pfx;
+
+ asprintf(&pfx, "%s ", prefix);
+
+ /* print out the children's info */
+ for (item = opal_list_get_first(&ptr->children);
+ item != opal_list_get_end(&ptr->children);
+ item = opal_list_get_next(item)) {
+ child = (orte_ns_replica_jobitem_t*)item;
+ asprintf(&tmp, "%sChild jobid: %ld Next vpid: %ld Num direct children: %ld\n",
+ pfx, (long)child->jobid, (long)child->next_vpid, (long)opal_list_get_size(&child->children));
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(buffer, &tmp, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ free(tmp);
+ if (ORTE_SUCCESS != (rc = dump_child_jobs(child, pfx, buffer))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ }
+ free(pfx);
+
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_replica_dump_jobs_fn(orte_buffer_t *buffer)
+{
+ orte_ns_replica_jobitem_t *root;
+ opal_list_item_t *item;
+ char *tmp;
+ int rc;
+ char *prefix = " ";
+
+ asprintf(&tmp, "Dump of Name Service Jobid Tracker\n");
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(buffer, &tmp, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ free(tmp);
+
+ for (item = opal_list_get_first(&orte_ns_replica.jobs);
+ item != opal_list_get_end(&orte_ns_replica.jobs);
+ item = opal_list_get_next(item)) {
+ root = (orte_ns_replica_jobitem_t*)item;
+ asprintf(&tmp, " Data for job family with root %ld\n", (long)root->jobid);
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(buffer, &tmp, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ free(tmp);
+ asprintf(&tmp, "%sNext vpid: %ld Num direct children: %ld\n",
+ prefix, (long)root->next_vpid, (long)opal_list_get_size(&root->children));
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(buffer, &tmp, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ free(tmp);
+ if (ORTE_SUCCESS != (rc = dump_child_jobs(root, prefix, buffer))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ }
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_replica_dump_tags(void)
+{
+ orte_buffer_t buffer;
+ int rc;
+
+ OBJ_CONSTRUCT(&buffer, orte_buffer_t);
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_tags_fn(&buffer))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&buffer))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&buffer);
+ return rc;
+ }
+
+ OBJ_DESTRUCT(&buffer);
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_replica_dump_tags_fn(orte_buffer_t *buffer)
+{
+ orte_std_cntr_t i;
+ orte_rml_tag_t j;
+ orte_ns_replica_tagitem_t **ptr;
+ char tmp_out[NS_REPLICA_MAX_STRING_SIZE], *tmp;
+ int rc;
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ tmp = tmp_out;
+ snprintf(tmp, NS_REPLICA_MAX_STRING_SIZE, "Dump of Name Service RML Tag Tracker\n");
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(buffer, &tmp, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+ ptr = (orte_ns_replica_tagitem_t**)(orte_ns_replica.tags)->addr;
+ for (i=0, j=0; j < orte_ns_replica.num_tags &&
+ i < (orte_ns_replica.tags)->size; i++) {
+ if (NULL != ptr[i]) {
+ j++;
+ snprintf(tmp, NS_REPLICA_MAX_STRING_SIZE, "Num: %lu\tTag id: %lu\tName: %s\n",
+ (unsigned long)j, (unsigned long)ptr[i]->tag, ptr[i]->name);
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(buffer, &tmp, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+ }
+ }
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_replica_dump_datatypes(void)
+{
+ orte_buffer_t buffer;
+ int rc;
+
+ OBJ_CONSTRUCT(&buffer, orte_buffer_t);
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_datatypes_fn(&buffer))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_base_print_dump(&buffer))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&buffer);
+ return rc;
+ }
+
+ OBJ_DESTRUCT(&buffer);
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_replica_dump_datatypes_fn(orte_buffer_t *buffer)
+{
+ orte_std_cntr_t i, j;
+ orte_ns_replica_dti_t **ptr;
+ char tmp_out[NS_REPLICA_MAX_STRING_SIZE], *tmp;
+ int rc;
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ tmp = tmp_out;
+ snprintf(tmp, NS_REPLICA_MAX_STRING_SIZE, "Dump of Name Service Datatype Tracker\n");
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(buffer, &tmp, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+ ptr = (orte_ns_replica_dti_t**)(orte_ns_replica.dts)->addr;
+ for (i=0, j=0; j < orte_ns_replica.num_dts &&
+ i < (orte_ns_replica.dts)->size; i++) {
+ if (NULL != ptr[i]) {
+ j++;
+ snprintf(tmp, NS_REPLICA_MAX_STRING_SIZE, "Num: %lu\tDatatype id: %lu\tName: %s\n",
+ (unsigned long)j, (unsigned long)ptr[i]->id, ptr[i]->name);
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(buffer, &tmp, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+ }
+ }
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+
+ return ORTE_SUCCESS;
+}
+
diff --git a/orte/mca/ns/replica/ns_replica_general_fns.c b/orte/mca/ns/replica/ns_replica_general_fns.c
new file mode 100644
index 0000000000..b56f107e29
--- /dev/null
+++ b/orte/mca/ns/replica/ns_replica_general_fns.c
@@ -0,0 +1,403 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+#include "orte_config.h"
+
+#include <stdio.h>
+#include <string.h>
+
+#include "opal/threads/mutex.h"
+#include "opal/util/output.h"
+#include "opal/util/trace.h"
+
+#include "orte/dss/dss.h"
+#include "orte/mca/rmgr/rmgr.h"
+#include "orte/mca/errmgr/errmgr.h"
+
+#include "orte/mca/ns/base/ns_private.h"
+#include "ns_replica.h"
+
+
+/*** GET PEERS ***/
+int orte_ns_replica_get_peers(orte_process_name_t **procs,
+ orte_std_cntr_t *num_procs, opal_list_t *attrs)
+{
+ orte_std_cntr_t i, isave, npeers;
+ orte_jobid_t *jptr;
+ orte_cellid_t *cptr;
+ orte_attribute_t *attr;
+ orte_ns_replica_jobitem_t *job_info, *child;
+ opal_list_item_t *item;
+ opal_list_t peerlist;
+ int rc;
+
+ OPAL_TRACE(1);
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ /* set default value */
+ *procs = NULL;
+ *num_procs = 0;
+
+ /* check the attributes to see if USE_JOB or USE_CELL has been set. If not, then this is
+ * a request for my own job peers - process that one locally
+ */
+
+ /* if the cell is given AND it matches my own, then we can process this
+ * quickly. Otherwise, we have to do some more work.
+ *
+ * RHC: when we go multi-cell, we need a way to find all the cells upon
+ * which a job is executing so we can make this work!
+ */
+ if (NULL != (attr = orte_rmgr.find_attribute(attrs, ORTE_NS_USE_CELL))) {
+ if (ORTE_SUCCESS != (rc = orte_dss.get((void**)&cptr, attr->value, ORTE_CELLID))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+ if (*cptr != ORTE_PROC_MY_NAME->cellid && *cptr != ORTE_CELLID_WILDCARD) {
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_IMPLEMENTED);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_NOT_IMPLEMENTED;
+ }
+ }
+
+ if (NULL == (attr = orte_rmgr.find_attribute(attrs, ORTE_NS_USE_JOBID))) {
+ /* get my own job peers, assuming all are on this cell */
+ *procs = (orte_process_name_t*)malloc(orte_process_info.num_procs * sizeof(orte_process_name_t));
+ if (NULL == *procs) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ for (i=0; i < orte_process_info.num_procs; i++) {
+ (*procs)[i].cellid = ORTE_PROC_MY_NAME->cellid;
+ (*procs)[i].jobid = ORTE_PROC_MY_NAME->jobid;
+ (*procs)[i].vpid = orte_process_info.vpid_start + i;
+ }
+
+ *num_procs = orte_process_info.num_procs;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+ }
+
+ /* we get here if the job attribute was passed to us - use that jobid */
+ if (ORTE_SUCCESS != (rc = orte_dss.get((void**)&jptr, attr->value, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+
+ /* look up this job's record on the tracking database */
+ if (NULL == (job_info = orte_ns_replica_find_job(*jptr))) {
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_NOT_FOUND;
+ }
+
+ if (NULL != (attr = orte_rmgr.find_attribute(attrs, ORTE_NS_INCLUDE_DESCENDANTS))) {
+ /* we want the peers from this job AND ALL of its descendants - start by constructing
+ * a flattened list of the descendant jobs
+ */
+ OBJ_CONSTRUCT(&peerlist, opal_list_t);
+ child = OBJ_NEW(orte_ns_replica_jobitem_t);
+ child->jobid = job_info->jobid;
+ child->next_vpid = job_info->next_vpid;
+ opal_list_append(&peerlist, &child->super); /* add the current job to the list */
+ orte_ns_replica_construct_flattened_tree(&peerlist, job_info);
+
+ i = opal_list_get_size(&peerlist);
+ if (0 < i) {
+ npeers = 0;
+ for (item = opal_list_get_first(&peerlist);
+ item != opal_list_get_end(&peerlist);
+ item = opal_list_get_next(item)) {
+ child = (orte_ns_replica_jobitem_t*)item;
+ npeers += child->next_vpid;
+ }
+ if (0 >= npeers) {
+ *num_procs = npeers;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+ }
+
+ *procs = (orte_process_name_t*)malloc(npeers * sizeof(orte_process_name_t));
+ if (NULL == *procs) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ /* populate it from the list */
+ isave = 0;
+ while (NULL != (item = opal_list_remove_first(&peerlist))) {
+ child = (orte_ns_replica_jobitem_t*)item;
+ for (i=0; i < child->next_vpid; i++) {
+ (*procs)[i+isave].cellid = ORTE_PROC_MY_NAME->cellid;
+ (*procs)[i+isave].jobid = child->jobid;
+ (*procs)[i+isave].vpid = i;
+ }
+ isave += child->next_vpid;
+ }
+ }
+ *num_procs = npeers;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+ }
+
+ if (NULL != (attr = orte_rmgr.find_attribute(attrs, ORTE_NS_INCLUDE_CHILDREN))) {
+ /* we want the peers from this job AND ONLY its immediate children */
+
+ /* determine the number of peers we are going to have */
+ npeers = job_info->next_vpid;
+ for (item = opal_list_get_first(&job_info->children);
+ item != opal_list_get_end(&job_info->children);
+ item = opal_list_get_next(item)) {
+ child = (orte_ns_replica_jobitem_t*)item;
+ npeers += child->next_vpid;
+ }
+
+ /* create the array */
+ if (0 < npeers) {
+ *procs = (orte_process_name_t*)malloc(npeers * sizeof(orte_process_name_t));
+ if (NULL == *procs) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ /* populate it, starting with the specified job followed by its children */
+ for (i=0; i < job_info->next_vpid; i++) {
+ (*procs)[i].cellid = ORTE_PROC_MY_NAME->cellid;
+ (*procs)[i].jobid = *jptr;
+ (*procs)[i].vpid = i;
+ }
+ isave = job_info->next_vpid;
+ for (item = opal_list_get_first(&job_info->children);
+ item != opal_list_get_end(&job_info->children);
+ item = opal_list_get_next(item)) {
+ child = (orte_ns_replica_jobitem_t*)item;
+ for (i=0; i < child->next_vpid; i++) {
+ (*procs)[i+isave].cellid = ORTE_PROC_MY_NAME->cellid;
+ (*procs)[i+isave].jobid = child->jobid;
+ (*procs)[i+isave].vpid = i;
+ }
+ isave += child->next_vpid;
+ }
+ }
+ *num_procs = npeers;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+ }
+
+ /* get here if we want just the peers for the specified job */
+
+ /* create the array of peers */
+ if (0 < job_info->next_vpid) {
+ *procs = (orte_process_name_t*)malloc(job_info->next_vpid * sizeof(orte_process_name_t));
+ if (NULL == *procs) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ for (i=0; i < job_info->next_vpid; i++) {
+ (*procs)[i].cellid = ORTE_PROC_MY_NAME->cellid;
+ (*procs)[i].jobid = *jptr;
+ (*procs)[i].vpid = i;
+ }
+ }
+
+ *num_procs = job_info->next_vpid;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
+
+
+/*
+ * TAG SERVER functions
+ */
+int orte_ns_replica_assign_rml_tag(orte_rml_tag_t *tag,
+ char *name)
+{
+ orte_ns_replica_tagitem_t *tagitem, **tags;
+ orte_std_cntr_t i;
+ orte_rml_tag_t j;
+ int rc;
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ if (NULL != name) {
+ /* see if this name is already in list - if so, return tag */
+ tags = (orte_ns_replica_tagitem_t**)orte_ns_replica.tags->addr;
+ for (i=0, j=0; j < orte_ns_replica.num_tags &&
+ i < (orte_ns_replica.tags)->size; i++) {
+ if (NULL != tags[i]) {
+ j++;
+ if (tags[i]->name != NULL &&
+ 0 == strcmp(name, tags[i]->name)) { /* found name on list */
+ *tag = tags[i]->tag;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+ }
+ }
+ }
+ }
+
+ /* not in list or not provided, so allocate next tag */
+ *tag = ORTE_RML_TAG_MAX;
+
+ /* check if tag is available - need to do this since the tag type
+ * is probably not going to be a orte_std_cntr_t, so we cannot just rely
+ * on the pointer_array's size limits to protect us. NOTE: need to
+ * reserve ORTE_RML_TAG_MAX as an invalid value, so can't let
+ * num_tags get there
+ */
+ if (ORTE_RML_TAG_MAX-2 < orte_ns_replica.num_tags) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ tagitem = OBJ_NEW(orte_ns_replica_tagitem_t);
+ if (NULL == tagitem) { /* out of memory */
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ if (ORTE_SUCCESS != (rc = orte_pointer_array_add(&i,
+ orte_ns_replica.tags, tagitem))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+ tagitem->tag = orte_ns_replica.num_tags + ORTE_RML_TAG_DYNAMIC;
+ (orte_ns_replica.num_tags)++;
+ if (NULL != name) { /* provided - can look it up later */
+ tagitem->name = strdup(name);
+ } else {
+ tagitem->name = NULL;
+ }
+
+ *tag = tagitem->tag;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
+
+/*
+ * DATA TYPE SERVER functions
+ */
+int orte_ns_replica_define_data_type(const char *name,
+ orte_data_type_t *type)
+{
+ orte_ns_replica_dti_t **dti, *dtip;
+ orte_std_cntr_t i, j;
+ int rc;
+
+ if (NULL == name || 0 < *type) {
+ ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
+ return ORTE_ERR_BAD_PARAM;
+ }
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ dti = (orte_ns_replica_dti_t**)orte_ns_replica.dts->addr;
+ for (i=0, j=0; j < orte_ns_replica.num_dts &&
+ i < orte_ns_replica.dts->size; i++) {
+ if (NULL != dti[i]) {
+ j++;
+ if (dti[i]->name != NULL &&
+ 0 == strcmp(name, dti[i]->name)) { /* found name on list */
+ *type = dti[i]->id;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+ }
+ }
+ }
+
+ /* not in list or not provided, so allocate next id */
+ *type = ORTE_DSS_ID_MAX;
+
+ /* check if id is available - need to do this since the data type
+ * is probably not going to be a orte_std_cntr_t, so we cannot just rely
+ * on the pointer_array's size limits to protect us.
+ */
+ if (ORTE_DSS_ID_MAX-2 < orte_ns_replica.num_dts) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ dtip = OBJ_NEW(orte_ns_replica_dti_t);
+ if (NULL == dtip) { /* out of memory */
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ dtip->name = strdup(name);
+ if (ORTE_SUCCESS != (rc = orte_pointer_array_add(&i,
+ orte_ns_replica.dts, dtip))) {
+ ORTE_ERROR_LOG(rc);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return rc;
+ }
+ dtip->id = orte_ns_replica.num_dts;
+ (orte_ns_replica.num_dts)++;
+
+ *type = dtip->id;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
+/*
+ * NAME functions
+ */
+int orte_ns_replica_create_my_name(void)
+{
+ orte_jobid_t jobid;
+ orte_vpid_t vpid;
+ opal_list_t attrs;
+ int rc;
+
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ if (ORTE_SUCCESS != (rc = orte_ns.create_jobid(&jobid, &attrs))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&attrs);
+ return rc;
+ }
+ OBJ_DESTRUCT(&attrs);
+
+ if (ORTE_SUCCESS != (rc = orte_ns.reserve_range(jobid, 1, &vpid))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns.create_process_name(&(orte_process_info.my_name),
+ 0, jobid, vpid))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+
+ return ORTE_SUCCESS;
+}
+
diff --git a/orte/mca/ns/replica/ns_replica_job_fns.c b/orte/mca/ns/replica/ns_replica_job_fns.c
new file mode 100644
index 0000000000..cabd7ba590
--- /dev/null
+++ b/orte/mca/ns/replica/ns_replica_job_fns.c
@@ -0,0 +1,295 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+#include "orte_config.h"
+
+#include <stdio.h>
+#include <string.h>
+
+#include "opal/threads/mutex.h"
+#include "opal/util/output.h"
+#include "opal/util/trace.h"
+
+#include "orte/dss/dss.h"
+#include "orte/mca/errmgr/errmgr.h"
+#include "orte/mca/rmgr/rmgr.h"
+
+#include "ns_replica.h"
+
+/*
+ * JOBID functions
+ */
+int orte_ns_replica_create_jobid(orte_jobid_t *jobid, opal_list_t *attrs)
+{
+ orte_ns_replica_jobitem_t *child, *parent, *root;
+ orte_jobid_t parent_job=ORTE_JOBID_INVALID, *jptr;
+ orte_attribute_t *attr;
+ int rc;
+
+ OPAL_TRACE(1);
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ *jobid = ORTE_JOBID_INVALID;
+
+ /* check for attributes */
+ if (NULL != (attr = orte_rmgr.find_attribute(attrs, ORTE_NS_USE_PARENT))) {
+ /* declares the specified jobid to be the parent of the new one */
+ if (ORTE_SUCCESS != (rc = orte_dss.get((void**)&jptr, attr->value, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ parent_job = *jptr;
+ } else if (NULL != (attr = orte_rmgr.find_attribute(attrs, ORTE_NS_USE_ROOT))) {
+ /* use the root of the specified job as the parent of the new one */
+ if (ORTE_SUCCESS != (rc = orte_dss.get((void**)&jptr, attr->value, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ if (NULL == (root = orte_ns_replica_find_root_job(*jptr))) {
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+ return ORTE_ERR_NOT_FOUND;
+ }
+ parent_job = root->jobid;
+ }
+
+ /* if the parent jobid is INVALID, then this is the root of a new
+ * job family - create it
+ */
+ if (ORTE_JOBID_INVALID == parent_job) {
+ root = OBJ_NEW(orte_ns_replica_jobitem_t);
+ if (NULL == root) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ root->jobid = orte_ns_replica.num_jobids;
+ opal_list_append(&orte_ns_replica.jobs, &root->super);
+ *jobid = root->jobid;
+ (orte_ns_replica.num_jobids)++;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+ }
+
+ /* if the parent jobid is not INVALID, then the request is for a
+ * new child for this parent. Find the job's record
+ */
+ if (NULL == (parent = orte_ns_replica_find_job(parent_job))) {
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_NOT_FOUND;
+ }
+
+ /* add this new job to the parent's list of children */
+ child = OBJ_NEW(orte_ns_replica_jobitem_t);
+ if (NULL == child) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+ opal_list_append(&parent->children, &child->super);
+ child->jobid = orte_ns_replica.num_jobids;
+ *jobid = child->jobid;
+ (orte_ns_replica.num_jobids)++;
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_replica_get_job_descendants(orte_jobid_t **descendants, orte_std_cntr_t *num_desc, orte_jobid_t job)
+{
+ orte_std_cntr_t i, num;
+ orte_ns_replica_jobitem_t *ptr, *newptr;
+ orte_jobid_t *descs;
+ opal_list_t desc_list;
+ opal_list_item_t *item;
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ /* default values */
+ *descendants = NULL;
+ *num_desc = 0;
+
+ /* find this job's record on the tree */
+ if (NULL == (ptr = orte_ns_replica_find_job(job))) {
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_NOT_FOUND;
+ }
+
+ /* construct a flattened list of its descendants - including ourself */
+ OBJ_CONSTRUCT(&desc_list, opal_list_t);
+ newptr = OBJ_NEW(orte_ns_replica_jobitem_t);
+ newptr->jobid = job;
+ opal_list_append(&desc_list, &newptr->super);
+
+ orte_ns_replica_construct_flattened_tree(&desc_list, ptr);
+
+ /* count number of entries */
+ num = opal_list_get_size(&desc_list);
+
+ /* allocate memory for the array */
+ descs = (orte_jobid_t*)malloc(num * sizeof(orte_jobid_t));
+ if (NULL == descs) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ /* now fill in the array */
+ i = 0;
+ while (NULL != (item = opal_list_remove_first(&desc_list))) {
+ ptr = (orte_ns_replica_jobitem_t*)item;
+ descs[i++] = ptr->jobid;
+ OBJ_RELEASE(ptr);
+ }
+ OBJ_DESTRUCT(&desc_list);
+
+ *descendants = descs;
+ *num_desc = num;
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_replica_get_job_children(orte_jobid_t **children, orte_std_cntr_t *num_childs, orte_jobid_t job)
+{
+ orte_std_cntr_t i, num;
+ orte_ns_replica_jobitem_t *ptr, *newptr;
+ orte_jobid_t *descs;
+ opal_list_item_t *item;
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ /* default values */
+ *children = NULL;
+ *num_childs = 0;
+
+ /* find this job's record on the tree */
+ if (NULL == (ptr = orte_ns_replica_find_job(job))) {
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_NOT_FOUND;
+ }
+
+ /* count number of entries in our direct children - include ourselves */
+ num = 1 + opal_list_get_size(&ptr->children);
+
+ /* allocate memory for the array */
+ descs = (orte_jobid_t*)malloc(num * sizeof(orte_jobid_t));
+ if (NULL == descs) {
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+ }
+
+ /* now fill in the array - put ourselves first */
+ descs[0] = job;
+ i = 1;
+ for (item = opal_list_get_first(&ptr->children);
+ item != opal_list_get_end(&ptr->children);
+ item = opal_list_get_next(item)) {
+ newptr = (orte_ns_replica_jobitem_t*)item;
+ descs[i++] = newptr->jobid;
+ }
+
+ *children = descs;
+ *num_childs = num;
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
+int orte_ns_replica_get_root_job(orte_jobid_t *root_job, orte_jobid_t job)
+{
+ orte_ns_replica_jobitem_t *root;
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ if (NULL == (root = orte_ns_replica_find_root_job(job))) {
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_NOT_FOUND;
+ }
+
+ *root_job = root->jobid;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_replica_get_parent_job(orte_jobid_t *parent_job, orte_jobid_t job)
+{
+ opal_list_item_t *item;
+ orte_ns_replica_jobitem_t *root, *ptr, *parent;
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ /* find this job's parent object */
+ for (item = opal_list_get_first(&orte_ns_replica.jobs);
+ item != opal_list_get_end(&orte_ns_replica.jobs);
+ item = opal_list_get_next(item)) {
+ root = (orte_ns_replica_jobitem_t*)item;
+ if (NULL != (ptr = down_search(root, &parent, job))) {
+ goto REPORT;
+ }
+ }
+ /* don't report an error if not found, just return invalid */
+ *parent_job = ORTE_JOBID_INVALID;
+ return ORTE_ERR_NOT_FOUND;
+
+REPORT:
+ /* return the info */
+ *parent_job = parent->jobid;
+
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+}
+
+
+int orte_ns_replica_reserve_range(orte_jobid_t job, orte_vpid_t range,
+ orte_vpid_t *start)
+{
+ orte_ns_replica_jobitem_t *ptr;
+
+ OPAL_TRACE(1);
+
+ OPAL_THREAD_LOCK(&orte_ns_replica.mutex);
+
+ /* find the job's record */
+ if (NULL == (ptr = orte_ns_replica_find_job(job))) {
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_NOT_FOUND;
+ }
+
+ if ((ORTE_VPID_MAX-range-(ptr->next_vpid)) > 0) {
+ *start = ptr->next_vpid;
+ ptr->next_vpid += range;
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_SUCCESS;
+ }
+
+ /* get here if the range isn't available */
+ ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
+ OPAL_THREAD_UNLOCK(&orte_ns_replica.mutex);
+ return ORTE_ERR_OUT_OF_RESOURCE;
+}
diff --git a/orte/mca/ns/replica/ns_replica_recv.c b/orte/mca/ns/replica/ns_replica_recv.c
new file mode 100644
index 0000000000..559ce1b235
--- /dev/null
+++ b/orte/mca/ns/replica/ns_replica_recv.c
@@ -0,0 +1,457 @@
+/* -*- C -*-
+*
+* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+* University Research and Technology
+* Corporation. All rights reserved.
+* Copyright (c) 2004-2005 The University of Tennessee and The University
+* of Tennessee Research Foundation. All rights
+* reserved.
+* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+* University of Stuttgart. All rights reserved.
+* Copyright (c) 2004-2005 The Regents of the University of California.
+* All rights reserved.
+* $COPYRIGHT$
+*
+* Additional copyrights may follow
+*
+* $HEADER$
+*/
+/** @file:
+*
+* The Open MPI Name Server
+*
+*/
+
+/*
+ * includes
+ */
+#include "orte_config.h"
+
+#include "orte/orte_constants.h"
+#include "orte/orte_types.h"
+
+#include "opal/threads/mutex.h"
+#include "opal/class/opal_list.h"
+#include "opal/util/output.h"
+
+#include "opal/mca/mca.h"
+#include "opal/mca/base/mca_base_param.h"
+
+#include "orte/util/proc_info.h"
+#include "orte/mca/errmgr/errmgr.h"
+#include "orte/mca/rml/rml.h"
+
+#include "orte/mca/ns/base/ns_private.h"
+#include "ns_replica.h"
+
+
+/*
+ * handle message from proxies
+ * NOTE: The incoming buffer "buffer" is OBJ_RELEASED by the calling program.
+ * DO NOT RELEASE THIS BUFFER IN THIS CODE
+ */
+
+void orte_ns_replica_recv(int status, orte_process_name_t* sender,
+ orte_buffer_t* buffer, orte_rml_tag_t tag,
+ void* cbdata)
+{
+ orte_buffer_t answer, error_answer;
+ orte_ns_cmd_flag_t command;
+ opal_list_t attrs;
+ orte_cellid_t cell;
+ orte_jobid_t job, root, *descendants;
+ orte_vpid_t startvpid, range;
+ char *tagname, *site, *resource;
+ orte_rml_tag_t oob_tag;
+ orte_data_type_t type;
+ orte_std_cntr_t count, nprocs, nret;
+ orte_process_name_t *procs;
+ int rc=ORTE_SUCCESS, ret;
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &command, &count, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ rc = ORTE_ERR_BAD_PARAM;
+ goto RETURN_ERROR;
+ }
+
+ OBJ_CONSTRUCT(&answer, orte_buffer_t);
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, &command, 1, ORTE_NS_CMD))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ switch (command) {
+ case ORTE_NS_CREATE_CELLID_CMD:
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &site, &count, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ rc = ORTE_ERR_BAD_PARAM;
+ goto RETURN_ERROR;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &resource, &count, ORTE_STRING))) {
+ ORTE_ERROR_LOG(rc);
+ rc = ORTE_ERR_BAD_PARAM;
+ goto RETURN_ERROR;
+ }
+
+ rc = orte_ns_replica_create_cellid(&cell, site, resource);
+
+ if (ORTE_SUCCESS != (ret = orte_dss.pack(&answer, &cell, 1, ORTE_CELLID))) {
+ ORTE_ERROR_LOG(ret);
+ goto RETURN_ERROR;
+ }
+
+ if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_GET_CELL_INFO_CMD:
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &cell, &count, ORTE_CELLID))) {
+ ORTE_ERROR_LOG(rc);
+ rc = ORTE_ERR_BAD_PARAM;
+ goto RETURN_ERROR;
+ }
+
+ site = NULL;
+ resource = NULL;
+ rc = orte_ns_replica_get_cell_info(cell, &site, &resource);
+
+ if (ORTE_SUCCESS != (ret = orte_dss.pack(&answer, &site, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(ret);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (ret = orte_dss.pack(&answer, &resource, 1, ORTE_STRING))) {
+ ORTE_ERROR_LOG(ret);
+ goto RETURN_ERROR;
+ }
+
+ if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_CREATE_NODEID_CMD:
+ case ORTE_NS_GET_NODE_INFO_CMD:
+ ORTE_ERROR_LOG(ORTE_ERR_NOT_IMPLEMENTED);
+ goto RETURN_ERROR;
+ break;
+
+ case ORTE_NS_CREATE_JOBID_CMD:
+ /* get the list of attributes */
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ count = 1;
+ if(ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &attrs, &count, ORTE_ATTR_LIST))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_create_jobid(&job, &attrs))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&attrs);
+ goto RETURN_ERROR;
+ }
+ OBJ_DESTRUCT(&attrs);
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&job, 1, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_GET_JOB_DESC_CMD:
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, (void*)&job, &count, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_get_job_descendants(&descendants, &nret, job))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&nret, 1, ORTE_STD_CNTR))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (0 < nret) {
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)descendants, nret, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+ }
+
+ if (0 > (rc = orte_rml.send_buffer(sender, &answer, tag, 0))) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_GET_JOB_CHILD_CMD:
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, (void*)&job, &count, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_get_job_children(&descendants, &nret, job))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&nret, 1, ORTE_STD_CNTR))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (0 < nret) {
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)descendants, nret, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+ }
+
+ if (0 > (rc = orte_rml.send_buffer(sender, &answer, tag, 0))) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_GET_ROOT_JOB_CMD:
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, (void*)&job, &count, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_get_root_job(&root, job))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&root, 1, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (0 > (rc = orte_rml.send_buffer(sender, &answer, tag, 0))) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_GET_PARENT_JOB_CMD:
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, (void*)&job, &count, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_get_parent_job(&root, job))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&root, 1, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (0 > (rc = orte_rml.send_buffer(sender, &answer, tag, 0))) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_RESERVE_RANGE_CMD:
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, (void*)&job, &count, ORTE_JOBID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, (void*)&range, &count, ORTE_VPID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_reserve_range(job, range, &startvpid))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&startvpid, 1, ORTE_VPID))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (0 > (rc = orte_rml.send_buffer(sender, &answer, tag, 0))) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_ASSIGN_OOB_TAG_CMD:
+ count = 1;
+ if (0 > orte_dss.unpack(buffer, &tagname, &count, ORTE_STRING)) {
+ rc = ORTE_ERR_UNPACK_FAILURE;
+ goto RETURN_ERROR;
+ }
+
+ if (0 == strncmp(tagname, "NULL", 4)) {
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_assign_rml_tag(&oob_tag, NULL))) {
+ goto RETURN_ERROR;
+ }
+ } else {
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_assign_rml_tag(&oob_tag, tagname))) {
+ goto RETURN_ERROR;
+ }
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&oob_tag, 1, ORTE_RML_TAG))) {
+ goto RETURN_ERROR;
+ }
+
+ if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_DEFINE_DATA_TYPE_CMD:
+ count = 1;
+ if (0 > orte_dss.unpack(buffer, &tagname, &count, ORTE_STRING)) {
+ rc = ORTE_ERR_UNPACK_FAILURE;
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_define_data_type(tagname, &type))) {
+ goto RETURN_ERROR;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&type, 1, ORTE_DATA_TYPE))) {
+ goto RETURN_ERROR;
+ }
+
+ if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_CREATE_MY_NAME_CMD:
+ /* ignore this command */
+ break;
+
+ case ORTE_NS_GET_PEERS_CMD:
+ /* get the list of attributes */
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ count = 1;
+ if(ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &attrs, &count, ORTE_ATTR_LIST))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ /* process the request */
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_get_peers(&procs, &nprocs, &attrs))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&attrs);
+ goto RETURN_ERROR;
+ }
+ OBJ_DESTRUCT(&attrs);
+
+ /* pack the answer */
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, &nprocs, 1, ORTE_STD_CNTR))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+
+ if (nprocs > 0) {
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, procs, nprocs, ORTE_NAME))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+ }
+
+ if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_DUMP_CELLS_CMD:
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_cells_fn(&answer))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+ if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_DUMP_JOBIDS_CMD:
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_jobs_fn(&answer))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+ if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_DUMP_TAGS_CMD:
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_tags_fn(&answer))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+ if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ case ORTE_NS_DUMP_DATATYPES_CMD:
+ if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_datatypes_fn(&answer))) {
+ ORTE_ERROR_LOG(rc);
+ goto RETURN_ERROR;
+ }
+ if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
+ ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
+ goto RETURN_ERROR;
+ }
+ break;
+
+ default:
+ goto RETURN_ERROR;
+ }
+ goto CLEANUP;
+
+RETURN_ERROR:
+ OBJ_CONSTRUCT(&error_answer, orte_buffer_t);
+ orte_dss.pack(&error_answer, (void*)&command, 1, ORTE_NS_CMD);
+ orte_dss.pack(&error_answer, (void*)&rc, 1, ORTE_INT32);
+ orte_rml.send_buffer(sender, &error_answer, tag, 0);
+ OBJ_DESTRUCT(&error_answer);
+
+CLEANUP:
+ /* cleanup */
+ OBJ_DESTRUCT(&answer);
+}
+
diff --git a/orte/mca/ns/replica/ns_replica_support_fns.c b/orte/mca/ns/replica/ns_replica_support_fns.c
new file mode 100644
index 0000000000..fa2c2da7fc
--- /dev/null
+++ b/orte/mca/ns/replica/ns_replica_support_fns.c
@@ -0,0 +1,120 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+/** @file:
+ *
+ */
+#include "orte_config.h"
+
+#include <stdio.h>
+#include <string.h>
+
+#include "opal/util/output.h"
+#include "opal/util/trace.h"
+
+#include "orte/mca/errmgr/errmgr.h"
+
+#include "ns_replica.h"
+
+orte_ns_replica_jobitem_t *down_search(orte_ns_replica_jobitem_t *ptr,
+ orte_ns_replica_jobitem_t **parent_ptr,
+ orte_jobid_t job)
+{
+ opal_list_item_t *item;
+ orte_ns_replica_jobitem_t *ptr2, *ptr3;
+
+ /* check if this is the specified job */
+ if (ptr->jobid == job) {
+ return ptr;
+ }
+
+ /* otherwise, look at the children of this ptr. call ourselves
+ * to check each one
+ */
+ for (item = opal_list_get_first(&ptr->children);
+ item != opal_list_get_end(&ptr->children);
+ item = opal_list_get_next(item)) {
+ ptr2 = (orte_ns_replica_jobitem_t*)item;
+ *parent_ptr = ptr;
+ if (NULL != (ptr3 = down_search(ptr2, parent_ptr, job))) {
+ return ptr3;
+ }
+ }
+
+ return NULL;
+}
+
+/* find a job's record, wherever it is on the tree */
+orte_ns_replica_jobitem_t* orte_ns_replica_find_job(orte_jobid_t job)
+{
+ opal_list_item_t *item;
+ orte_ns_replica_jobitem_t *root, *ptr, *parent;
+
+ for (item = opal_list_get_first(&orte_ns_replica.jobs);
+ item != opal_list_get_end(&orte_ns_replica.jobs);
+ item = opal_list_get_next(item)) {
+ root = (orte_ns_replica_jobitem_t*)item;
+ if (NULL != (ptr = down_search(root, &parent, job))) {
+ return ptr;
+ }
+ }
+
+ /* don't report an error if not found, just return NULL */
+ return NULL;
+}
+
+/* given a jobid, find it's root job's object */
+orte_ns_replica_jobitem_t* orte_ns_replica_find_root_job(orte_jobid_t job)
+{
+ opal_list_item_t *item;
+ orte_ns_replica_jobitem_t *root, *ptr, *parent;
+
+ for (item = opal_list_get_first(&orte_ns_replica.jobs);
+ item != opal_list_get_end(&orte_ns_replica.jobs);
+ item = opal_list_get_next(item)) {
+ root = (orte_ns_replica_jobitem_t*)item;
+
+ if (NULL != (ptr = down_search(root, &parent, job))) {
+ return root;
+ }
+ }
+
+ /* don't report an error if not found, just return NULL */
+ return NULL;
+}
+
+/* given a job's record, construct a flattened list of the descendants below it,
+ * including the starting point
+ */
+void orte_ns_replica_construct_flattened_tree(opal_list_t *tree, orte_ns_replica_jobitem_t *ptr)
+{
+ orte_ns_replica_jobitem_t *job, *newjob;
+ opal_list_item_t *item;
+
+ for (item = opal_list_get_first(&ptr->children);
+ item != opal_list_get_end(&ptr->children);
+ item = opal_list_get_next(item)) {
+ job = (orte_ns_replica_jobitem_t*)item;
+
+ newjob = OBJ_NEW(orte_ns_replica_jobitem_t);
+ newjob->jobid = job->jobid;
+ newjob->next_vpid = job->next_vpid;
+ opal_list_append(tree, &newjob->super);
+
+ orte_ns_replica_construct_flattened_tree(tree, job); /* get anyone below this one */
+ }
+}
diff --git a/orte/mca/ns/replica/src/Makefile.extra b/orte/mca/ns/replica/src/Makefile.extra
deleted file mode 100644
index 3c3478e9c5..0000000000
--- a/orte/mca/ns/replica/src/Makefile.extra
+++ /dev/null
@@ -1,24 +0,0 @@
-# -*- makefile -*-
-#
-# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
-# University Research and Technology
-# Corporation. All rights reserved.
-# Copyright (c) 2004-2005 The University of Tennessee and The University
-# of Tennessee Research Foundation. All rights
-# reserved.
-# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
-# University of Stuttgart. All rights reserved.
-# Copyright (c) 2004-2005 The Regents of the University of California.
-# All rights reserved.
-# $COPYRIGHT$
-#
-# Additional copyrights may follow
-#
-# $HEADER$
-#
-
-sources += \
- src/ns_replica.h \
- src/ns_replica.c \
- src/ns_replica_component.c
-
diff --git a/orte/mca/ns/replica/src/ns_replica_component.c b/orte/mca/ns/replica/src/ns_replica_component.c
deleted file mode 100644
index b126bf36fa..0000000000
--- a/orte/mca/ns/replica/src/ns_replica_component.c
+++ /dev/null
@@ -1,702 +0,0 @@
-/* -*- C -*-
- *
- * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
- * University Research and Technology
- * Corporation. All rights reserved.
- * Copyright (c) 2004-2005 The University of Tennessee and The University
- * of Tennessee Research Foundation. All rights
- * reserved.
- * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
- * University of Stuttgart. All rights reserved.
- * Copyright (c) 2004-2005 The Regents of the University of California.
- * All rights reserved.
- * $COPYRIGHT$
- *
- * Additional copyrights may follow
- *
- * $HEADER$
- */
-/** @file:
- *
- * The Open MPI Name Server
- *
- * The Open MPI Name Server provides unique name ranges for processes
- * within the universe. Each universe will have one name server
- * running within the seed daemon. This is done to prevent the
- * inadvertent duplication of names.
- */
-
-/*
- * includes
- */
-#include "orte_config.h"
-
-#include "orte/orte_constants.h"
-#include "orte/orte_types.h"
-
-#include "opal/threads/mutex.h"
-#include "orte/util/proc_info.h"
-#include "opal/util/output.h"
-
-#include "opal/mca/mca.h"
-#include "opal/mca/base/mca_base_param.h"
-#include "orte/mca/errmgr/errmgr.h"
-#include "orte/mca/rml/rml.h"
-#include "ns_replica.h"
-
-
-/*
- * Struct of function pointers that need to be initialized
- */
-mca_ns_base_component_t mca_ns_replica_component = {
- {
- MCA_NS_BASE_VERSION_1_0_0,
-
- "replica", /* MCA module name */
- ORTE_MAJOR_VERSION, /* MCA module major version */
- ORTE_MINOR_VERSION, /* MCA module minor version */
- ORTE_RELEASE_VERSION, /* MCA module release version */
- orte_ns_replica_open, /* module open */
- orte_ns_replica_close /* module close */
- },
- {
- false /* checkpoint / restart */
- },
- orte_ns_replica_init, /* module init */
- orte_ns_replica_finalize /* module shutdown */
-};
-
-/*
- * setup the function pointers for the module
- */
-static mca_ns_base_module_t orte_ns_replica_module = {
- /* init */
- orte_ns_replica_module_init,
- /* cell functions */
- orte_ns_replica_create_cellid,
- orte_ns_base_get_cellid,
- orte_ns_replica_get_cell_info,
- orte_ns_base_assign_cellid_to_process,
- orte_ns_base_get_cellid_string,
- orte_ns_base_convert_cellid_to_string,
- orte_ns_base_convert_string_to_cellid,
- /* jobid functions */
- orte_ns_replica_create_jobid,
- orte_ns_base_get_jobid,
- orte_ns_base_get_jobid_string,
- orte_ns_base_convert_jobid_to_string,
- orte_ns_base_convert_string_to_jobid,
- /* vpid functions */
- orte_ns_replica_reserve_range,
- orte_ns_base_get_vpid,
- orte_ns_base_get_vpid_string,
- orte_ns_base_convert_vpid_to_string,
- orte_ns_base_convert_string_to_vpid,
- /* name functions */
- orte_ns_base_create_process_name,
- orte_ns_replica_create_my_name,
- orte_ns_base_copy_process_name,
- orte_ns_base_convert_string_to_process_name,
- orte_ns_base_free_name,
- orte_ns_base_get_proc_name_string,
- orte_ns_base_compare,
- /* peer functions */
- orte_ns_base_get_peers,
- orte_ns_replica_get_job_peers,
- /* tag server functions */
- orte_ns_replica_assign_rml_tag,
- /* data type functions */
- orte_ns_replica_define_data_type,
- /* diagnostic functions */
- orte_ns_replica_dump_cells,
- orte_ns_replica_dump_jobs,
- orte_ns_replica_dump_tags,
- orte_ns_replica_dump_datatypes
-};
-
-/*
- * Whether or not we allowed this component to be selected
- */
-static bool initialized = false;
-
-
-/* constructor - used to initialize state of cell_tracker instance */
-static void orte_ns_replica_cell_tracker_construct(orte_ns_replica_cell_tracker_t* cell_tracker)
-{
- cell_tracker->cell = 0;
- cell_tracker->site = NULL;
- cell_tracker->resource = NULL;
-}
-
-/* destructor - used to free any resources held by instance */
-static void orte_ns_replica_cell_tracker_destructor(orte_ns_replica_cell_tracker_t* cell_tracker)
-{
- if (NULL != cell_tracker->site) free(cell_tracker->site);
- if (NULL != cell_tracker->resource) free(cell_tracker->resource);
-}
-
-/* define instance of opal_class_t */
-OBJ_CLASS_INSTANCE(
- orte_ns_replica_cell_tracker_t, /* type name */
- opal_object_t, /* parent "class" name */
- orte_ns_replica_cell_tracker_construct, /* constructor */
- orte_ns_replica_cell_tracker_destructor); /* destructor */
-
-
-/* constructor - used to initialize state of jobid_tracker instance */
-static void orte_ns_replica_jobid_tracker_construct(orte_ns_replica_jobid_tracker_t* jobid_tracker)
-{
- jobid_tracker->jobid = ORTE_JOBID_MAX;
- jobid_tracker->next_vpid = 0;
-}
-
-/* destructor - used to free any resources held by instance */
-static void orte_ns_replica_jobid_tracker_destructor(orte_ns_replica_jobid_tracker_t* jobid_tracker){
-}
-
-/* define instance of opal_class_t */
-OBJ_CLASS_INSTANCE(
- orte_ns_replica_jobid_tracker_t, /* type name */
- opal_object_t, /* parent "class" name */
- orte_ns_replica_jobid_tracker_construct, /* constructor */
- orte_ns_replica_jobid_tracker_destructor); /* destructor */
-
-
-/* constructor - used to initialize state of taglist instance */
-static void orte_ns_replica_tagitem_construct(orte_ns_replica_tagitem_t* tagitem)
-{
- tagitem->tag = ORTE_RML_TAG_MAX;
- tagitem->name = NULL;
-}
-
-/* destructor - used to free any resources held by instance */
-static void orte_ns_replica_tagitem_destructor(orte_ns_replica_tagitem_t* tagitem)
-{
- if (NULL != tagitem->name) {
- free(tagitem->name);
- }
-}
-
-/* define instance of opal_class_t */
-OBJ_CLASS_INSTANCE(
- orte_ns_replica_tagitem_t, /* type name */
- opal_object_t, /* parent "class" name */
- orte_ns_replica_tagitem_construct, /* constructor */
- orte_ns_replica_tagitem_destructor); /* destructor */
-
-
-/* constructor - used to initialize state of dtilist instance */
-static void orte_ns_replica_dti_construct(orte_ns_replica_dti_t* dti)
-{
- dti->id = ORTE_DSS_ID_MAX;
- dti->name = NULL;
-}
-
-/* destructor - used to free any resources held by instance */
-static void orte_ns_replica_dti_destructor(orte_ns_replica_dti_t* dti)
-{
- if (NULL != dti->name) {
- free(dti->name);
- }
-}
-
-/* define instance of opal_class_t */
-OBJ_CLASS_INSTANCE(
- orte_ns_replica_dti_t, /* type name */
- opal_object_t, /* parent "class" name */
- orte_ns_replica_dti_construct, /* constructor */
- orte_ns_replica_dti_destructor); /* destructor */
-
-/*
- * globals needed within replica component
- */
-orte_ns_replica_globals_t orte_ns_replica;
-
-/*
- * don't really need this function - could just put NULL in the above structure
- * Just holding the place in case we decide there is something we need to do
- */
-int orte_ns_replica_open(void)
-{
- int id, param;
-
- id = mca_base_param_register_int("ns", "replica", "debug", NULL, (int)false);
- mca_base_param_lookup_int(id, &orte_ns_replica.debug);
-
- id = mca_base_param_register_int("ns", "replica", "isolate", NULL, (int)false);
- mca_base_param_lookup_int(id, ¶m);
- if (param) {
- orte_ns_replica.isolate = true;
- } else {
- orte_ns_replica.isolate = false;
- }
-
- id = mca_base_param_register_int("ns", "replica", "maxsize", NULL,
- ORTE_NS_ARRAY_MAX_SIZE);
- mca_base_param_lookup_int(id, ¶m);
- orte_ns_replica.max_size = (size_t)param;
-
- id = mca_base_param_register_int("ns", "replica", "blocksize", NULL,
- ORTE_NS_ARRAY_BLOCK_SIZE);
- mca_base_param_lookup_int(id, ¶m);
- orte_ns_replica.block_size = (size_t)param;
-
- return ORTE_SUCCESS;
-}
-
-/*
- * ditto for this one
- */
-int orte_ns_replica_close(void)
-{
- return ORTE_SUCCESS;
-}
-
-mca_ns_base_module_t* orte_ns_replica_init(int *priority)
-{
- int rc;
-
- /* If we are to host a replica, then we want to be selected, so do all the
- setup and return the module */
-
- if (NULL == orte_process_info.ns_replica_uri) {
-
- /* Return a module (choose an arbitrary, positive priority --
- it's only relevant compared to other ns components). If
- we're not the seed, then we don't want to be selected, so
- return NULL. */
-
- *priority = 50;
-
- /* initialize the cell info tracker */
- if (ORTE_SUCCESS != (rc = orte_pointer_array_init(&(orte_ns_replica.cells),
- (orte_std_cntr_t)orte_ns_replica.block_size,
- (orte_std_cntr_t)orte_ns_replica.max_size,
- (orte_std_cntr_t)orte_ns_replica.block_size))) {
- ORTE_ERROR_LOG(rc);
- return NULL;
- }
- orte_ns_replica.num_cells = 0;
-
- /* initialize the job id tracker */
- if (ORTE_SUCCESS != (rc = orte_pointer_array_init(&(orte_ns_replica.jobids),
- (orte_std_cntr_t)orte_ns_replica.block_size,
- (orte_std_cntr_t)orte_ns_replica.max_size,
- (orte_std_cntr_t)orte_ns_replica.block_size))) {
- ORTE_ERROR_LOG(rc);
- return NULL;
- }
- orte_ns_replica.num_jobids = 0;
-
- /* initialize the taglist */
-
- if (ORTE_SUCCESS != (rc = orte_pointer_array_init(&(orte_ns_replica.tags),
- (orte_std_cntr_t)orte_ns_replica.block_size,
- (orte_std_cntr_t)orte_ns_replica.max_size,
- (orte_std_cntr_t)orte_ns_replica.block_size))) {
- ORTE_ERROR_LOG(rc);
- return NULL;
- }
- orte_ns_replica.num_tags = 0;
-
- /* initialize the dtlist */
-
- if (ORTE_SUCCESS != (rc = orte_pointer_array_init(&(orte_ns_replica.dts),
- (orte_std_cntr_t)orte_ns_replica.block_size,
- (orte_std_cntr_t)orte_ns_replica.max_size,
- (orte_std_cntr_t)orte_ns_replica.block_size))) {
- ORTE_ERROR_LOG(rc);
- return NULL;
- }
- orte_ns_replica.num_dts = 0;
-
- /* setup the thread lock */
- OBJ_CONSTRUCT(&orte_ns_replica.mutex, opal_mutex_t);
-
- /* Return the module */
-
- initialized = true;
- return &orte_ns_replica_module;
- } else {
- return NULL;
- }
-}
-
-int orte_ns_replica_module_init(void)
-{
- int rc;
- if (orte_ns_replica.isolate) {
- return ORTE_SUCCESS;
- }
-
- /* issue non-blocking receive for call_back function */
- rc = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY, ORTE_RML_TAG_NS, ORTE_RML_PERSISTENT, orte_ns_replica_recv, NULL);
- if(rc < 0) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
- return ORTE_SUCCESS;
-}
-
-
-/*
- * finalize routine
- */
-int orte_ns_replica_finalize(void)
-{
- orte_ns_replica_cell_tracker_t **cptr;
- orte_ns_replica_jobid_tracker_t **jptr;
- orte_ns_replica_tagitem_t **tag;
- orte_ns_replica_dti_t **dti;
- orte_std_cntr_t i;
-
- /* free all tracking storage, but only if this component was initialized */
-
- if (initialized) {
- cptr = (orte_ns_replica_cell_tracker_t**)(orte_ns_replica.cells)->addr;
- for (i=0; i < (orte_ns_replica.cells)->size; i++) {
- if (NULL != cptr[i]) {
- OBJ_RELEASE(cptr[i]);
- }
- }
- OBJ_RELEASE(orte_ns_replica.cells);
-
- jptr = (orte_ns_replica_jobid_tracker_t**)(orte_ns_replica.jobids)->addr;
- for (i=0; i < (orte_ns_replica.jobids)->size; i++) {
- if (NULL != jptr[i]) {
- OBJ_RELEASE(jptr[i]);
- }
- }
- OBJ_RELEASE(orte_ns_replica.jobids);
-
- tag = (orte_ns_replica_tagitem_t**)(orte_ns_replica.tags)->addr;
- for (i=0; i < (orte_ns_replica.tags)->size; i++) {
- if (NULL != tag[i]) OBJ_RELEASE(tag[i]);
- }
- OBJ_RELEASE(orte_ns_replica.tags);
-
- dti = (orte_ns_replica_dti_t**)(orte_ns_replica.dts)->addr;
- for (i=0; i < (orte_ns_replica.dts)->size; i++) {
- if (NULL != dti[i]) OBJ_RELEASE(dti[i]);
- }
- OBJ_RELEASE(orte_ns_replica.dts);
-
- initialized = false;
- }
-
- /* All done */
- if (orte_ns_replica.isolate) {
- return ORTE_SUCCESS;
- }
-
- orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_NS);
- return ORTE_SUCCESS;
-}
-
-
-/*
- * handle message from proxies
- * NOTE: The incoming buffer "buffer" is OBJ_RELEASED by the calling program.
- * DO NOT RELEASE THIS BUFFER IN THIS CODE
- */
-
-void orte_ns_replica_recv(int status, orte_process_name_t* sender,
- orte_buffer_t* buffer, orte_rml_tag_t tag,
- void* cbdata)
-{
- orte_buffer_t answer, error_answer;
- orte_ns_cmd_flag_t command;
- orte_cellid_t cell;
- orte_jobid_t job;
- orte_vpid_t startvpid, range;
- char *tagname, *site, *resource;
- orte_rml_tag_t oob_tag;
- orte_data_type_t type;
- orte_std_cntr_t count, nprocs;
- orte_process_name_t *procs;
- int rc=ORTE_SUCCESS, ret;
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &command, &count, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- rc = ORTE_ERR_BAD_PARAM;
- goto RETURN_ERROR;
- }
-
- OBJ_CONSTRUCT(&answer, orte_buffer_t);
-
- switch (command) {
- case ORTE_NS_CREATE_CELLID_CMD:
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, &command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &site, &count, ORTE_STRING))) {
- ORTE_ERROR_LOG(rc);
- rc = ORTE_ERR_BAD_PARAM;
- goto RETURN_ERROR;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &resource, &count, ORTE_STRING))) {
- ORTE_ERROR_LOG(rc);
- rc = ORTE_ERR_BAD_PARAM;
- goto RETURN_ERROR;
- }
-
- rc = orte_ns_replica_create_cellid(&cell, site, resource);
-
- if (ORTE_SUCCESS != (ret = orte_dss.pack(&answer, &cell, 1, ORTE_CELLID))) {
- ORTE_ERROR_LOG(ret);
- goto RETURN_ERROR;
- }
- if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- goto RETURN_ERROR;
- }
- break;
-
- case ORTE_NS_GET_CELL_INFO_CMD:
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, &command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &cell, &count, ORTE_CELLID))) {
- ORTE_ERROR_LOG(rc);
- rc = ORTE_ERR_BAD_PARAM;
- goto RETURN_ERROR;
- }
-
- site = NULL;
- resource = NULL;
- rc = orte_ns_replica_get_cell_info(cell, &site, &resource);
-
- if (ORTE_SUCCESS != (ret = orte_dss.pack(&answer, &site, 1, ORTE_STRING))) {
- ORTE_ERROR_LOG(ret);
- goto RETURN_ERROR;
- }
-
- if (ORTE_SUCCESS != (ret = orte_dss.pack(&answer, &resource, 1, ORTE_STRING))) {
- ORTE_ERROR_LOG(ret);
- goto RETURN_ERROR;
- }
-
- if (ORTE_SUCCESS != (ret = orte_dss.pack(&answer, &rc, 1, ORTE_INT))) {
- ORTE_ERROR_LOG(ret);
- goto RETURN_ERROR;
- }
-
- if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- goto RETURN_ERROR;
- }
- break;
-
- case ORTE_NS_CREATE_JOBID_CMD:
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- if (ORTE_SUCCESS != (rc = orte_ns_replica_create_jobid(&job))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&job, 1, ORTE_JOBID))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- goto RETURN_ERROR;
- }
- break;
-
- case ORTE_NS_RESERVE_RANGE_CMD:
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, (void*)&job, &count, ORTE_JOBID))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, (void*)&range, &count, ORTE_VPID))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- if (ORTE_SUCCESS != (rc = orte_ns_replica_reserve_range(job, range, &startvpid))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&command, 1, ORTE_NS_CMD))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&startvpid, 1, ORTE_VPID))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- if (0 > (rc = orte_rml.send_buffer(sender, &answer, tag, 0))) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- goto RETURN_ERROR;
- }
- break;
-
- case ORTE_NS_ASSIGN_OOB_TAG_CMD:
- count = 1;
- if (0 > orte_dss.unpack(buffer, &tagname, &count, ORTE_STRING)) {
- rc = ORTE_ERR_UNPACK_FAILURE;
- goto RETURN_ERROR;
- }
-
- if (0 == strncmp(tagname, "NULL", 4)) {
- if (ORTE_SUCCESS != (rc = orte_ns_replica_assign_rml_tag(&oob_tag, NULL))) {
- goto RETURN_ERROR;
- }
- } else {
- if (ORTE_SUCCESS != (rc = orte_ns_replica_assign_rml_tag(&oob_tag, tagname))) {
- goto RETURN_ERROR;
- }
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&command, 1, ORTE_NS_CMD))) {
- goto RETURN_ERROR;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&oob_tag, 1, ORTE_RML_TAG))) {
- goto RETURN_ERROR;
- }
-
- if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- goto RETURN_ERROR;
- }
- break;
-
- case ORTE_NS_DEFINE_DATA_TYPE_CMD:
- count = 1;
- if (0 > orte_dss.unpack(buffer, &tagname, &count, ORTE_STRING)) {
- rc = ORTE_ERR_UNPACK_FAILURE;
- goto RETURN_ERROR;
- }
-
- if (ORTE_SUCCESS != (rc = orte_ns_replica_define_data_type(tagname, &type))) {
- goto RETURN_ERROR;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&command, 1, ORTE_NS_CMD))) {
- goto RETURN_ERROR;
- }
-
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, (void*)&type, 1, ORTE_DATA_TYPE))) {
- goto RETURN_ERROR;
- }
-
- if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- goto RETURN_ERROR;
- }
- break;
-
- case ORTE_NS_CREATE_MY_NAME_CMD:
- /* ignore this command */
- break;
-
- case ORTE_NS_GET_JOB_PEERS_CMD:
- /* unpack the jobid */
- count = 1;
- if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &job, &count, ORTE_JOBID))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
- /* process the request */
- if (ORTE_SUCCESS != (rc = orte_ns_replica_get_job_peers(&procs, &nprocs, job))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- /* pack the answer */
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, &nprocs, 1, ORTE_STD_CNTR))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
-
- if (nprocs > 0) {
- if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, &procs, nprocs, ORTE_NAME))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
- }
- break;
-
- case ORTE_NS_DUMP_CELLS_CMD:
- if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_cells_fn(&answer))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
- if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- goto RETURN_ERROR;
- }
- break;
-
- case ORTE_NS_DUMP_JOBIDS_CMD:
- if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_jobs_fn(&answer))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
- if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- goto RETURN_ERROR;
- }
- break;
-
- case ORTE_NS_DUMP_TAGS_CMD:
- if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_tags_fn(&answer))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
- if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- goto RETURN_ERROR;
- }
- break;
-
- case ORTE_NS_DUMP_DATATYPES_CMD:
- if (ORTE_SUCCESS != (rc = orte_ns_replica_dump_datatypes_fn(&answer))) {
- ORTE_ERROR_LOG(rc);
- goto RETURN_ERROR;
- }
- if (0 > orte_rml.send_buffer(sender, &answer, tag, 0)) {
- ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
- goto RETURN_ERROR;
- }
- break;
-
- default:
- goto RETURN_ERROR;
- }
- goto CLEANUP;
-
-RETURN_ERROR:
- OBJ_CONSTRUCT(&error_answer, orte_buffer_t);
- orte_dss.pack(&error_answer, (void*)&command, 1, ORTE_NS_CMD);
- orte_dss.pack(&error_answer, (void*)&rc, 1, ORTE_INT32);
- orte_rml.send_buffer(sender, &error_answer, tag, 0);
- OBJ_DESTRUCT(&error_answer);
-
-CLEANUP:
- /* cleanup */
- OBJ_DESTRUCT(&answer);
-}
-
diff --git a/orte/mca/odls/bproc/odls_bproc.c b/orte/mca/odls/bproc/odls_bproc.c
index 1e3feb40dc..f029b013f2 100644
--- a/orte/mca/odls/bproc/odls_bproc.c
+++ b/orte/mca/odls/bproc/odls_bproc.c
@@ -119,7 +119,7 @@ static char *
return NULL;
}
- rc = orte_ns_base_convert_jobid_to_string(&job, jobid);
+ rc = orte_ns.convert_jobid_to_string(&job, jobid);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return NULL;
@@ -634,7 +634,7 @@ orte_odls_bproc_launch_local_procs(orte_gpr_notify_data_t *data, char **base_env
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
}
- rc = mca_oob_send_packed_nb(ORTE_RML_NAME_SEED, ack, ORTE_RML_TAG_BPROC, 0,
+ rc = mca_oob_send_packed_nb(ORTE_PROC_MY_HNP, ack, ORTE_RML_TAG_BPROC, 0,
odls_bproc_send_cb, NULL);
if (0 > rc) {
ORTE_ERROR_LOG(rc);
@@ -666,7 +666,7 @@ int orte_odls_bproc_kill_local_procs(orte_jobid_t job, bool set_state)
* @param signal The signal to send
* @retval ORTE_SUCCESS
*/
-int orte_odls_bproc_signal_local_procs(orte_process_name_t* proc, int32_t signal)
+int orte_odls_bproc_signal_local_procs(const orte_process_name_t* proc, int32_t signal)
{
orte_iof.iof_flush();
return ORTE_SUCCESS;
diff --git a/orte/mca/odls/default/odls_default_module.c b/orte/mca/odls/default/odls_default_module.c
index 5724e6714b..d6526ea0d1 100644
--- a/orte/mca/odls/default/odls_default_module.c
+++ b/orte/mca/odls/default/odls_default_module.c
@@ -372,6 +372,8 @@ static void odls_default_wait_local_proc(pid_t pid, int status, void* cbdata)
struct stat buf;
int rc;
+ opal_output(orte_odls_globals.output, "odls: child process terminated");
+
/* since we are going to be working with the global list of
* children, we need to protect that list from modification
* by other threads. This will also be used to protect us
@@ -384,6 +386,8 @@ static void odls_default_wait_local_proc(pid_t pid, int status, void* cbdata)
item != opal_list_get_end(&orte_odls_default.children);
item = opal_list_get_next(item)) {
child = (orte_odls_child_t*)item;
+ opal_output(orte_odls_globals.output, "odls: checking child [%ld,%ld,%ld] alive %s",
+ ORTE_NAME_ARGS(child->name), (child->alive ? "true" : "dead"));
if (child->alive && pid == child->pid) { /* found it */
goto GOTCHILD;
}
@@ -398,8 +402,14 @@ static void odls_default_wait_local_proc(pid_t pid, int status, void* cbdata)
return;
GOTCHILD:
+ opal_output(orte_odls_globals.output, "odls: flushing output for [%ld,%ld,%ld]",
+ ORTE_NAME_ARGS(child->name));
+
orte_iof.iof_flush();
+ opal_output(orte_odls_globals.output, "odls: output for [%ld,%ld,%ld] flushed",
+ ORTE_NAME_ARGS(child->name));
+
/* determine the state of this process */
aborted = false;
if(WIFEXITED(status)) {
@@ -426,6 +436,9 @@ GOTCHILD:
job, vpid, "abort", NULL );
free(job);
free(vpid);
+ opal_output(orte_odls_globals.output, "odls: stat'ing file %s for [%ld,%ld,%ld]",
+ abort_file, ORTE_NAME_ARGS(child->name));
+
if (0 == stat(abort_file, &buf)) {
/* the abort file must exist - there is nothing in it we need. It's
* meer existence indicates that an abnormal termination occurred
@@ -435,14 +448,14 @@ GOTCHILD:
aborted = true;
free(abort_file);
} else {
- opal_output(orte_odls_globals.output, "odls: child [%ld,%ld,%ld] died naturally",
+ opal_output(orte_odls_globals.output, "odls: child process [%ld,%ld,%ld] terminated normally",
ORTE_NAME_ARGS(child->name));
}
} else {
/* the process was terminated with a signal! That's definitely
* abnormal, so indicate that condition
*/
- opal_output(orte_odls_globals.output, "odls: child [%ld,%ld,%ld] died by signal",
+ opal_output(orte_odls_globals.output, "odls: child process [%ld,%ld,%ld] terminated with signal",
ORTE_NAME_ARGS(child->name));
aborted = true;
}
diff --git a/orte/mca/oob/base/base.h b/orte/mca/oob/base/base.h
index 2adfcd82b4..8a1ac8468a 100644
--- a/orte/mca/oob/base/base.h
+++ b/orte/mca/oob/base/base.h
@@ -48,13 +48,6 @@ extern "C" {
*/
ORTE_DECLSPEC extern bool orte_oob_base_timing;
-/*
- * Well known address
- */
-
-ORTE_DECLSPEC extern orte_process_name_t mca_oob_name_any;
-ORTE_DECLSPEC extern orte_process_name_t mca_oob_name_seed;
-
/*
* OOB API
*/
@@ -187,7 +180,7 @@ ORTE_DECLSPEC int mca_oob_send_packed(
/**
* Similiar to unix readv(2)
*
-* @param peer (IN/OUT) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive. In the
+* @param peer (IN/OUT) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive. In the
* case of a wildcard receive, will be modified to return the matched peer name.
* @param msg (IN) Array of iovecs describing user buffers and lengths.
* @param count (IN) Number of elements in iovec array.
@@ -223,7 +216,7 @@ ORTE_DECLSPEC int mca_oob_recv(
/**
* Similiar to unix read(2)
*
-* @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+* @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param buf (OUT) Array of iovecs describing user buffers and lengths.
* @param tag (IN/OUT) User defined tag for matching send/recv.
* @return OMPI error code (<0) on error or number of bytes actually received.
@@ -338,7 +331,7 @@ ORTE_DECLSPEC int mca_oob_send_packed_nb(
/**
* Non-blocking version of mca_oob_recv().
*
-* @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+* @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param msg (IN) Array of iovecs describing user buffers and lengths.
* @param count (IN) Number of elements in iovec array.
* @param tag (IN) User defined tag for matching send/recv.
@@ -363,7 +356,7 @@ ORTE_DECLSPEC int mca_oob_recv_nb(
/**
* Routine to cancel pending non-blocking recvs.
*
-* @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+* @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param tag (IN) User defined tag for matching send/recv.
* @return OMPI error code (<0) on error or number of bytes actually received.
*/
@@ -375,7 +368,7 @@ ORTE_DECLSPEC int mca_oob_recv_cancel(
/**
* Non-blocking version of mca_oob_recv_packed().
*
-* @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+* @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param buffer (IN) Array of iovecs describing user buffers and lengths.
* @param count (IN) Number of elements in iovec array.
* @param tag (IN) User defined tag for matching send/recv.
diff --git a/orte/mca/oob/base/oob_base_init.c b/orte/mca/oob/base/oob_base_init.c
index c48db9d3ab..b717d37edb 100644
--- a/orte/mca/oob/base/oob_base_init.c
+++ b/orte/mca/oob/base/oob_base_init.c
@@ -48,9 +48,6 @@ OBJ_CLASS_INSTANCE(
NULL
);
-orte_process_name_t mca_oob_name_seed = { 0, 0, 0 };
-orte_process_name_t mca_oob_name_any = { ORTE_CELLID_MAX, ORTE_JOBID_MAX, ORTE_VPID_MAX };
-
/**
* Parse contact info string into process name and list of uri strings.
*/
diff --git a/orte/mca/oob/base/oob_base_recv.c b/orte/mca/oob/base/oob_base_recv.c
index fa8ace5aa3..6066c34386 100644
--- a/orte/mca/oob/base/oob_base_recv.c
+++ b/orte/mca/oob/base/oob_base_recv.c
@@ -34,7 +34,7 @@
/*
* Similiar to unix recv(2)
*
-* @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+* @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param msg (IN) Array of iovecs describing user buffers and lengths.
* @param types (IN) Parallel array to iovecs describing data type of each iovec element.
* @param count (IN) Number of elements in iovec array.
@@ -51,7 +51,7 @@ int mca_oob_recv(orte_process_name_t* peer, struct iovec *msg, int count, int ta
/*
* Similiar to unix recv(2)
*
-* @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+* @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param buffer (OUT) Buffer that the OOB creates to recv this message...
* @param tag (IN) User defined tag for matching send/recv.
* iovec array without removing the message from the queue.
diff --git a/orte/mca/oob/base/oob_base_recv_nb.c b/orte/mca/oob/base/oob_base_recv_nb.c
index 3fa5456218..b898cfbddf 100644
--- a/orte/mca/oob/base/oob_base_recv_nb.c
+++ b/orte/mca/oob/base/oob_base_recv_nb.c
@@ -53,7 +53,7 @@ static void mca_oob_recv_callback(
/*
* Non-blocking version of mca_oob_recv_nb().
*
- * @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+ * @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param msg (IN) Array of iovecs describing user buffers and lengths.
* @param count (IN) Number of elements in iovec array.
* @param flags (IN) May be MCA_OOB_PEEK to return up to size bytes of msg w/out removing it from the queue,
@@ -71,7 +71,7 @@ int mca_oob_recv_nb(orte_process_name_t* peer, struct iovec* msg, int count, int
/*
* Cancel non-blocking recv.j
*
- * @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+ * @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param tag (IN) User defined tag for message matching.
* @return OMPI success or error code (<0) on error.
*/
@@ -84,7 +84,7 @@ int mca_oob_recv_cancel(orte_process_name_t* peer, int tag)
/**
* Non-blocking version of mca_oob_recv_packed().
*
-* @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+* @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param buffer (IN) Array of iovecs describing user buffers and lengths.
* @param count (IN) Number of elements in iovec array.
* @param tag (IN) User defined tag for matching send/recv.
diff --git a/orte/mca/oob/base/oob_base_xcast.c b/orte/mca/oob/base/oob_base_xcast.c
index a8475b4988..01ec3f5b5f 100644
--- a/orte/mca/oob/base/oob_base_xcast.c
+++ b/orte/mca/oob/base/oob_base_xcast.c
@@ -149,7 +149,7 @@ int mca_oob_xcast(
orte_gpr_notify_message_t *msg;
OBJ_CONSTRUCT(&rbuf, orte_buffer_t);
- rc = mca_oob_recv_packed(MCA_OOB_NAME_ANY, &rbuf, tag);
+ rc = mca_oob_recv_packed(ORTE_NAME_WILDCARD, &rbuf, tag);
if(rc < 0) {
OBJ_DESTRUCT(&rbuf);
return rc;
diff --git a/orte/mca/oob/oob.h b/orte/mca/oob/oob.h
index f3abd671b5..c61ddbe278 100644
--- a/orte/mca/oob/oob.h
+++ b/orte/mca/oob/oob.h
@@ -110,7 +110,7 @@ typedef int (*mca_oob_base_module_send_fn_t)(
/**
* Implementation of mca_oob_recv().
*
-* @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+* @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param msg (IN) Array of iovecs describing user buffers and lengths.
* @param types (IN) Parallel array to iovecs describing data type of each iovec element.
* @param count (IN) Number of elements in iovec array.
@@ -153,7 +153,7 @@ typedef int (*mca_oob_base_module_send_nb_fn_t)(
/**
* Implementation of mca_oob_recv_nb().
*
-* @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+* @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param msg (IN) Array of iovecs describing user buffers and lengths.
* @param count (IN) Number of elements in iovec array.
* @param tag (IN) User defined tag for matching send/recv.
@@ -175,7 +175,7 @@ typedef int (*mca_oob_base_module_recv_nb_fn_t)(
/**
* Implementation of mca_oob_recv_cancel().
*
-* @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+* @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param tag (IN) User defined tag for matching send/recv.
* @return OMPI error code (<0) on error or number of bytes actually received.
*/
diff --git a/orte/mca/oob/oob_types.h b/orte/mca/oob/oob_types.h
index f0e6d85e76..ae1c62135b 100644
--- a/orte/mca/oob/oob_types.h
+++ b/orte/mca/oob/oob_types.h
@@ -25,24 +25,5 @@
#include "orte_config.h"
#include "orte/orte_constants.h"
-#include <limits.h>
-
-#include "orte/mca/rml/rml_types.h"
-/*
- * Other constants
- */
-
-/**
- * The wildcard for receives from any peer.
- */
-#define MCA_OOB_NAME_ANY &mca_oob_name_any
-/**
- * Process name of self
- */
-#define MCA_OOB_NAME_SELF orte_process_info.my_name
-/**
- * Process name of seed
- */
-#define MCA_OOB_NAME_SEED &mca_oob_name_seed
#endif /* MCA_OOB_TYPES_H */
diff --git a/orte/mca/oob/tcp/oob_tcp.c b/orte/mca/oob/tcp/oob_tcp.c
index f65bde51f9..4f9e09433a 100644
--- a/orte/mca/oob/tcp/oob_tcp.c
+++ b/orte/mca/oob/tcp/oob_tcp.c
@@ -690,20 +690,18 @@ static void mca_oob_tcp_recv_connect(int sd, mca_oob_tcp_hdr_t* hdr)
}
}
- /* check for wildcard name - if this is true - we allocate a name from the name server
+ /* check for invalid name - if this is true - we allocate a name from the name server
* and return to the peer
*/
- cmpval = orte_ns.compare(ORTE_NS_CMP_ALL, &hdr->msg_src, MCA_OOB_NAME_ANY);
- if (cmpval == 0) {
- if (ORTE_SUCCESS != orte_ns.create_jobid(&hdr->msg_src.jobid)) {
+ cmpval = orte_ns.compare_fields(ORTE_NS_CMP_ALL, &hdr->msg_src, ORTE_NAME_INVALID);
+ if (cmpval == ORTE_EQUAL) {
+ if (ORTE_SUCCESS != orte_ns.create_jobid(&hdr->msg_src.jobid, NULL)) {
return;
}
if (ORTE_SUCCESS != orte_ns.reserve_range(hdr->msg_src.jobid, 1, &hdr->msg_src.vpid)) {
return;
}
- if (ORTE_SUCCESS != orte_ns.assign_cellid_to_process(&hdr->msg_src)) {
- return;
- }
+ hdr->msg_src.cellid = ORTE_PROC_MY_NAME->cellid;
}
/* lookup the corresponding process */
@@ -1049,12 +1047,8 @@ int mca_oob_tcp_init(void)
#endif
/* get my jobid */
- if (ORTE_SUCCESS != (rc = orte_ns.get_jobid(&jobid,
- orte_process_info.my_name))) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
-
+ jobid = ORTE_PROC_MY_NAME->jobid;
+
/* create a listen socket */
if (OOB_TCP_EVENT == mca_oob_tcp_component.tcp_listen_type) {
if(mca_oob_tcp_create_listen() != ORTE_SUCCESS) {
@@ -1286,12 +1280,17 @@ int mca_oob_tcp_fini(void)
* Note that the definition of < or > is somewhat arbitrary -
* just needs to be consistently applied to maintain an ordering
* when process names are used as indices.
+*
+* Currently, this function is ONLY used in one place - in oob_tcp_send.c to
+* determine if the recipient of the message-to-be-sent is ourselves. Hence,
+* this comparison is okay to be LITERAL and can/should use the ns.compare_fields
+* function
*/
int mca_oob_tcp_process_name_compare(const orte_process_name_t* n1, const orte_process_name_t* n2)
{
- return orte_ns.compare(ORTE_NS_CMP_ALL, n1, n2);
+ return orte_ns.compare_fields(ORTE_NS_CMP_ALL, n1, n2);
}
diff --git a/orte/mca/oob/tcp/oob_tcp.h b/orte/mca/oob/tcp/oob_tcp.h
index 7900f982df..add933b1ad 100644
--- a/orte/mca/oob/tcp/oob_tcp.h
+++ b/orte/mca/oob/tcp/oob_tcp.h
@@ -119,7 +119,7 @@ int mca_oob_tcp_send(
/**
* Similiar to unix readv(2)
*
- * @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+ * @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param msg (IN) Array of iovecs describing user buffers and lengths.
* @param count (IN) Number of elements in iovec array.
* @param tag (IN) User defined tag for matching send/recv.
@@ -166,7 +166,7 @@ int mca_oob_tcp_send_nb(
/**
* Non-blocking version of mca_oob_recv().
*
- * @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+ * @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param msg (IN) Array of iovecs describing user buffers and lengths.
* @param count (IN) Number of elements in iovec array.
* @param tag (IN) User defined tag for matching send/recv.
@@ -188,7 +188,7 @@ int mca_oob_tcp_recv_nb(
/**
* Cancel non-blocking receive.
*
- * @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+ * @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param tag (IN) User defined tag for matching send/recv.
* @return OMPI error code (<0) on error or number of bytes actually received.
*/
diff --git a/orte/mca/oob/tcp/oob_tcp_addr.c b/orte/mca/oob/tcp/oob_tcp_addr.c
index 6857bea799..8a2061f609 100644
--- a/orte/mca/oob/tcp/oob_tcp_addr.c
+++ b/orte/mca/oob/tcp/oob_tcp_addr.c
@@ -32,6 +32,11 @@
#include <string.h>
#include "orte/orte_constants.h"
#include "opal/util/if.h"
+
+#include "orte/mca/ns/ns_types.h"
+#include "orte/util/proc_info.h"
+#include "orte/dss/dss.h"
+
#include "oob_tcp.h"
#include "oob_tcp_addr.h"
@@ -65,7 +70,7 @@ int mca_oob_tcp_addr_pack(orte_buffer_t* buffer)
int i;
int rc;
- rc = orte_dss.pack(buffer, orte_process_info.my_name, 1, ORTE_NAME);
+ rc = orte_dss.pack(buffer, ORTE_PROC_MY_NAME, 1, ORTE_NAME);
if(rc != ORTE_SUCCESS)
return rc;
diff --git a/orte/mca/oob/tcp/oob_tcp_msg.c b/orte/mca/oob/tcp/oob_tcp_msg.c
index c02d403e68..62352c8eea 100644
--- a/orte/mca/oob/tcp/oob_tcp_msg.c
+++ b/orte/mca/oob/tcp/oob_tcp_msg.c
@@ -404,7 +404,9 @@ void mca_oob_tcp_msg_recv_complete(mca_oob_tcp_msg_t* msg, mca_oob_tcp_peer_t* p
}
/**
- * Process an ident message.
+ * Process an ident message. In this case, we insist that the two process names
+ * exactly match - hence, we use the orte_ns.compare_fields function, which
+ * checks each field in a literal manner (i.e., no wildcards).
*/
static void mca_oob_tcp_msg_ident(mca_oob_tcp_msg_t* msg, mca_oob_tcp_peer_t* peer)
@@ -412,7 +414,7 @@ static void mca_oob_tcp_msg_ident(mca_oob_tcp_msg_t* msg, mca_oob_tcp_peer_t* pe
orte_process_name_t src = msg->msg_hdr.msg_src;
OPAL_THREAD_LOCK(&mca_oob_tcp_component.tcp_lock);
- if (orte_ns.compare(ORTE_NS_CMP_ALL, &peer->peer_name, &src) != 0) {
+ if (orte_ns.compare_fields(ORTE_NS_CMP_ALL, &peer->peer_name, &src) != ORTE_EQUAL) {
orte_hash_table_remove_proc(&mca_oob_tcp_component.tcp_peers, &peer->peer_name);
peer->peer_name = src;
orte_hash_table_set_proc(&mca_oob_tcp_component.tcp_peers, &peer->peer_name, peer);
@@ -558,9 +560,7 @@ mca_oob_tcp_msg_t* mca_oob_tcp_msg_match_recv(orte_process_name_t* name, int tag
msg != (mca_oob_tcp_msg_t*) opal_list_get_end(&mca_oob_tcp_component.tcp_msg_recv);
msg = (mca_oob_tcp_msg_t*) opal_list_get_next(msg)) {
- int cmpval1 = orte_ns.compare(ORTE_NS_CMP_ALL, name, MCA_OOB_NAME_ANY);
- int cmpval2 = orte_ns.compare(ORTE_NS_CMP_ALL, name, &msg->msg_peer);
- if((0 == cmpval1) || (0 == cmpval2)) {
+ if(ORTE_EQUAL == orte_dss.compare(name, &msg->msg_peer, ORTE_NAME)) {
if (tag == msg->msg_hdr.msg_tag) {
return msg;
}
@@ -585,10 +585,7 @@ mca_oob_tcp_msg_t* mca_oob_tcp_msg_match_post(orte_process_name_t* name, int tag
msg != (mca_oob_tcp_msg_t*) opal_list_get_end(&mca_oob_tcp_component.tcp_msg_post);
msg = (mca_oob_tcp_msg_t*) opal_list_get_next(msg)) {
- int cmpval1 = orte_ns.compare(ORTE_NS_CMP_ALL, &msg->msg_peer, MCA_OOB_NAME_ANY);
- int cmpval2 = orte_ns.compare(ORTE_NS_CMP_ALL, name, &msg->msg_peer);
-
- if((0 == cmpval1) || (0 == cmpval2)) {
+ if(ORTE_EQUAL == orte_dss.compare(name, &msg->msg_peer, ORTE_NAME)) {
if (msg->msg_hdr.msg_tag == tag) {
if((msg->msg_flags & MCA_OOB_PERSISTENT) == 0) {
opal_list_remove_item(&mca_oob_tcp_component.tcp_msg_post, &msg->super.super);
diff --git a/orte/mca/oob/tcp/oob_tcp_peer.c b/orte/mca/oob/tcp/oob_tcp_peer.c
index 0fe1be9f73..34e88ff62e 100644
--- a/orte/mca/oob/tcp/oob_tcp_peer.c
+++ b/orte/mca/oob/tcp/oob_tcp_peer.c
@@ -503,7 +503,7 @@ void mca_oob_tcp_peer_close(mca_oob_tcp_peer_t* peer)
}
/* if we lose the connection to the seed - abort */
- if(memcmp(&peer->peer_name,&mca_oob_name_seed,sizeof(mca_oob_name_seed)) == 0) {
+ if(memcmp(&peer->peer_name,ORTE_PROC_MY_HNP,sizeof(orte_process_name_t)) == 0) {
/* If we are not already inside orte_finalize, then call abort */
if (ORTE_UNIVERSE_STATE_FINALIZE > orte_universe_info.state) {
/* Should free the peer lock before we abort so we don't
@@ -554,7 +554,7 @@ static int mca_oob_tcp_peer_send_connect_ack(mca_oob_tcp_peer_t* peer)
mca_oob_tcp_hdr_t hdr;
memset(&hdr,0,sizeof(hdr));
if (NULL == orte_process_info.my_name) { /* my name isn't defined yet */
- hdr.msg_src = *MCA_OOB_NAME_ANY;
+ hdr.msg_src = *ORTE_NAME_INVALID;
} else {
hdr.msg_src = *(orte_process_info.my_name);
}
@@ -597,11 +597,13 @@ static int mca_oob_tcp_peer_recv_connect_ack(mca_oob_tcp_peer_t* peer)
return ORTE_ERR_UNREACH;
}
- /* if we have a wildcard name - use the name returned by the peer */
+ /* if we have an invalid name or do not have one assigned at all - use the name returned by the peer.
+ * This needs to be a LITERAL comparison - we do NOT want wildcard values to return EQUAL
+ */
if(orte_process_info.my_name == NULL) {
orte_ns.create_process_name(&orte_process_info.my_name,
hdr.msg_dst.cellid, hdr.msg_dst.jobid, hdr.msg_dst.vpid);
- } else if(orte_ns.compare(ORTE_NS_CMP_ALL, orte_process_info.my_name, &mca_oob_name_any) == 0) {
+ } else if (orte_ns.compare_fields(ORTE_NS_CMP_ALL, orte_process_info.my_name, ORTE_NAME_INVALID) == ORTE_EQUAL) {
*orte_process_info.my_name = hdr.msg_dst;
}
@@ -876,18 +878,24 @@ static void mca_oob_tcp_peer_dump(mca_oob_tcp_peer_t* peer, const char* msg)
/*
- * Accept incoming connection - if not already connected.
+ * Accept incoming connection - if not already connected. We compare the name of the
+ * peer to our own name using the ns.compare_fields function as we want this to be
+ * a LITERAL comparison - i.e., there is no occasion when the peer's name should
+ * be a wildcard value.
+ *
+ * To avoid competing reciprocal connection attempts, we only accept connections from
+ * processes whose names are "greater" than our own.
*/
bool mca_oob_tcp_peer_accept(mca_oob_tcp_peer_t* peer, int sd)
{
int cmpval;
OPAL_THREAD_LOCK(&peer->peer_lock);
- cmpval = orte_ns.compare(ORTE_NS_CMP_ALL, &peer->peer_name, orte_process_info.my_name);
+ cmpval = orte_ns.compare_fields(ORTE_NS_CMP_ALL, &peer->peer_name, orte_process_info.my_name);
if ((peer->peer_state == MCA_OOB_TCP_CLOSED) ||
(peer->peer_state == MCA_OOB_TCP_RESOLVE) ||
(peer->peer_state != MCA_OOB_TCP_CONNECTED &&
- cmpval < 0)) {
+ cmpval == ORTE_VALUE1_GREATER)) {
if(peer->peer_state != MCA_OOB_TCP_CLOSED) {
mca_oob_tcp_peer_close(peer);
diff --git a/orte/mca/oob/tcp/oob_tcp_ping.c b/orte/mca/oob/tcp/oob_tcp_ping.c
index 6a89a8d4a4..48a79f1bfa 100644
--- a/orte/mca/oob/tcp/oob_tcp_ping.c
+++ b/orte/mca/oob/tcp/oob_tcp_ping.c
@@ -47,7 +47,10 @@
#include <signal.h>
#endif
#include "opal/event/event.h"
+
#include "orte/mca/ns/ns_types.h"
+#include "orte/util/proc_info.h"
+
#include "orte/mca/oob/tcp/oob_tcp.h"
/*
@@ -148,7 +151,7 @@ int mca_oob_tcp_ping(
if(orte_process_info.my_name != NULL) {
hdr.msg_src = *orte_process_info.my_name;
} else {
- hdr.msg_src = mca_oob_name_any;
+ hdr.msg_src = *ORTE_NAME_INVALID;
}
hdr.msg_dst = *name;
hdr.msg_type = MCA_OOB_TCP_PROBE;
diff --git a/orte/mca/oob/tcp/oob_tcp_recv.c b/orte/mca/oob/tcp/oob_tcp_recv.c
index 382dd76c37..7d6c31e281 100644
--- a/orte/mca/oob/tcp/oob_tcp_recv.c
+++ b/orte/mca/oob/tcp/oob_tcp_recv.c
@@ -24,7 +24,7 @@
/*
* Similiar to unix readv(2)
*
- * @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+ * @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param msg (IN) Array of iovecs describing user buffers and lengths.
* @param types (IN) Parallel array to iovecs describing data type of each iovec element.
* @param count (IN) Number of elements in iovec array.
@@ -114,7 +114,7 @@ int mca_oob_tcp_recv(
msg->msg_hdr.msg_type = MCA_OOB_TCP_DATA;
msg->msg_hdr.msg_src = *peer;
if (NULL == orte_process_info.my_name) {
- msg->msg_hdr.msg_dst = *MCA_OOB_NAME_ANY;
+ msg->msg_hdr.msg_dst = *ORTE_NAME_INVALID;
} else {
msg->msg_hdr.msg_dst = *orte_process_info.my_name;
}
@@ -202,7 +202,7 @@ static void mca_oob_tcp_msg_matched(mca_oob_tcp_msg_t* msg, mca_oob_tcp_msg_t* m
/*
* Non-blocking version of mca_oob_recv().
*
- * @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+ * @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param msg (IN) Array of iovecs describing user buffers and lengths.
* @param count (IN) Number of elements in iovec array.
* @param tag (IN) User supplied tag for matching send/recv.
@@ -242,7 +242,11 @@ int mca_oob_tcp_recv_nb(
}
/* fill in the header */
- msg->msg_hdr.msg_src = *orte_process_info.my_name;
+ if (NULL == orte_process_info.my_name) {
+ msg->msg_hdr.msg_src = *ORTE_NAME_INVALID;
+ } else {
+ msg->msg_hdr.msg_src = *orte_process_info.my_name;
+ }
msg->msg_hdr.msg_dst = *peer;
msg->msg_hdr.msg_size = size;
msg->msg_hdr.msg_tag = tag;
@@ -286,7 +290,7 @@ int mca_oob_tcp_recv_nb(
/*
* Cancel non-blocking recv.
*
- * @param peer (IN) Opaque name of peer process or MCA_OOB_NAME_ANY for wildcard receive.
+ * @param peer (IN) Opaque name of peer process or ORTE_NAME_WILDCARD for wildcard receive.
* @param tag (IN) User supplied tag for matching send/recv.
* @return OMPI error code (<0) on error or number of bytes actually received.
*/
@@ -295,7 +299,7 @@ int mca_oob_tcp_recv_cancel(
orte_process_name_t* name,
int tag)
{
- int matched = 0, cmpval1, cmpval2;
+ int matched = 0;
opal_list_item_t *item, *next;
/* wait for any previously matched messages to be processed */
@@ -317,9 +321,7 @@ int mca_oob_tcp_recv_cancel(
mca_oob_tcp_msg_t* msg = (mca_oob_tcp_msg_t*)item;
next = opal_list_get_next(item);
- cmpval1 = orte_ns.compare(ORTE_NS_CMP_ALL, name, MCA_OOB_NAME_ANY);
- cmpval2 = orte_ns.compare(ORTE_NS_CMP_ALL, &msg->msg_peer, name);
- if ((0 == cmpval1) || (0 == cmpval2)) {
+ if (ORTE_EQUAL == orte_dss.compare(name, &msg->msg_peer, ORTE_NAME)) {
if (msg->msg_hdr.msg_tag == tag) {
opal_list_remove_item(&mca_oob_tcp_component.tcp_msg_post, &msg->super.super);
MCA_OOB_TCP_MSG_RETURN(msg);
diff --git a/orte/mca/oob/tcp/oob_tcp_send.c b/orte/mca/oob/tcp/oob_tcp_send.c
index 0b57f9c93e..6e11513668 100644
--- a/orte/mca/oob/tcp/oob_tcp_send.c
+++ b/orte/mca/oob/tcp/oob_tcp_send.c
@@ -18,6 +18,7 @@
#include "orte_config.h"
#include "orte/mca/ns/ns_types.h"
+#include "orte/util/proc_info.h"
#include "orte/mca/oob/tcp/oob_tcp.h"
@@ -119,7 +120,7 @@ int mca_oob_tcp_send(
msg->msg_hdr.msg_size = size;
msg->msg_hdr.msg_tag = tag;
if (NULL == orte_process_info.my_name) {
- msg->msg_hdr.msg_src = *MCA_OOB_NAME_ANY;
+ msg->msg_hdr.msg_src = *ORTE_NAME_INVALID;
} else {
msg->msg_hdr.msg_src = *orte_process_info.my_name;
}
@@ -144,7 +145,7 @@ int mca_oob_tcp_send(
msg->msg_peer = peer->peer_name;
if (NULL != name && NULL != orte_process_info.my_name &&
- 0 == mca_oob_tcp_process_name_compare(name, orte_process_info.my_name)) { /* local delivery */
+ ORTE_EQUAL == mca_oob_tcp_process_name_compare(name, orte_process_info.my_name)) { /* local delivery */
return mca_oob_tcp_send_self(peer,msg,iov,count);
}
@@ -206,7 +207,11 @@ int mca_oob_tcp_send_nb(
msg->msg_hdr.msg_type = MCA_OOB_TCP_DATA;
msg->msg_hdr.msg_size = size;
msg->msg_hdr.msg_tag = tag;
- msg->msg_hdr.msg_src = *orte_process_info.my_name;
+ if (NULL == orte_process_info.my_name) {
+ msg->msg_hdr.msg_src = *ORTE_NAME_INVALID;
+ } else {
+ msg->msg_hdr.msg_src = *orte_process_info.my_name;
+ }
msg->msg_hdr.msg_dst = *name;
/* create one additional iovect that will hold the size of the message */
@@ -227,7 +232,7 @@ int mca_oob_tcp_send_nb(
msg->msg_complete = false;
msg->msg_peer = peer->peer_name;
- if (0 == mca_oob_tcp_process_name_compare(name, orte_process_info.my_name)) { /* local delivery */
+ if (ORTE_EQUAL == mca_oob_tcp_process_name_compare(name, orte_process_info.my_name)) { /* local delivery */
return mca_oob_tcp_send_self(peer,msg,iov,count);
}
diff --git a/orte/mca/pls/base/pls_base_dmn_registry_fns.c b/orte/mca/pls/base/pls_base_dmn_registry_fns.c
index 44c75a5150..829ccd9610 100644
--- a/orte/mca/pls/base/pls_base_dmn_registry_fns.c
+++ b/orte/mca/pls/base/pls_base_dmn_registry_fns.c
@@ -28,6 +28,7 @@
#include "orte/mca/ns/ns.h"
#include "orte/mca/gpr/gpr.h"
#include "orte/mca/errmgr/errmgr.h"
+#include "orte/mca/rmgr/rmgr.h"
#include "orte/mca/pls/base/pls_private.h"
@@ -128,10 +129,7 @@ CLEANUP:
return rc;
}
-/*
- * Retrieve a list of the active daemons for a job
- */
-int orte_pls_base_get_active_daemons(opal_list_t *daemons, orte_jobid_t job)
+static int get_daemons(opal_list_t *daemons, orte_jobid_t job)
{
orte_gpr_value_t **values;
orte_gpr_keyval_t *kv;
@@ -149,7 +147,7 @@ int orte_pls_base_get_active_daemons(opal_list_t *daemons, orte_jobid_t job)
orte_pls_daemon_info_t *dmn;
bool found_name, found_node, found_cell;
int rc;
-
+
/* setup the key */
if (ORTE_SUCCESS != (rc = orte_ns.convert_jobid_to_string(&jobid_string, job))) {
ORTE_ERROR_LOG(rc);
@@ -172,11 +170,11 @@ int orte_pls_base_get_active_daemons(opal_list_t *daemons, orte_jobid_t job)
/* loop through the answers and construct the list */
for (i=0; i < cnt; i++) {
/* for systems such as bproc, the node segment holds containers
- * for nodes that we may not have launched upon. Each container
- * will send us back a value object, so we have to ensure here
- * that we only create daemon objects on the list for those nodes
- * that DO provide a valid object
- */
+ * for nodes that we may not have launched upon. Each container
+ * will send us back a value object, so we have to ensure here
+ * that we only create daemon objects on the list for those nodes
+ * that DO provide a valid object
+ */
found_name = found_node = found_cell = false;
for (j=0; j < values[i]->cnt; j++) {
kv = values[i]->keyvals[j];
@@ -204,10 +202,10 @@ int orte_pls_base_get_active_daemons(opal_list_t *daemons, orte_jobid_t job)
found_cell = true;
continue;
}
- }
+ }
/* if we found everything, then this is a valid entry - create
- * it and add it to the list
- */
+ * it and add it to the list
+ */
if (found_name && found_node && found_cell) {
dmn = OBJ_NEW(orte_pls_daemon_info_t);
if (NULL == dmn) {
@@ -230,14 +228,59 @@ int orte_pls_base_get_active_daemons(opal_list_t *daemons, orte_jobid_t job)
}
OBJ_RELEASE(values[i]);
}
-
+
CLEANUP:
for (i=0; i < cnt; i++) {
if (NULL != values[i]) OBJ_RELEASE(values[i]);
}
if (NULL != values) free(values);
free(keys[0]);
+
+ return rc;
+}
+/*
+ * Retrieve a list of the active daemons for a job
+ */
+int orte_pls_base_get_active_daemons(opal_list_t *daemons, orte_jobid_t job, opal_list_t *attrs)
+{
+ orte_jobid_t *jobs;
+ orte_std_cntr_t njobs, i;
+ bool allocated;
+ int rc;
+
+ if (NULL != orte_rmgr.find_attribute(attrs, ORTE_NS_INCLUDE_DESCENDANTS)) {
+ /* need to include all descendants in list */
+ if (ORTE_SUCCESS != (rc = orte_ns.get_job_descendants(&jobs, &njobs, job))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ allocated = true;
+ } else if (NULL != orte_rmgr.find_attribute(attrs, ORTE_NS_INCLUDE_CHILDREN)) {
+ /* just include the direct children of the job */
+ if (ORTE_SUCCESS != (rc = orte_ns.get_job_children(&jobs, &njobs, job))) {
+ ORTE_ERROR_LOG(rc);
+ return rc;
+ }
+ allocated = true;
+ } else {
+ /* just want daemons for this one job */
+ jobs = &job;
+ njobs = 1;
+ allocated = false;
+ }
+
+ /* loop through all the jobs and get their info */
+ for (i=0; i < njobs; i++) {
+ if (ORTE_SUCCESS != (rc = get_daemons(daemons, jobs[i]))) {
+ ORTE_ERROR_LOG(rc);
+ goto CLEANUP;
+ }
+ }
+
+CLEANUP:
+ if (allocated) free(jobs);
+
return rc;
}
@@ -246,19 +289,9 @@ CLEANUP:
*/
int orte_pls_base_remove_daemon(orte_pls_daemon_info_t *info)
{
- opal_list_t daemons;
- int rc;
-
- OBJ_CONSTRUCT(&daemons, opal_list_t);
-
- /* We actually don't want to do this - instead, we need to do a registry
- * delete function call targeting this entry
+ /* We need to do a registry
+ * delete function call targeting the entry
*/
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, info->active_job))) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
-
- /* find this item in the list */
+
return ORTE_SUCCESS;
}
diff --git a/orte/mca/pls/base/pls_base_receive.c b/orte/mca/pls/base/pls_base_receive.c
index 29cbb7d07d..810233564b 100644
--- a/orte/mca/pls/base/pls_base_receive.c
+++ b/orte/mca/pls/base/pls_base_receive.c
@@ -51,7 +51,7 @@ int orte_pls_base_comm_start(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY,
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_PLS,
ORTE_RML_PERSISTENT,
orte_pls_base_recv,
@@ -72,7 +72,7 @@ int orte_pls_base_comm_stop(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_PLS))) {
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_PLS))) {
ORTE_ERROR_LOG(rc);
}
recv_issued = false;
@@ -97,6 +97,8 @@ void orte_pls_base_recv(int status, orte_process_name_t* sender,
orte_jobid_t job;
orte_process_name_t *name;
int32_t signal;
+ opal_list_t attrs;
+ opal_list_item_t *item;
int rc;
count = 1;
@@ -130,10 +132,21 @@ void orte_pls_base_recv(int status, orte_process_name_t* sender,
ORTE_ERROR_LOG(rc);
goto SEND_ANSWER;
}
+
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &attrs, &count, ORTE_ATTR_LIST))) {
+ ORTE_ERROR_LOG(rc);
+ goto SEND_ANSWER;
+ }
- if (ORTE_SUCCESS != (rc = orte_pls.terminate_job(job))) {
+
+ if (ORTE_SUCCESS != (rc = orte_pls.terminate_job(job, &attrs))) {
ORTE_ERROR_LOG(rc);
}
+
+ while (NULL != (item = opal_list_remove_first(&attrs))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attrs);
break;
case ORTE_PLS_TERMINATE_ORTEDS_CMD:
@@ -143,9 +156,19 @@ void orte_pls_base_recv(int status, orte_process_name_t* sender,
goto SEND_ANSWER;
}
- if (ORTE_SUCCESS != (rc = orte_pls.terminate_orteds(job))) {
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &attrs, &count, ORTE_ATTR_LIST))) {
+ ORTE_ERROR_LOG(rc);
+ goto SEND_ANSWER;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_pls.terminate_orteds(job, &attrs))) {
ORTE_ERROR_LOG(rc);
}
+
+ while (NULL != (item = opal_list_remove_first(&attrs))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attrs);
break;
case ORTE_PLS_SIGNAL_JOB_CMD:
@@ -161,9 +184,19 @@ void orte_pls_base_recv(int status, orte_process_name_t* sender,
goto SEND_ANSWER;
}
- if (ORTE_SUCCESS != (rc = orte_pls.signal_job(job, signal))) {
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ count = 1;
+ if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &attrs, &count, ORTE_ATTR_LIST))) {
+ ORTE_ERROR_LOG(rc);
+ goto SEND_ANSWER;
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_pls.signal_job(job, signal, &attrs))) {
ORTE_ERROR_LOG(rc);
}
+
+ while (NULL != (item = opal_list_remove_first(&attrs))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attrs);
break;
case ORTE_PLS_TERMINATE_PROC_CMD:
diff --git a/orte/mca/pls/base/pls_private.h b/orte/mca/pls/base/pls_private.h
index f2c971ae55..34b0b3ec96 100644
--- a/orte/mca/pls/base/pls_private.h
+++ b/orte/mca/pls/base/pls_private.h
@@ -79,7 +79,7 @@ extern "C" {
int orte_pls_base_orted_signal_local_procs(opal_list_t *daemons, int32_t signal);
int orte_pls_base_orted_add_local_procs(opal_list_t *daemons, orte_gpr_notify_data_t *ndat);
- int orte_pls_base_get_active_daemons(opal_list_t *daemons, orte_jobid_t job);
+ int orte_pls_base_get_active_daemons(opal_list_t *daemons, orte_jobid_t job, opal_list_t *attrs);
int orte_pls_base_store_active_daemons(opal_list_t *daemons);
int orte_pls_base_remove_daemon(orte_pls_daemon_info_t *info);
diff --git a/orte/mca/pls/bproc/pls_bproc.c b/orte/mca/pls/bproc/pls_bproc.c
index 9fa58e65cf..5953169d04 100644
--- a/orte/mca/pls/bproc/pls_bproc.c
+++ b/orte/mca/pls/bproc/pls_bproc.c
@@ -42,6 +42,7 @@
#endif /* HAVE_STRING_H */
#include "opal/install_dirs.h"
+#include "opal/class/opal_list.h"
#include "opal/event/event.h"
#include "opal/mca/base/mca_base_param.h"
#include "opal/util/argv.h"
@@ -342,7 +343,7 @@ static void orte_pls_bproc_waitpid_daemon_cb(pid_t wpid, int status, void *data)
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
}
- rc = mca_oob_send_packed(ORTE_RML_NAME_SELF, &ack, ORTE_RML_TAG_BPROC, 0);
+ rc = mca_oob_send_packed(ORTE_PROC_MY_NAME, &ack, ORTE_RML_TAG_BPROC, 0);
if(0 > rc) {
ORTE_ERROR_LOG(rc);
}
@@ -429,8 +430,7 @@ static void orte_pls_bproc_setup_env(char *** env)
/* ns replica contact info */
if(NULL == orte_process_info.ns_replica) {
- orte_ns.copy_process_name(&orte_process_info.ns_replica,
- orte_process_info.my_name);
+ orte_dss.copy((void**)&orte_process_info.ns_replica, orte_process_info.my_name, ORTE_NAME);
orte_process_info.ns_replica_uri = orte_rml.get_uri();
}
var = mca_base_param_environ_variable("ns","replica","uri");
@@ -451,8 +451,7 @@ static void orte_pls_bproc_setup_env(char *** env)
/* gpr replica contact info */
if(NULL == orte_process_info.gpr_replica) {
- orte_ns.copy_process_name(&orte_process_info.gpr_replica,
- orte_process_info.my_name);
+ orte_dss.copy((void**)&orte_process_info.gpr_replica, orte_process_info.my_name, ORTE_NAME);
orte_process_info.gpr_replica_uri = orte_rml.get_uri();
}
var = mca_base_param_environ_variable("gpr","replica","uri");
@@ -832,13 +831,13 @@ orte_pls_bproc_check_node_state(orte_gpr_notify_data_t *notify_data,
orte_schema.extract_jobid_from_segment_name(&jobid, value->tokens[0]);
printf("killing jobid %d\n", jobid);
if(jobid != 0)
- orte_pls_bproc_terminate_job(jobid);
+ orte_pls_bproc_terminate_job(jobid, NULL);
}
/*
* and kill everyone else
*/
printf("and go bye-bye...\n");
- orte_pls_bproc_terminate_job(0);
+ orte_pls_bproc_terminate_job(0, NULL);
/* shouldn't ever get here.. */
exit(1);
}
@@ -1240,7 +1239,7 @@ int orte_pls_bproc_launch(orte_jobid_t jobid) {
for(j = 0; j < num_daemons; j++) {
orte_buffer_t ack;
OBJ_CONSTRUCT(&ack, orte_buffer_t);
- rc = mca_oob_recv_packed(ORTE_RML_NAME_ANY, &ack, ORTE_RML_TAG_BPROC);
+ rc = mca_oob_recv_packed(ORTE_NAME_WILDCARD, &ack, ORTE_RML_TAG_BPROC);
if(0 > rc) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&ack);
@@ -1265,7 +1264,7 @@ int orte_pls_bproc_launch(orte_jobid_t jobid) {
}
rc = ORTE_ERROR;
ORTE_ERROR_LOG(rc);
- orte_pls_bproc_terminate_job(jobid);
+ orte_pls_bproc_terminate_job(jobid, NULL);
goto cleanup;
}
}
@@ -1307,7 +1306,7 @@ cleanup:
/**
* Terminate all processes associated with this job */
-int orte_pls_bproc_terminate_job(orte_jobid_t jobid) {
+int orte_pls_bproc_terminate_job(orte_jobid_t jobid, opal_list_t *attrs) {
pid_t* pids;
orte_std_cntr_t i, num_pids;
int rc;
@@ -1319,7 +1318,7 @@ int orte_pls_bproc_terminate_job(orte_jobid_t jobid) {
}
/* kill application process */
- if(ORTE_SUCCESS != (rc = orte_pls_bproc_get_proc_pids(jobid, &pids, &num_pids)))
+ if(ORTE_SUCCESS != (rc = orte_pls_bproc_get_proc_pids(jobid, &pids, &num_pids, attrs)))
return rc;
for(i=0; i<num_pids; i++) {
if(mca_pls_bproc_component.debug) {
@@ -1337,7 +1336,7 @@ int orte_pls_bproc_terminate_job(orte_jobid_t jobid) {
/**
* Terminate the orteds for a given job
*/
-int orte_pls_bproc_terminate_orteds(orte_jobid_t jobid)
+int orte_pls_bproc_terminate_orteds(orte_jobid_t jobid, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -1347,7 +1346,7 @@ int orte_pls_bproc_terminate_orteds(orte_jobid_t jobid)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
@@ -1394,7 +1393,7 @@ int orte_pls_bproc_terminate_proc(const orte_process_name_t* proc_name) {
/**
* Signal all processes associated with this job
*/
-int orte_pls_bproc_signal_job(orte_jobid_t jobid, int32_t signal) {
+int orte_pls_bproc_signal_job(orte_jobid_t jobid, int32_t signal, opal_list_t *attrs) {
pid_t* pids;
orte_std_cntr_t i, num_pids;
int rc;
@@ -1402,7 +1401,7 @@ int orte_pls_bproc_signal_job(orte_jobid_t jobid, int32_t signal) {
OPAL_TRACE(1);
/* signal application process */
- if(ORTE_SUCCESS != (rc = orte_pls_bproc_get_proc_pids(jobid, &pids, &num_pids)))
+ if(ORTE_SUCCESS != (rc = orte_pls_bproc_get_proc_pids(jobid, &pids, &num_pids, attrs)))
return rc;
for(i=0; i<num_pids; i++) {
if(mca_pls_bproc_component.debug) {
diff --git a/orte/mca/pls/bproc/pls_bproc.h b/orte/mca/pls/bproc/pls_bproc.h
index 001b636656..55e7635f8d 100644
--- a/orte/mca/pls/bproc/pls_bproc.h
+++ b/orte/mca/pls/bproc/pls_bproc.h
@@ -72,10 +72,10 @@ int orte_pls_bproc_finalize(void);
* Interface
*/
int orte_pls_bproc_launch(orte_jobid_t);
-int orte_pls_bproc_terminate_job(orte_jobid_t);
+int orte_pls_bproc_terminate_job(orte_jobid_t, opal_list_t*);
int orte_pls_bproc_terminate_proc(const orte_process_name_t* proc_name);
-int orte_pls_bproc_terminate_orteds(orte_jobid_t jobid);
-int orte_pls_bproc_signal_job(orte_jobid_t, int32_t);
+int orte_pls_bproc_terminate_orteds(orte_jobid_t jobid, opal_list_t*);
+int orte_pls_bproc_signal_job(orte_jobid_t, int32_t, opal_list_t*);
int orte_pls_bproc_signal_proc(const orte_process_name_t* proc_name, int32_t);
/* Utility routine to get/set process pid */
@@ -84,7 +84,10 @@ ORTE_DECLSPEC int orte_pls_bproc_get_proc_pid(const orte_process_name_t*, pid_t*
/**
* Utility routine to retreive all process pids w/in a specified job.
*/
-ORTE_DECLSPEC int orte_pls_bproc_get_proc_pids(orte_jobid_t jobid, pid_t** pids, orte_std_cntr_t* num_pids);
+ORTE_DECLSPEC int orte_pls_bproc_get_proc_pids(orte_jobid_t jobid, pid_t** pids,
+ orte_std_cntr_t* num_pids,
+ opal_list_t *attrs);
+
/**
* Utility routine to get/set daemon pid
*/
diff --git a/orte/mca/pls/bproc/pls_bproc_state.c b/orte/mca/pls/bproc/pls_bproc_state.c
index 74671971ad..2196be2cdc 100644
--- a/orte/mca/pls/bproc/pls_bproc_state.c
+++ b/orte/mca/pls/bproc/pls_bproc_state.c
@@ -173,7 +173,7 @@ cleanup:
/**
* Retrieve all process pids for the specified job.
*/
-int orte_pls_bproc_get_proc_pids(orte_jobid_t jobid, pid_t **pids, orte_std_cntr_t* num_pids)
+int orte_pls_bproc_get_proc_pids(orte_jobid_t jobid, pid_t **pids, orte_std_cntr_t* num_pids, opal_list_t *attrs)
{
char *segment;
char *keys[2];
@@ -352,7 +352,7 @@ int orte_pls_bproc_comm_start(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY,
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_BPROC_ABORT,
ORTE_RML_PERSISTENT,
orte_pls_bproc_recv,
@@ -373,7 +373,7 @@ int orte_pls_bproc_comm_stop(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_BPROC_ABORT))) {
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_BPROC_ABORT))) {
ORTE_ERROR_LOG(rc);
}
mca_pls_bproc_component.recv_issued = false;
diff --git a/orte/mca/pls/cnos/pls_cnos.c b/orte/mca/pls/cnos/pls_cnos.c
index 6e8bd8611d..7c3a10ce7a 100644
--- a/orte/mca/pls/cnos/pls_cnos.c
+++ b/orte/mca/pls/cnos/pls_cnos.c
@@ -38,10 +38,10 @@
static int orte_pls_cnos_launch_job(orte_jobid_t jobid);
-static int orte_pls_cnos_terminate_job(orte_jobid_t jobid);
-static int orte_pls_cnos_terminate_orteds(orte_jobid_t jobid);
+static int orte_pls_cnos_terminate_job(orte_jobid_t jobid, opal_list_t *attrs);
+static int orte_pls_cnos_terminate_orteds(orte_jobid_t jobid, opal_list_t *attrs);
static int orte_pls_cnos_terminate_proc(const orte_process_name_t* proc_name);
-static int orte_pls_cnos_signal_job(orte_jobid_t jobid, int32_t signal);
+static int orte_pls_cnos_signal_job(orte_jobid_t jobid, int32_t signal, opal_list_t *attrs);
static int orte_pls_cnos_signal_proc(const orte_process_name_t* proc_name, int32_t signal);
static int orte_pls_cnos_finalize(void);
@@ -68,12 +68,12 @@ static int orte_pls_cnos_launch_job(orte_jobid_t jobid)
extern int killrank(rank_t RANK, int SIG);
#endif
-static int orte_pls_cnos_terminate_job(orte_jobid_t jobid)
+static int orte_pls_cnos_terminate_job(orte_jobid_t jobid, opal_list_t *attrs)
{
#ifdef HAVE_KILLRANK
orte_jobid_t my_jobid;
- orte_ns.get_jobid(&my_jobid, orte_process_info.my_name);
+ my_jobid = orte_process_info.my_name->jobid;
/* make sure it's my job */
if (jobid == my_jobid) {
@@ -89,12 +89,12 @@ static int orte_pls_cnos_terminate_job(orte_jobid_t jobid)
}
-static int orte_pls_cnos_terminate_orteds(orte_jobid_t jobid)
+static int orte_pls_cnos_terminate_orteds(orte_jobid_t jobid, opal_list_t *attrs)
{
#ifdef HAVE_KILLRANK
orte_jobid_t my_jobid;
- orte_ns.get_jobid(&my_jobid, orte_process_info.my_name);
+ my_jobid = orte_process_info.my_name->jobid;
/* make sure it's my job */
if (jobid == my_jobid) {
@@ -136,7 +136,7 @@ static int orte_pls_cnos_terminate_proc(const orte_process_name_t* proc_name)
}
-static int orte_pls_cnos_signal_job(orte_jobid_t jobid, int32_t signal)
+static int orte_pls_cnos_signal_job(orte_jobid_t jobid, int32_t signal, opal_list_t *attrs)
{
return ORTE_ERR_NOT_SUPPORTED;
}
diff --git a/orte/mca/pls/gridengine/pls_gridengine.h b/orte/mca/pls/gridengine/pls_gridengine.h
index efdfbd9ee5..e141261809 100644
--- a/orte/mca/pls/gridengine/pls_gridengine.h
+++ b/orte/mca/pls/gridengine/pls_gridengine.h
@@ -113,10 +113,10 @@ int orte_pls_gridengine_finalize(void);
* Interface
*/
int orte_pls_gridengine_launch_job(orte_jobid_t);
-int orte_pls_gridengine_terminate_job(orte_jobid_t);
-int orte_pls_gridengine_terminate_orteds(orte_jobid_t);
+int orte_pls_gridengine_terminate_job(orte_jobid_t, opal_list_t *attrs);
+int orte_pls_gridengine_terminate_orteds(orte_jobid_t, opal_list_t *attrs);
int orte_pls_gridengine_terminate_proc(const orte_process_name_t*);
-int orte_pls_gridengine_signal_job(orte_jobid_t, int32_t);
+int orte_pls_gridengine_signal_job(orte_jobid_t, int32_t, opal_list_t *attrs);
int orte_pls_gridengine_signal_proc(const orte_process_name_t*, int32_t);
/**
diff --git a/orte/mca/pls/gridengine/pls_gridengine_module.c b/orte/mca/pls/gridengine/pls_gridengine_module.c
index 8f6f2f6293..0cab075ec3 100644
--- a/orte/mca/pls/gridengine/pls_gridengine_module.c
+++ b/orte/mca/pls/gridengine/pls_gridengine_module.c
@@ -759,7 +759,7 @@ static int update_slot_keyval(orte_ras_node_t* ras_node, int* slot_cnt)
/**
* Query the registry for all nodes participating in the job
*/
-int orte_pls_gridengine_terminate_job(orte_jobid_t jobid)
+int orte_pls_gridengine_terminate_job(orte_jobid_t jobid, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -767,7 +767,7 @@ int orte_pls_gridengine_terminate_job(orte_jobid_t jobid)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
@@ -794,7 +794,7 @@ int orte_pls_gridengine_terminate_proc(const orte_process_name_t* proc)
/**
* Terminate the orteds for a given job
*/
-int orte_pls_gridengine_terminate_orteds(orte_jobid_t jobid)
+int orte_pls_gridengine_terminate_orteds(orte_jobid_t jobid, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -802,7 +802,7 @@ int orte_pls_gridengine_terminate_orteds(orte_jobid_t jobid)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
@@ -823,7 +823,7 @@ CLEANUP:
/**
* Signal all processes associated with this job
*/
-int orte_pls_gridengine_signal_job(orte_jobid_t jobid, int32_t signal)
+int orte_pls_gridengine_signal_job(orte_jobid_t jobid, int32_t signal, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -831,7 +831,7 @@ int orte_pls_gridengine_signal_job(orte_jobid_t jobid, int32_t signal)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&daemons);
return rc;
diff --git a/orte/mca/pls/pls.h b/orte/mca/pls/pls.h
index cef06ccb7a..2968ad7f49 100644
--- a/orte/mca/pls/pls.h
+++ b/orte/mca/pls/pls.h
@@ -202,12 +202,12 @@ typedef int (*orte_pls_base_module_launch_job_fn_t)(orte_jobid_t);
* Terminate any processes launched for the respective jobid by
* this component.
*/
-typedef int (*orte_pls_base_module_terminate_job_fn_t)(orte_jobid_t);
+typedef int (*orte_pls_base_module_terminate_job_fn_t)(orte_jobid_t, opal_list_t *attrs);
/**
* Terminate the daemons associated with this jobid
*/
-typedef int (*orte_pls_base_module_terminate_orteds_fn_t)(orte_jobid_t);
+typedef int (*orte_pls_base_module_terminate_orteds_fn_t)(orte_jobid_t, opal_list_t *attrs);
/**
* Terminate a specific process.
@@ -218,7 +218,7 @@ typedef int (*orte_pls_base_module_terminate_proc_fn_t)(const orte_process_name_
* Signal any processes launched for the respective jobid by
* this component.
*/
-typedef int (*orte_pls_base_module_signal_job_fn_t)(orte_jobid_t, int32_t);
+typedef int (*orte_pls_base_module_signal_job_fn_t)(orte_jobid_t, int32_t, opal_list_t *attrs);
/**
* Signal a specific process.
diff --git a/orte/mca/pls/poe/pls_poe_module.c b/orte/mca/pls/poe/pls_poe_module.c
index 4fc4adb57b..82e73f2b5f 100644
--- a/orte/mca/pls/poe/pls_poe_module.c
+++ b/orte/mca/pls/poe/pls_poe_module.c
@@ -60,10 +60,10 @@ extern char **environ;
* Local functions
*/
static int pls_poe_launch_job(orte_jobid_t jobid);
-static int pls_poe_terminate_job(orte_jobid_t jobid);
-static int pls_poe_terminate_orteds(orte_jobid_t jobid);
+static int pls_poe_terminate_job(orte_jobid_t jobid, opal_list_t *attrs);
+static int pls_poe_terminate_orteds(orte_jobid_t jobid, opal_list_t *attrs);
static int pls_poe_terminate_proc(const orte_process_name_t *name);
-static int pls_poe_signal_job(orte_jobid_t jobid, int32_t signal);
+static int pls_poe_signal_job(orte_jobid_t jobid, int32_t signal, opal_list_t *attrs);
static int pls_poe_signal_proc(const orte_process_name_t *name, int32_t signal);
static int pls_poe_finalize(void);
@@ -601,7 +601,7 @@ static int pls_poe_launch_job(orte_jobid_t jobid)
return ORTE_ERR_NOT_IMPLEMENTED;
}
-static int pls_poe_terminate_job(orte_jobid_t jobid)
+static int pls_poe_terminate_job(orte_jobid_t jobid, opal_list_t *attrs)
{
return ORTE_ERR_NOT_IMPLEMENTED;
}
@@ -612,12 +612,12 @@ static int pls_poe_terminate_proc(const orte_process_name_t *name)
return ORTE_ERR_NOT_IMPLEMENTED;
}
-static int pls_poe_terminate_orteds(orte_jobid_t jobid)
+static int pls_poe_terminate_orteds(orte_jobid_t jobid, opal_list_t *attrs)
{
return ORTE_ERR_NOT_IMPLEMENTED;
}
-static int pls_poe_signal_job(orte_jobid_t jobid, int32_t signal)
+static int pls_poe_signal_job(orte_jobid_t jobid, int32_t signal, opal_list_t *attrs)
{
return ORTE_ERR_NOT_IMPLEMENTED;
}
diff --git a/orte/mca/pls/proxy/pls_proxy.c b/orte/mca/pls/proxy/pls_proxy.c
index 1a0470dd2d..bdc6603fef 100644
--- a/orte/mca/pls/proxy/pls_proxy.c
+++ b/orte/mca/pls/proxy/pls_proxy.c
@@ -110,7 +110,7 @@ int orte_pls_proxy_launch(orte_jobid_t job)
return ORTE_SUCCESS;
}
-int orte_pls_proxy_terminate_job(orte_jobid_t job)
+int orte_pls_proxy_terminate_job(orte_jobid_t job, opal_list_t *attrs)
{
orte_buffer_t* cmd;
orte_buffer_t* answer;
@@ -140,6 +140,12 @@ int orte_pls_proxy_terminate_job(orte_jobid_t job)
return rc;
}
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, attrs, 1, ORTE_ATTR_LIST))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
if (0 > orte_rml.send_buffer(orte_pls_proxy_replica, cmd, ORTE_RML_TAG_PLS, 0)) {
ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
OBJ_RELEASE(cmd);
@@ -176,7 +182,7 @@ int orte_pls_proxy_terminate_job(orte_jobid_t job)
return ORTE_SUCCESS;
}
-int orte_pls_proxy_terminate_orteds(orte_jobid_t job)
+int orte_pls_proxy_terminate_orteds(orte_jobid_t job, opal_list_t *attrs)
{
orte_buffer_t* cmd;
orte_buffer_t* answer;
@@ -206,6 +212,12 @@ int orte_pls_proxy_terminate_orteds(orte_jobid_t job)
return rc;
}
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, attrs, 1, ORTE_ATTR_LIST))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
if (0 > orte_rml.send_buffer(orte_pls_proxy_replica, cmd, ORTE_RML_TAG_PLS, 0)) {
ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
OBJ_RELEASE(cmd);
@@ -242,7 +254,7 @@ int orte_pls_proxy_terminate_orteds(orte_jobid_t job)
return ORTE_SUCCESS;
}
-int orte_pls_proxy_signal_job(orte_jobid_t job, int32_t signal)
+int orte_pls_proxy_signal_job(orte_jobid_t job, int32_t signal, opal_list_t *attrs)
{
orte_buffer_t* cmd;
orte_buffer_t* answer;
@@ -278,6 +290,12 @@ int orte_pls_proxy_signal_job(orte_jobid_t job, int32_t signal)
return rc;
}
+ if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, attrs, 1, ORTE_ATTR_LIST))) {
+ ORTE_ERROR_LOG(rc);
+ OBJ_RELEASE(cmd);
+ return rc;
+ }
+
if (0 > orte_rml.send_buffer(orte_pls_proxy_replica, cmd, ORTE_RML_TAG_PLS, 0)) {
ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
OBJ_RELEASE(cmd);
diff --git a/orte/mca/pls/proxy/pls_proxy.h b/orte/mca/pls/proxy/pls_proxy.h
index df47d732fb..10db57fa4f 100644
--- a/orte/mca/pls/proxy/pls_proxy.h
+++ b/orte/mca/pls/proxy/pls_proxy.h
@@ -48,10 +48,10 @@ int orte_pls_proxy_finalize(void);
* proxy function prototypes
*/
int orte_pls_proxy_launch(orte_jobid_t job);
-int orte_pls_proxy_terminate_job(orte_jobid_t job);
-int orte_pls_proxy_terminate_orteds(orte_jobid_t job);
+int orte_pls_proxy_terminate_job(orte_jobid_t job, opal_list_t *attrs);
+int orte_pls_proxy_terminate_orteds(orte_jobid_t job, opal_list_t *attrs);
int orte_pls_proxy_terminate_proc(const orte_process_name_t* name);
-int orte_pls_proxy_signal_job(orte_jobid_t job, int32_t signal);
+int orte_pls_proxy_signal_job(orte_jobid_t job, int32_t signal, opal_list_t *attrs);
int orte_pls_proxy_signal_proc(const orte_process_name_t* name, int32_t signal);
diff --git a/orte/mca/pls/rsh/pls_rsh.h b/orte/mca/pls/rsh/pls_rsh.h
index 7770e83444..554161d9e2 100644
--- a/orte/mca/pls/rsh/pls_rsh.h
+++ b/orte/mca/pls/rsh/pls_rsh.h
@@ -49,10 +49,10 @@ int orte_pls_rsh_finalize(void);
* Interface
*/
int orte_pls_rsh_launch(orte_jobid_t);
-int orte_pls_rsh_terminate_job(orte_jobid_t);
-int orte_pls_rsh_terminate_orteds(orte_jobid_t);
+int orte_pls_rsh_terminate_job(orte_jobid_t, opal_list_t*);
+int orte_pls_rsh_terminate_orteds(orte_jobid_t, opal_list_t*);
int orte_pls_rsh_terminate_proc(const orte_process_name_t* proc_name);
-int orte_pls_rsh_signal_job(orte_jobid_t, int32_t);
+int orte_pls_rsh_signal_job(orte_jobid_t, int32_t, opal_list_t*);
int orte_pls_rsh_signal_proc(const orte_process_name_t* proc_name, int32_t);
/**
@@ -66,6 +66,7 @@ struct orte_pls_rsh_component_t {
bool timing;
bool reap;
bool assume_same_shell;
+ bool force_rsh;
int delay;
int priority;
char *agent_param;
diff --git a/orte/mca/pls/rsh/pls_rsh_component.c b/orte/mca/pls/rsh/pls_rsh_component.c
index dbe5a68c04..095fe2e002 100644
--- a/orte/mca/pls/rsh/pls_rsh_component.c
+++ b/orte/mca/pls/rsh/pls_rsh_component.c
@@ -134,6 +134,11 @@ int orte_pls_rsh_component_open(void)
}
mca_pls_rsh_component.num_concurrent = tmp;
+ mca_base_param_reg_int(c, "force_rsh",
+ "Force the launcher to always use rsh, even for local daemons",
+ false, false, false, &tmp);
+ mca_pls_rsh_component.force_rsh = OPAL_INT_TO_BOOL(tmp);
+
if (mca_pls_rsh_component.debug == 0) {
mca_base_param_reg_int_name("orte", "debug",
"Whether or not to enable debugging output for all ORTE components (0 or 1)",
diff --git a/orte/mca/pls/rsh/pls_rsh_module.c b/orte/mca/pls/rsh/pls_rsh_module.c
index f04b72ee88..ad0ab530b8 100644
--- a/orte/mca/pls/rsh/pls_rsh_module.c
+++ b/orte/mca/pls/rsh/pls_rsh_module.c
@@ -832,8 +832,9 @@ int orte_pls_rsh_launch(orte_jobid_t jobid)
* current nodename, which must be local. If that doesn't
* match, check using ifislocal().
*/
- if (0 == strcmp(rmaps_node->nodename, orte_system_info.nodename) ||
- opal_ifislocal(rmaps_node->nodename)) {
+ if (!mca_pls_rsh_component.force_rsh &&
+ (0 == strcmp(rmaps_node->nodename, orte_system_info.nodename) ||
+ opal_ifislocal(rmaps_node->nodename))) {
if (mca_pls_rsh_component.debug) {
opal_output(0, "pls:rsh: %s is a LOCAL node\n",
rmaps_node->nodename);
@@ -1103,7 +1104,7 @@ cleanup:
/**
* Terminate all processes for a given job
*/
-int orte_pls_rsh_terminate_job(orte_jobid_t jobid)
+int orte_pls_rsh_terminate_job(orte_jobid_t jobid, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -1113,7 +1114,7 @@ int orte_pls_rsh_terminate_job(orte_jobid_t jobid)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
@@ -1135,7 +1136,7 @@ CLEANUP:
/**
* Terminate the orteds for a given job
*/
-int orte_pls_rsh_terminate_orteds(orte_jobid_t jobid)
+int orte_pls_rsh_terminate_orteds(orte_jobid_t jobid, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -1145,7 +1146,7 @@ int orte_pls_rsh_terminate_orteds(orte_jobid_t jobid)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
@@ -1173,7 +1174,7 @@ int orte_pls_rsh_terminate_proc(const orte_process_name_t* proc)
return ORTE_ERR_NOT_IMPLEMENTED;
}
-int orte_pls_rsh_signal_job(orte_jobid_t jobid, int32_t signal)
+int orte_pls_rsh_signal_job(orte_jobid_t jobid, int32_t signal, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -1183,7 +1184,7 @@ int orte_pls_rsh_signal_job(orte_jobid_t jobid, int32_t signal)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&daemons);
return rc;
diff --git a/orte/mca/pls/slurm/pls_slurm_module.c b/orte/mca/pls/slurm/pls_slurm_module.c
index 575440699d..31f187c5b5 100644
--- a/orte/mca/pls/slurm/pls_slurm_module.c
+++ b/orte/mca/pls/slurm/pls_slurm_module.c
@@ -73,10 +73,10 @@
* Local functions
*/
static int pls_slurm_launch_job(orte_jobid_t jobid);
-static int pls_slurm_terminate_job(orte_jobid_t jobid);
-static int pls_slurm_terminate_orteds(orte_jobid_t jobid);
+static int pls_slurm_terminate_job(orte_jobid_t jobid, opal_list_t *attrs);
+static int pls_slurm_terminate_orteds(orte_jobid_t jobid, opal_list_t *attrs);
static int pls_slurm_terminate_proc(const orte_process_name_t *name);
-static int pls_slurm_signal_job(orte_jobid_t jobid, int32_t signal);
+static int pls_slurm_signal_job(orte_jobid_t jobid, int32_t signal, opal_list_t *attrs);
static int pls_slurm_signal_proc(const orte_process_name_t *name, int32_t signal);
static int pls_slurm_finalize(void);
@@ -421,7 +421,7 @@ cleanup:
}
-static int pls_slurm_terminate_job(orte_jobid_t jobid)
+static int pls_slurm_terminate_job(orte_jobid_t jobid, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -429,7 +429,7 @@ static int pls_slurm_terminate_job(orte_jobid_t jobid)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
@@ -452,7 +452,7 @@ CLEANUP:
/**
* Terminate the orteds for a given job
*/
-static int pls_slurm_terminate_orteds(orte_jobid_t jobid)
+static int pls_slurm_terminate_orteds(orte_jobid_t jobid, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -460,7 +460,7 @@ static int pls_slurm_terminate_orteds(orte_jobid_t jobid)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
@@ -493,7 +493,7 @@ static int pls_slurm_terminate_proc(const orte_process_name_t *name)
/**
* Signal all the processes in the child srun by sending the signal directly to it
*/
-static int pls_slurm_signal_job(orte_jobid_t jobid, int32_t signal)
+static int pls_slurm_signal_job(orte_jobid_t jobid, int32_t signal, opal_list_t *attrs)
{
if (0 != srun_pid) {
kill(srun_pid, (int)signal);
diff --git a/orte/mca/pls/tm/pls_tm_module.c b/orte/mca/pls/tm/pls_tm_module.c
index 3c022097f4..ea969daaf4 100644
--- a/orte/mca/pls/tm/pls_tm_module.c
+++ b/orte/mca/pls/tm/pls_tm_module.c
@@ -79,10 +79,10 @@
* Local functions
*/
static int pls_tm_launch_job(orte_jobid_t jobid);
-static int pls_tm_terminate_job(orte_jobid_t jobid);
-static int pls_tm_terminate_orteds(orte_jobid_t jobid);
+static int pls_tm_terminate_job(orte_jobid_t jobid, opal_list_t *attrs);
+static int pls_tm_terminate_orteds(orte_jobid_t jobid, opal_list_t *attrs);
static int pls_tm_terminate_proc(const orte_process_name_t *name);
-static int pls_tm_signal_job(orte_jobid_t jobid, int32_t signal);
+static int pls_tm_signal_job(orte_jobid_t jobid, int32_t signal, opal_list_t *attrs);
static int pls_tm_signal_proc(const orte_process_name_t *name, int32_t signal);
static int pls_tm_finalize(void);
@@ -545,7 +545,7 @@ static int pls_tm_launch_job(orte_jobid_t jobid)
}
-static int pls_tm_terminate_job(orte_jobid_t jobid)
+static int pls_tm_terminate_job(orte_jobid_t jobid, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -553,7 +553,7 @@ static int pls_tm_terminate_job(orte_jobid_t jobid)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
@@ -576,7 +576,7 @@ CLEANUP:
/**
* Terminate the orteds for a given job
*/
-int pls_tm_terminate_orteds(orte_jobid_t jobid)
+int pls_tm_terminate_orteds(orte_jobid_t jobid, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -584,7 +584,7 @@ int pls_tm_terminate_orteds(orte_jobid_t jobid)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
@@ -614,7 +614,7 @@ static int pls_tm_terminate_proc(const orte_process_name_t *name)
}
-static int pls_tm_signal_job(orte_jobid_t jobid, int32_t signal)
+static int pls_tm_signal_job(orte_jobid_t jobid, int32_t signal, opal_list_t *attrs)
{
int rc;
opal_list_t daemons;
@@ -622,7 +622,7 @@ static int pls_tm_signal_job(orte_jobid_t jobid, int32_t signal)
/* construct the list of active daemons on this job */
OBJ_CONSTRUCT(&daemons, opal_list_t);
- if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(&daemons, jobid, attrs))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&daemons);
return rc;
diff --git a/orte/mca/ras/base/ras_base_receive.c b/orte/mca/ras/base/ras_base_receive.c
index e0e725e923..a24abafd6d 100644
--- a/orte/mca/ras/base/ras_base_receive.c
+++ b/orte/mca/ras/base/ras_base_receive.c
@@ -50,7 +50,7 @@ int orte_ras_base_comm_start(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY,
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_RAS,
ORTE_RML_PERSISTENT,
orte_ras_base_recv,
@@ -71,7 +71,7 @@ int orte_ras_base_comm_stop(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_RAS))) {
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_RAS))) {
ORTE_ERROR_LOG(rc);
}
recv_issued = false;
diff --git a/orte/mca/rds/base/rds_base_receive.c b/orte/mca/rds/base/rds_base_receive.c
index 4ab39fdf39..da03378bc4 100644
--- a/orte/mca/rds/base/rds_base_receive.c
+++ b/orte/mca/rds/base/rds_base_receive.c
@@ -49,7 +49,7 @@ int orte_rds_base_comm_start(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY,
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_RDS,
ORTE_RML_PERSISTENT,
orte_rds_base_recv,
@@ -70,7 +70,7 @@ int orte_rds_base_comm_stop(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_RDS))) {
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_RDS))) {
ORTE_ERROR_LOG(rc);
}
recv_issued = false;
diff --git a/orte/mca/rmaps/base/rmaps_base_receive.c b/orte/mca/rmaps/base/rmaps_base_receive.c
index 45c197eda3..6b70e9dd47 100644
--- a/orte/mca/rmaps/base/rmaps_base_receive.c
+++ b/orte/mca/rmaps/base/rmaps_base_receive.c
@@ -50,7 +50,7 @@ int orte_rmaps_base_comm_start(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY,
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_RMAPS,
ORTE_RML_PERSISTENT,
orte_rmaps_base_recv,
@@ -71,7 +71,7 @@ int orte_rmaps_base_comm_stop(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_RMAPS))) {
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_RMAPS))) {
ORTE_ERROR_LOG(rc);
}
recv_issued = false;
diff --git a/orte/mca/rmgr/base/data_type_support/rmgr_data_type_packing_fns.c b/orte/mca/rmgr/base/data_type_support/rmgr_data_type_packing_fns.c
index 2b7528a6a5..2d18892008 100644
--- a/orte/mca/rmgr/base/data_type_support/rmgr_data_type_packing_fns.c
+++ b/orte/mca/rmgr/base/data_type_support/rmgr_data_type_packing_fns.c
@@ -214,8 +214,13 @@ int orte_rmgr_base_pack_attr_list(orte_buffer_t *buffer, void *src,
opal_list_item_t *item;
orte_std_cntr_t num_attr;
- /* get the number of attributes and pack it */
- num_attr = (orte_std_cntr_t)opal_list_get_size(attrs);
+ /* if the list is NULL, we have zero attributes */
+ if (NULL == src) {
+ num_attr = 0;
+ } else {
+ /* get the number of attributes */
+ num_attr = (orte_std_cntr_t)opal_list_get_size(attrs);
+ }
if (ORTE_SUCCESS != (rc = orte_dss_pack_buffer(buffer, (void*)&num_attr, 1, ORTE_STD_CNTR))) {
ORTE_ERROR_LOG(rc);
return rc;
diff --git a/orte/mca/rmgr/base/rmgr_base_attribute_fns.c b/orte/mca/rmgr/base/rmgr_base_attribute_fns.c
index a12f65d756..bc6c838059 100644
--- a/orte/mca/rmgr/base/rmgr_base_attribute_fns.c
+++ b/orte/mca/rmgr/base/rmgr_base_attribute_fns.c
@@ -43,6 +43,11 @@ orte_attribute_t* orte_rmgr_base_find_attribute(opal_list_t* attr_list, char* ke
opal_list_item_t *item;
orte_attribute_t *kval;
+ if (NULL == attr_list) {
+ /* if the list is NULL, then by definition we couldn't find it! */
+ return NULL;
+ }
+
for (item = opal_list_get_first(attr_list);
item != opal_list_get_end(attr_list);
item = opal_list_get_next(item)) {
@@ -69,6 +74,11 @@ int orte_rmgr_base_add_attribute(opal_list_t* attr_list, char* key,
orte_gpr_keyval_t *kval;
orte_attribute_t *attr;
+ /* protect against NULL case */
+ if (NULL == attr_list) {
+ return ORTE_ERR_BAD_PARAM;
+ }
+
/* see if this attribute is already present */
if (NULL != (attr = orte_rmgr_base_find_attribute(attr_list, key))) {
/** found it - do we want to replace this value? */
@@ -108,6 +118,11 @@ int orte_rmgr_base_merge_attributes(opal_list_t* target, opal_list_t* source, bo
opal_list_item_t *item;
orte_attribute_t *attr;
+ /* protect against NULL cases */
+ if (NULL == target || NULL == source) {
+ return ORTE_ERR_BAD_PARAM;
+ }
+
/* Since the add_attribute function takes care of the override issue, we just
* need to cycle through the source list and "add" everything to the target
*/
@@ -136,6 +151,11 @@ int orte_rmgr_base_delete_attribute(opal_list_t* attr_list, char* key)
opal_list_item_t *item;
orte_attribute_t *kval;
+ /* protect against the NULL case */
+ if (NULL == attr_list) {
+ return ORTE_SUCCESS;
+ }
+
for (item = opal_list_get_first(attr_list);
item != opal_list_get_end(attr_list);
item = opal_list_get_next(item)) {
diff --git a/orte/mca/rmgr/base/rmgr_base_receive.c b/orte/mca/rmgr/base/rmgr_base_receive.c
index c05acdbc50..2137a7c54f 100644
--- a/orte/mca/rmgr/base/rmgr_base_receive.c
+++ b/orte/mca/rmgr/base/rmgr_base_receive.c
@@ -51,7 +51,7 @@ int orte_rmgr_base_comm_start(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY,
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_RMGR,
ORTE_RML_PERSISTENT,
orte_rmgr_base_recv,
@@ -71,7 +71,7 @@ int orte_rmgr_base_comm_stop(void)
return ORTE_SUCCESS;
}
- if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_RML_NAME_ANY, ORTE_RML_TAG_RMGR))) {
+ if (ORTE_SUCCESS != (rc = orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_RMGR))) {
ORTE_ERROR_LOG(rc);
}
recv_issued = false;
@@ -148,7 +148,7 @@ void orte_rmgr_base_recv(int status, orte_process_name_t* sender,
count = 1;
if(ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &attrs, &count, ORTE_ATTR_LIST))) {
ORTE_ERROR_LOG(rc);
- goto CLEANUP_SPAWN;
+ goto SEND_ANSWER;
}
/* process the request */
@@ -156,7 +156,11 @@ void orte_rmgr_base_recv(int status, orte_process_name_t* sender,
ORTE_ERROR_LOG(rc);
goto CLEANUP_SPAWN;
}
-
+ while (NULL != (item = opal_list_remove_first(&attrs))) {
+ OBJ_RELEASE(item);
+ }
+ OBJ_DESTRUCT(&attrs);
+
/* return the new jobid */
if(ORTE_SUCCESS != (rc = orte_dss.pack(&answer, &job, 1, ORTE_JOBID))) {
ORTE_ERROR_LOG(rc);
diff --git a/orte/mca/rmgr/base/rmgr_base_stage_gate.c b/orte/mca/rmgr/base/rmgr_base_stage_gate.c
index 5654ffd62d..813fca199d 100644
--- a/orte/mca/rmgr/base/rmgr_base_stage_gate.c
+++ b/orte/mca/rmgr/base/rmgr_base_stage_gate.c
@@ -62,6 +62,8 @@ int orte_rmgr_base_proc_stage_gate_mgr(orte_gpr_notify_message_t *msg)
orte_std_cntr_t n=0;
int rc;
orte_jobid_t job;
+ opal_list_t attrs;
+ opal_list_item_t *item;
OPAL_TRACE(1);
@@ -88,13 +90,21 @@ int orte_rmgr_base_proc_stage_gate_mgr(orte_gpr_notify_message_t *msg)
OPAL_TRACE_ARG1(1, job);
/* need the list of peers for this job so we can send them the xcast.
- * obtain this list from the name service's get_job_peers function
- */
- if (ORTE_SUCCESS != (rc = orte_ns.get_job_peers(&recipients, &n, job))) {
+ * obtain this list from the name service's get_job_peers function
+ */
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ if (ORTE_SUCCESS != (rc = orte_rmgr.add_attribute(&attrs, ORTE_NS_USE_JOBID, ORTE_JOBID,
+ &job, ORTE_RMGR_ATTR_OVERRIDE))) {
ORTE_ERROR_LOG(rc);
+ OBJ_DESTRUCT(&attrs);
return rc;
}
-
+
+ if (ORTE_SUCCESS != (rc = orte_ns.get_peers(&recipients, &n, &attrs))) {
+ ORTE_ERROR_LOG(rc);
+ goto CLEANUP;
+ }
+
/* set the job state to the appropriate level */
if (orte_schema.check_std_trigger_name(msg->target, ORTE_ALL_LAUNCHED_TRIGGER)) {
if (ORTE_SUCCESS != (rc = orte_smr.set_job_state(job, ORTE_JOB_STATE_LAUNCHED))) {
@@ -146,5 +156,11 @@ int orte_rmgr_base_proc_stage_gate_mgr(orte_gpr_notify_message_t *msg)
CLEANUP:
if (NULL != recipients) free(recipients);
+
+ while (NULL != (item = opal_list_remove_first(&attrs))) {
+ OBJ_RELEASE(item);
+ }
+ OBJ_DESTRUCT(&attrs);
+
return rc;
}
diff --git a/orte/mca/rmgr/proxy/rmgr_proxy.c b/orte/mca/rmgr/proxy/rmgr_proxy.c
index 45a28d88aa..0e1186c828 100644
--- a/orte/mca/rmgr/proxy/rmgr_proxy.c
+++ b/orte/mca/rmgr/proxy/rmgr_proxy.c
@@ -122,7 +122,7 @@ static int orte_rmgr_proxy_setup_job(orte_app_context_t** app_context,
return rc;
}
- /* pack any attributes */
+ /* pack the attributes */
if (ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, attrs, 1, ORTE_ATTR_LIST))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&cmd);
@@ -130,7 +130,7 @@ static int orte_rmgr_proxy_setup_job(orte_app_context_t** app_context,
}
/* send the command */
- if(0 > (rc = orte_rml.send_buffer(ORTE_RML_NAME_SEED, &cmd, ORTE_RML_TAG_RMGR, 0))) {
+ if(0 > (rc = orte_rml.send_buffer(ORTE_PROC_MY_HNP, &cmd, ORTE_RML_TAG_RMGR, 0))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&cmd);
return rc;
@@ -139,7 +139,7 @@ static int orte_rmgr_proxy_setup_job(orte_app_context_t** app_context,
/* wait for response */
OBJ_CONSTRUCT(&rsp, orte_buffer_t);
- if(0 > (rc = orte_rml.recv_buffer(ORTE_RML_NAME_SEED, &rsp, ORTE_RML_TAG_RMGR))) {
+ if(0 > (rc = orte_rml.recv_buffer(ORTE_PROC_MY_HNP, &rsp, ORTE_RML_TAG_RMGR))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&rsp);
return rc;
@@ -197,7 +197,7 @@ static int orte_rmgr_proxy_setup_stage_gates(orte_jobid_t jobid)
}
/* send the command */
- if(0 > (rc = orte_rml.send_buffer(ORTE_RML_NAME_SEED, &cmd, ORTE_RML_TAG_RMGR, 0))) {
+ if(0 > (rc = orte_rml.send_buffer(ORTE_PROC_MY_HNP, &cmd, ORTE_RML_TAG_RMGR, 0))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&cmd);
return rc;
@@ -206,7 +206,7 @@ static int orte_rmgr_proxy_setup_stage_gates(orte_jobid_t jobid)
/* wait for response */
OBJ_CONSTRUCT(&rsp, orte_buffer_t);
- if(0 > (rc = orte_rml.recv_buffer(ORTE_RML_NAME_SEED, &rsp, ORTE_RML_TAG_RMGR))) {
+ if(0 > (rc = orte_rml.recv_buffer(ORTE_PROC_MY_HNP, &rsp, ORTE_RML_TAG_RMGR))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&rsp);
return rc;
@@ -398,7 +398,7 @@ static int orte_rmgr_proxy_spawn_job(
*/
if (flags & ORTE_RMGR_SETUP) {
if (ORTE_SUCCESS !=
- (rc = orte_rmgr_proxy_setup_job(app_context,num_context,jobid, attributes))) {
+ (rc = orte_rmgr_proxy_setup_job(app_context, num_context, jobid, attributes))) {
ORTE_ERROR_LOG(rc);
return rc;
}
diff --git a/orte/mca/rmgr/rmgr.h b/orte/mca/rmgr/rmgr.h
index 4089bfc38d..b14b56a294 100644
--- a/orte/mca/rmgr/rmgr.h
+++ b/orte/mca/rmgr/rmgr.h
@@ -66,7 +66,7 @@ typedef int (*orte_rmgr_base_module_setup_job_fn_t)(
orte_app_context_t** app_context,
orte_std_cntr_t num_context,
orte_jobid_t *jobid,
- opal_list_t *attributes);
+ opal_list_t *attrs);
/*
* Callback function for resource manager
diff --git a/orte/mca/rmgr/urm/rmgr_urm.c b/orte/mca/rmgr/urm/rmgr_urm.c
index ea4b456c5f..1c0fc59465 100644
--- a/orte/mca/rmgr/urm/rmgr_urm.c
+++ b/orte/mca/rmgr/urm/rmgr_urm.c
@@ -133,7 +133,7 @@ static int orte_rmgr_urm_setup_job(orte_app_context_t** app_context,
*jobid = *jptr;
} else {
/* allocate a jobid */
- if (ORTE_SUCCESS != (rc = orte_ns.create_jobid(jobid))) {
+ if (ORTE_SUCCESS != (rc = orte_ns.create_jobid(jobid, attrs))) {
ORTE_ERROR_LOG(rc);
return rc;
}
@@ -349,15 +349,12 @@ static int orte_rmgr_urm_spawn_job(
* Initialize job segment and allocate resources
*/ /* JJH Insert C/N mapping stuff here */
- /* Only do this step if we have been asked to do it via the ORTE_RMGR_SPAWN_FLOW
- * attribute, or if no flow flags were provided
- * If the jobid = ORTE_JOBID_INVALID, then we need to
- * get one assigned to us. Otherwise, we are entering
- * with a valid jobid, so no need to get one
+ /* Only do this step if we have been asked to do it via the
+ * ORTE_RMGR_SPAWN_FLOW attribute
*/
if (flags & ORTE_RMGR_SETUP) {
if (ORTE_SUCCESS !=
- (rc = orte_rmgr_urm_setup_job(app_context,num_context,jobid,attributes))) {
+ (rc = orte_rmgr_urm_setup_job(app_context, num_context, jobid, attributes))) {
ORTE_ERROR_LOG(rc);
return rc;
}
diff --git a/orte/mca/rml/base/rml_base_open.c b/orte/mca/rml/base/rml_base_open.c
index 65067144cc..7406929aff 100644
--- a/orte/mca/rml/base/rml_base_open.c
+++ b/orte/mca/rml/base/rml_base_open.c
@@ -43,8 +43,6 @@
orte_rml_base_t orte_rml_base;
orte_rml_module_t orte_rml;
-orte_process_name_t orte_rml_name_any = { ORTE_CELLID_MAX, ORTE_JOBID_MAX, ORTE_VPID_MAX };
-orte_process_name_t orte_rml_name_seed = { 0, 0, 0 };
/**
* Function for finding and opening either all MCA components, or the one
diff --git a/orte/mca/rml/rml.h b/orte/mca/rml/rml.h
index dfca8306fb..d4366a50de 100644
--- a/orte/mca/rml/rml.h
+++ b/orte/mca/rml/rml.h
@@ -41,14 +41,6 @@
extern "C" {
#endif
-/*
- * Well known addresses
- */
-
-ORTE_DECLSPEC extern orte_process_name_t orte_rml_name_any;
-ORTE_DECLSPEC extern orte_process_name_t orte_rml_name_seed;
-
-
/*
* RML Module function prototypes.
*/
diff --git a/orte/mca/rml/rml_types.h b/orte/mca/rml/rml_types.h
index 2dc7a362d9..81bb9c4756 100644
--- a/orte/mca/rml/rml_types.h
+++ b/orte/mca/rml/rml_types.h
@@ -73,20 +73,5 @@ typedef uint32_t orte_rml_tag_t;
#define ORTE_RML_PERSISTENT 0x08 /**< posted non-blocking recv is persistent */
#define ORTE_RML_NON_PERSISTENT 0x00
-/**
- * The wildcard for receives from any peer.
- */
-#define ORTE_RML_NAME_ANY &orte_rml_name_any
-
-/**
- * Process name of seed
- */
-#define ORTE_RML_NAME_SEED &orte_rml_name_seed
-
-/**
- * Process name of self
- */
-#define ORTE_RML_NAME_SELF orte_process_info.my_name
-
#endif /* RML_TYPES */
diff --git a/orte/mca/schema/schema_types.h b/orte/mca/schema/schema_types.h
index dc837ca56a..b32a0eabaf 100644
--- a/orte/mca/schema/schema_types.h
+++ b/orte/mca/schema/schema_types.h
@@ -23,6 +23,16 @@
#include "orte/mca/ns/ns_types.h"
#include "opal/util/error.h"
+/**
+* Standard characters used in ORTE
+ */
+#define ORTE_SCHEMA_DELIMITER_CHAR '.'
+#define ORTE_SCHEMA_DELIMITER_STRING "."
+#define ORTE_SCHEMA_WILDCARD_CHAR '*'
+#define ORTE_SCHEMA_WILDCARD_STRING "*"
+#define ORTE_SCHEMA_INVALID_CHAR '$'
+#define ORTE_SCHEMA_INVALID_STRING "$"
+
/*
* Standard names used across the system
*/
diff --git a/orte/mca/sds/bproc/sds_bproc_module.c b/orte/mca/sds/bproc/sds_bproc_module.c
index e5bb0637b0..7c5fd7bc4e 100644
--- a/orte/mca/sds/bproc/sds_bproc_module.c
+++ b/orte/mca/sds/bproc/sds_bproc_module.c
@@ -52,7 +52,7 @@ int orte_sds_bproc_set_name(void)
id = mca_base_param_register_string("ns", "nds", "name", NULL, NULL);
mca_base_param_lookup_string(id, &name_string);
if(name_string != NULL) {
- if (ORTE_SUCCESS != (rc = orte_ns_base_convert_string_to_process_name(
+ if (ORTE_SUCCESS != (rc = orte_ns.convert_string_to_process_name(
&(orte_process_info.my_name),
name_string))) {
ORTE_ERROR_LOG(rc);
diff --git a/orte/mca/sds/env/sds_env_module.c b/orte/mca/sds/env/sds_env_module.c
index 08104432e3..7b5b388898 100644
--- a/orte/mca/sds/env/sds_env_module.c
+++ b/orte/mca/sds/env/sds_env_module.c
@@ -28,7 +28,7 @@
#include "opal/mca/base/mca_base_param.h"
#include "orte/mca/ns/ns.h"
#include "orte/mca/errmgr/errmgr.h"
-#include "orte/mca/ns/base/base.h"
+#include "orte/mca/ns/ns.h"
orte_sds_base_module_t orte_sds_env_module = {
@@ -50,7 +50,7 @@ orte_sds_env_set_name(void)
mca_base_param_lookup_string(id, &name_string);
if(name_string != NULL) {
- if (ORTE_SUCCESS != (rc = orte_ns_base_convert_string_to_process_name(
+ if (ORTE_SUCCESS != (rc = orte_ns.convert_string_to_process_name(
&(orte_process_info.my_name),
name_string))) {
ORTE_ERROR_LOG(rc);
diff --git a/orte/mca/sds/pipe/sds_pipe_module.c b/orte/mca/sds/pipe/sds_pipe_module.c
index 821226667e..9c0b8e4d46 100644
--- a/orte/mca/sds/pipe/sds_pipe_module.c
+++ b/orte/mca/sds/pipe/sds_pipe_module.c
@@ -29,6 +29,8 @@
#include "orte/util/proc_info.h"
#include "opal/util/output.h"
#include "opal/mca/base/mca_base_param.h"
+
+#include "orte/dss/dss.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/ns/ns.h"
#include "orte/mca/ns/base/base.h"
@@ -59,7 +61,7 @@ orte_sds_pipe_set_name(void)
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
- if(ORTE_SUCCESS != (rc = orte_ns.copy_process_name(&orte_process_info.my_name, &name))) {
+ if(ORTE_SUCCESS != (rc = orte_dss.copy((void**)&orte_process_info.my_name, &name, ORTE_NAME))) {
ORTE_ERROR_LOG(rc);
return rc;
}
diff --git a/orte/mca/sds/singleton/sds_singleton_module.c b/orte/mca/sds/singleton/sds_singleton_module.c
index c6982b7013..6f481a2999 100644
--- a/orte/mca/sds/singleton/sds_singleton_module.c
+++ b/orte/mca/sds/singleton/sds_singleton_module.c
@@ -48,10 +48,7 @@ orte_sds_singleton_set_name(void)
return rc;
}
- if (ORTE_SUCCESS != (rc = orte_ns.get_vpid(&vpid, orte_process_info.my_name))) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
+ vpid = ORTE_PROC_MY_NAME->vpid;
orte_process_info.num_procs = 1;
orte_process_info.vpid_start = vpid;
diff --git a/orte/mca/sds/slurm/sds_slurm_module.c b/orte/mca/sds/slurm/sds_slurm_module.c
index d441c03267..5432516bc8 100644
--- a/orte/mca/sds/slurm/sds_slurm_module.c
+++ b/orte/mca/sds/slurm/sds_slurm_module.c
@@ -68,7 +68,7 @@ orte_sds_slurm_set_name(void)
if(name_string != NULL) {
if (ORTE_SUCCESS !=
- (rc = orte_ns_base_convert_string_to_process_name(&(orte_process_info.my_name),
+ (rc = orte_ns.convert_string_to_process_name(&(orte_process_info.my_name),
name_string))) {
ORTE_ERROR_LOG(rc);
free(name_string);
diff --git a/orte/mca/smr/base/smr_base_get_proc_state.c b/orte/mca/smr/base/smr_base_get_proc_state.c
index 035978bbf5..e730bdcb33 100644
--- a/orte/mca/smr/base/smr_base_get_proc_state.c
+++ b/orte/mca/smr/base/smr_base_get_proc_state.c
@@ -47,10 +47,7 @@ int orte_smr_base_get_proc_state(orte_proc_state_t *state,
orte_jobid_t jobid;
bool found1=false, found2=false;
- if (ORTE_SUCCESS != (rc = orte_ns.get_jobid(&jobid, proc))) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
+ jobid = proc->jobid;
if (ORTE_SUCCESS != (rc = orte_schema.get_job_segment_name(&segment, jobid))) {
ORTE_ERROR_LOG(rc);
diff --git a/orte/mca/smr/base/smr_base_set_proc_state.c b/orte/mca/smr/base/smr_base_set_proc_state.c
index d7b8c98181..58fef4be4a 100644
--- a/orte/mca/smr/base/smr_base_set_proc_state.c
+++ b/orte/mca/smr/base/smr_base_set_proc_state.c
@@ -44,10 +44,7 @@ int orte_smr_base_set_proc_state(orte_process_name_t *proc,
orte_exit_code_t exit_code;
char *segment;
- if (ORTE_SUCCESS != (rc = orte_ns.get_jobid(&jobid, proc))) {
- ORTE_ERROR_LOG(rc);
- return rc;
- }
+ jobid = proc->jobid;
if (ORTE_SUCCESS != (rc = orte_schema.get_job_segment_name(&segment, jobid))) {
ORTE_ERROR_LOG(rc);
@@ -61,12 +58,7 @@ int orte_smr_base_set_proc_state(orte_process_name_t *proc,
return rc;
}
- if (ORTE_SUCCESS != (rc = orte_ns.get_vpid(&vpid, proc))) {
- ORTE_ERROR_LOG(rc);
- OBJ_RELEASE(value);
- free(segment);
- return rc;
- }
+ vpid = proc->vpid;
if (ORTE_VPID_MAX != vpid) { /* check for wildcard case - leave tokens alone if so */
if (ORTE_SUCCESS != (rc = orte_schema.get_proc_tokens(&(value->tokens), &(value->num_tokens), proc))) {
diff --git a/orte/runtime/orte_init_stage1.c b/orte/runtime/orte_init_stage1.c
index c1b4cb1800..f7f161edf9 100644
--- a/orte/runtime/orte_init_stage1.c
+++ b/orte/runtime/orte_init_stage1.c
@@ -371,11 +371,7 @@ int orte_init_stage1(bool infrastructure)
char *site, *resource;
orte_app_context_t *app;
- if (ORTE_SUCCESS != (ret = orte_ns.get_jobid(&my_jobid, orte_process_info.my_name))) {
- ORTE_ERROR_LOG(ret);
- error = "orte_ns.get_jobid for singleton/seed";
- goto error;
- }
+ my_jobid = ORTE_PROC_MY_NAME->jobid;
/* If there is no existing cellid, create one */
my_cellid = 0; /* JJH Assertion/Repair until cellid's are fixed */
@@ -399,11 +395,7 @@ int orte_init_stage1(bool infrastructure)
goto error;
}
- if (ORTE_SUCCESS != (ret = orte_ns.get_cellid(&my_cellid, orte_process_info.my_name))) {
- ORTE_ERROR_LOG(ret);
- error = "orte_ns.get_cellid for singleton/seed";
- goto error;
- }
+ my_cellid = ORTE_PROC_MY_NAME->cellid;
/* set the rest of the infrastructure */
app = OBJ_NEW(orte_app_context_t);
diff --git a/orte/runtime/orte_restart.c b/orte/runtime/orte_restart.c
index cd2ab1ac76..5e805854a4 100644
--- a/orte/runtime/orte_restart.c
+++ b/orte/runtime/orte_restart.c
@@ -33,6 +33,8 @@
#include "opal/mca/mca.h"
#include "opal/mca/base/base.h"
#include "opal/mca/base/mca_base_param.h"
+
+#include "orte/dss/dss.h"
#include "orte/mca/iof/base/base.h"
#include "orte/mca/rml/base/base.h"
#include "orte/mca/errmgr/base/base.h"
@@ -60,11 +62,11 @@ int orte_restart(orte_process_name_t *name, const char* uri)
orte_process_name_t* old_name;
orte_process_name_t* new_name;
- if (ORTE_SUCCESS != (rc = orte_ns.copy_process_name(&old_name, orte_process_info.my_name))) {
+ if (ORTE_SUCCESS != (rc = orte_dss.copy((void**)&old_name, orte_process_info.my_name, ORTE_NAME))) {
ORTE_ERROR_LOG(rc);
return rc;
}
- if (ORTE_SUCCESS != (rc = orte_ns.copy_process_name(&new_name, name))) {
+ if (ORTE_SUCCESS != (rc = orte_dss.copy((void**)&new_name, name, ORTE_NAME))) {
ORTE_ERROR_LOG(rc);
return rc;
}
diff --git a/orte/runtime/orte_setup_hnp.c b/orte/runtime/orte_setup_hnp.c
index 222d590b9c..d54d5b2a5a 100644
--- a/orte/runtime/orte_setup_hnp.c
+++ b/orte/runtime/orte_setup_hnp.c
@@ -333,7 +333,7 @@ MOVEON:
OBJ_CONSTRUCT(&orte_setup_hnp_condition, opal_condition_t);
/* get a jobid for the probe */
- rc = orte_ns.create_jobid(&jobid);
+ rc = orte_ns.create_jobid(&jobid, NULL);
if (ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return rc;
@@ -451,7 +451,7 @@ MOVEON:
}
/* issue the non-blocking recv to get the probe's findings */
- rc = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY, ORTE_RML_TAG_PROBE,
+ rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_PROBE,
0, orte_setup_hnp_recv, NULL);
if(rc < 0) {
ORTE_ERROR_LOG(rc);
diff --git a/orte/test/unit/ns/Makefile b/orte/test/unit/ns/Makefile
new file mode 100644
index 0000000000..51e5d404f4
--- /dev/null
+++ b/orte/test/unit/ns/Makefile
@@ -0,0 +1,12 @@
+PROGS = ns_string_fns ns_peers
+
+all: $(PROGS)
+
+CC = ortecc
+CFLAGS = -g
+CXX = ortec++
+CXXFLAGS = -g
+FFLAGS = -g
+
+clean:
+ rm -f $(PROGS) *~
diff --git a/orte/test/unit/ns/ns_peers.c b/orte/test/unit/ns/ns_peers.c
new file mode 100644
index 0000000000..0dafc419c1
--- /dev/null
+++ b/orte/test/unit/ns/ns_peers.c
@@ -0,0 +1,204 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+
+#include "orte_config.h"
+#include "orte/orte_constants.h"
+
+#include <stdio.h>
+#include <string.h>
+
+#include "orte/mca/errmgr/errmgr.h"
+#include "orte/mca/rmgr/rmgr.h"
+
+#include "orte/mca/ns/base/base.h"
+
+int main(int argc, char **argv)
+{
+ orte_process_name_t *test_name;
+ orte_process_name_t *peers, *jptr;
+ orte_jobid_t parent, jobs[5], *jobdesc, root;
+ orte_std_cntr_t j, num_jobs, npeers;
+ orte_vpid_t vpid, vpids[5];
+ int i, rc;
+ opal_list_t attrs;
+ opal_list_item_t *item;
+ orte_attribute_t *attr;
+
+ if (ORTE_SUCCESS != orte_init(true)) {
+ fprintf(stderr, "failed to start ORTE\n");
+ exit (1);
+ }
+
+ /* create parent jobid */
+ if (ORTE_SUCCESS != (rc = orte_ns.create_jobid(&parent, NULL))) { /* got error */
+ fprintf(stderr, "create parent jobid: error with error %s\n", ORTE_ERROR_NAME(rc));
+ goto FINALIZE;
+ } else {
+ fprintf(stderr, "parent jobid created: %lu\n", (unsigned long) parent);
+ }
+
+ /* get range of vpids */
+ if (ORTE_SUCCESS != (rc = orte_ns.reserve_range(parent, 25, &vpid))) { /* got error */
+ fprintf(stderr, "reserve range: error with error %s\n",
+ ORTE_ERROR_NAME(rc));
+ goto FINALIZE;
+ } else {
+ fprintf(stderr, "range reserved: %lu\n", (unsigned long) vpid);
+ }
+
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+
+ for (i=0; i<5; i++) { /* loop through several vpid ranges */
+ if (0 == i) {
+ orte_rmgr.add_attribute(&attrs, ORTE_NS_USE_PARENT, ORTE_JOBID, &parent, ORTE_RMGR_ATTR_OVERRIDE);
+ } else if (2 == i) {
+ orte_rmgr.add_attribute(&attrs, ORTE_NS_USE_ROOT, ORTE_JOBID, &parent, ORTE_RMGR_ATTR_OVERRIDE);
+ } else {
+ orte_rmgr.add_attribute(&attrs, ORTE_NS_USE_PARENT, ORTE_JOBID, &jobs[i-1], ORTE_RMGR_ATTR_OVERRIDE);
+ }
+
+ if (ORTE_SUCCESS != (rc = orte_ns.create_jobid(&jobs[i], &attrs))) {
+ ORTE_ERROR_LOG(rc);
+ goto FINALIZE;
+ }
+ fprintf(stderr, "create jobid on step %d: jobid %ld\n", i, (long)jobs[i]);
+
+ /* get range of vpids */
+ if (ORTE_SUCCESS != (rc = orte_ns.reserve_range(jobs[i], 10*(i+1), &vpids[i]))) { /* got error */
+ fprintf(stderr, "reserve range: error with error %s\n",
+ ORTE_ERROR_NAME(rc));
+ goto FINALIZE;
+ } else {
+ fprintf(stderr, "range reserved: %lu\n", (unsigned long) vpids[i]);
+ }
+
+ orte_rmgr.delete_attribute(&attrs, ORTE_NS_USE_PARENT);
+ orte_rmgr.delete_attribute(&attrs, ORTE_NS_USE_ROOT);
+ }
+ while (NULL != (item = opal_list_remove_first(&attrs))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attrs);
+
+ orte_ns.dump_jobs();
+
+ /*** DESCENDANTS ***/
+ num_jobs = 0;
+ if (ORTE_SUCCESS != (rc = orte_ns.get_job_descendants(&jobdesc, &num_jobs, parent))) {
+ fprintf(stderr, "get job descendants: failed with error %s\n", ORTE_ERROR_NAME(rc));
+ goto FINALIZE;
+ }
+ for (j=0; j < num_jobs; j++) {
+ fprintf(stderr, "job descendants: job %ld\n", jobdesc[j]);
+ }
+ free(jobdesc);
+
+
+ /*** ROOT JOB ***/
+ if (ORTE_SUCCESS != (rc = orte_ns.get_root_job(&root, jobs[4]))) {
+ fprintf(stderr, "get root job: failed with error %s\n", ORTE_ERROR_NAME(rc));
+ goto FINALIZE;
+ }
+ fprintf(stderr, "got root job for job %ld - root was %ld\n", (long)jobs[4], root);
+
+
+ /*** PARENT JOB ***/
+ if (ORTE_SUCCESS != (rc = orte_ns.get_parent_job(&root, jobs[4]))) {
+ fprintf(stderr, "get parent job: failed with error %s\n", ORTE_ERROR_NAME(rc));
+ goto FINALIZE;
+ }
+ fprintf(stderr, "got parent job for job %ld - parent was %ld\n", (long)jobs[4], root);
+
+
+ /*** PEERS FUNCTIONS ***/
+ if (ORTE_SUCCESS != (rc = orte_ns.get_peers(&peers, &npeers, NULL))) {
+ fprintf(stderr, "get peers local: failed with error %s\n", ORTE_ERROR_NAME(rc));
+ goto FINALIZE;
+ }
+
+ jptr = peers;
+ for (j=0; j < npeers; j++) {
+ fprintf(stderr, "get peers local: peer %ld, %ld, %ld\n", ORTE_NAME_ARGS(jptr));
+ jptr++;
+ }
+ free(peers);
+
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ orte_rmgr.add_attribute(&attrs, ORTE_NS_USE_JOBID, ORTE_JOBID, &jobs[1], ORTE_RMGR_ATTR_OVERRIDE);
+
+ if (ORTE_SUCCESS != (rc = orte_ns.get_peers(&peers, &npeers, &attrs))) {
+ fprintf(stderr, "get peers for job %ld: failed with error %s\n", (long)jobs[1], ORTE_ERROR_NAME(rc));
+ goto FINALIZE;
+ }
+
+ jptr = peers;
+ for (j=0; j < npeers; j++) {
+ fprintf(stderr, "get peers for job %ld: peer %ld, %ld, %ld\n", (long)jobs[1], ORTE_NAME_ARGS(jptr));
+ jptr++;
+ }
+ if (NULL != peers) free(peers);
+ while (NULL != (item = opal_list_remove_first(&attrs))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attrs);
+
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ orte_rmgr.add_attribute(&attrs, ORTE_NS_USE_JOBID, ORTE_JOBID, &parent, ORTE_RMGR_ATTR_OVERRIDE);
+ orte_rmgr.add_attribute(&attrs, ORTE_NS_INCLUDE_DESCENDANTS, ORTE_UNDEF, NULL, ORTE_RMGR_ATTR_OVERRIDE);
+
+ if (ORTE_SUCCESS != (rc = orte_ns.get_peers(&peers, &npeers, &attrs))) {
+ fprintf(stderr, "get peers with descendants for job %ld: failed with error %s\n", (long)parent, ORTE_ERROR_NAME(rc));
+ goto FINALIZE;
+ }
+
+ jptr = peers;
+ for (j=0; j < npeers; j++) {
+ fprintf(stderr, "get peers with descendants for job %ld: peer %ld, %ld, %ld\n", (long)parent, ORTE_NAME_ARGS(jptr));
+ jptr++;
+ }
+ if (NULL != peers) free(peers);
+ while (NULL != (item = opal_list_remove_first(&attrs))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attrs);
+
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ orte_rmgr.add_attribute(&attrs, ORTE_NS_USE_JOBID, ORTE_JOBID, &parent, ORTE_RMGR_ATTR_OVERRIDE);
+ orte_rmgr.add_attribute(&attrs, ORTE_NS_INCLUDE_CHILDREN, ORTE_UNDEF, NULL, ORTE_RMGR_ATTR_OVERRIDE);
+
+ if (ORTE_SUCCESS != (rc = orte_ns.get_peers(&peers, &npeers, &attrs))) {
+ fprintf(stderr, "get peers job with children only: failed with error %s\n", ORTE_ERROR_NAME(rc));
+ goto FINALIZE;
+ }
+
+ jptr = peers;
+ for (j=0; j < npeers; j++) {
+ fprintf(stderr, "get peers with children only for job %ld: peer %ld, %ld, %ld\n", (long)parent, ORTE_NAME_ARGS(jptr));
+ jptr++;
+ }
+ if (NULL != peers) free(peers);
+ while (NULL != (item = opal_list_remove_first(&attrs))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attrs);
+
+FINALIZE:
+ /* finalize and see if memory cleared */
+ orte_ns_base_close();
+
+ orte_proc_info_finalize();
+ mca_base_close();
+ opal_malloc_finalize();
+ opal_output_finalize();
+
+ fclose( stderr );
+
+ return(0);
+}
diff --git a/orte/test/unit/ns/ns_string_fns.c b/orte/test/unit/ns/ns_string_fns.c
new file mode 100644
index 0000000000..8e206dd956
--- /dev/null
+++ b/orte/test/unit/ns/ns_string_fns.c
@@ -0,0 +1,157 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+ * University Research and Technology
+ * Corporation. All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ * of Tennessee Research Foundation. All rights
+ * reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ * University of Stuttgart. All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ * All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+
+#include "orte_config.h"
+#include "orte/orte_constants.h"
+
+#include <stdio.h>
+#include <string.h>
+
+#include "orte/mca/errmgr/errmgr.h"
+
+#include "orte/mca/ns/base/base.h"
+
+int main(int argc, char **argv)
+{
+ orte_process_name_t *test_name;
+ orte_cellid_t cell;
+ orte_jobid_t job;
+ orte_vpid_t vpid;
+ int i, j, rc;
+ char *tmp, *site, *resource;
+
+ if (ORTE_SUCCESS != orte_init(true)) {
+ fprintf(stderr, "failed to start ORTE\n");
+ exit (1);
+ }
+
+ /* create a name */
+ if (ORTE_SUCCESS != (rc = orte_ns.create_process_name(&test_name, 0, 1, 1))) { /* got error */
+ fprintf(stderr, "create process name failed with error %s\n",
+ ORTE_ERROR_NAME(rc));
+ exit(1);
+ } else {
+ fprintf(stderr, "got process name: %ld %ld %ld\n", ORTE_NAME_ARGS(test_name));
+ }
+ free(test_name);
+
+ /* convert a string to a name */
+ tmp = strdup("1234.5678.0010");
+ if (ORTE_SUCCESS != (rc = orte_ns.convert_string_to_process_name(&test_name, tmp))) { /* got error */
+ fprintf(stderr, "convert string to process name failed with error %s\n",
+ ORTE_ERROR_NAME(rc));
+ exit(1);
+ } else {
+ fprintf(stderr, "got process name: %ld %ld %ld\n", ORTE_NAME_ARGS(test_name));
+ }
+ free(tmp);
+ free(test_name);
+
+ /* create a cellid */
+ if (ORTE_SUCCESS != (rc = orte_ns.create_cellid(&cell, "dummy-site", "dummy-resource"))) { /* got error */
+ fprintf(stderr, "create cellid: error with error %s\n", ORTE_ERROR_NAME(rc));
+ exit(1);
+ } else {
+ fprintf(stderr, "cellid created: %lu\n", (unsigned long) cell);
+ }
+
+ /* get cellid info */
+ if (ORTE_SUCCESS != (rc = orte_ns.get_cell_info(cell, &site, &resource))) { /* got error */
+ fprintf(stderr, "get_cell_info: error with error %s\n", ORTE_ERROR_NAME(rc));
+ exit(1);
+ } else {
+ fprintf(stderr, "get_cell_info: %lu %s %s\n", (unsigned long) cell, site, resource);
+ }
+
+ for (i=0; i<10; i++) { /* loop through */
+ /* create jobid */
+ if (ORTE_SUCCESS != (rc = orte_ns.create_jobid(&job, NULL))) { /* got error */
+ fprintf(stderr, "create jobid: error with error %s\n", ORTE_ERROR_NAME(rc));
+ exit(1);
+ } else {
+ fprintf(stderr, "jobid created: %lu\n", (unsigned long) job);
+ }
+
+ for (j=0; j<5; j++) { /* loop through several vpid ranges */
+ /* get range of vpids */
+ if (ORTE_SUCCESS != (rc = orte_ns.reserve_range(job, 250, &vpid))) { /* got error */
+ fprintf(stderr, "reserve range: error with error %s\n",
+ ORTE_ERROR_NAME(rc));
+ exit(1);
+ } else {
+ fprintf(stderr, "range reserved: %lu\n",
+ (unsigned long) vpid);
+ }
+
+ /* create a name */
+ if (ORTE_SUCCESS != (rc = orte_ns.create_process_name(&test_name, (orte_cellid_t)i,
+ job, vpid))) {
+ fprintf(stderr, "test_ns_replica: failed to create proc name after vpid range with error %s\n",
+ ORTE_ERROR_NAME(rc));
+ exit(1);
+ }
+
+ /* get and print its string values */
+ if (ORTE_SUCCESS != (rc = orte_ns.get_proc_name_string(&tmp, test_name))) {
+ fprintf(stderr, "test_ns_replica: failed to get proc_name_string with error %s\n",
+ ORTE_ERROR_NAME(rc));
+ exit(1);
+ } else {
+ fprintf(stderr, "(%d) strings: name - %s\n", i, tmp);
+ }
+ free(tmp);
+ if (ORTE_SUCCESS != (rc = orte_ns.get_vpid_string(&tmp, test_name))) {
+ fprintf(stderr, "test_ns_replica: failed to get vpid_string with error %s\n",
+ ORTE_ERROR_NAME(rc));
+ exit(1);
+ } else {
+ fprintf(stderr, "(%d) strings: vpid - %s\n", i, tmp);
+ }
+ free(tmp);
+ if (ORTE_SUCCESS != (rc = orte_ns.get_jobid_string(&tmp, test_name))) {
+ fprintf(stderr, "test_ns_replica: failed to get jobid_string with error %s\n",
+ ORTE_ERROR_NAME(rc));
+ exit(1);
+ } else {
+ fprintf(stderr, "(%d) strings: jobid - %s\n", i, tmp);
+ }
+ free(tmp);
+ if (ORTE_SUCCESS != (rc = orte_ns.get_cellid_string(&tmp, test_name))) {
+ fprintf(stderr, "test_ns_replica: failed to get cellid_string with error %s\n",
+ ORTE_ERROR_NAME(rc));
+ exit(1);
+ } else {
+ fprintf(stderr, "(%d) strings: cellid - %s\n", i, tmp);
+ }
+ free(tmp);
+
+ }
+ }
+
+ /* finalize and see if memory cleared */
+ orte_ns_base_close();
+
+ orte_proc_info_finalize();
+ mca_base_close();
+ opal_malloc_finalize();
+ opal_output_finalize();
+
+ fclose( stderr );
+
+ return(0);
+}
diff --git a/orte/tools/orted/orted.c b/orte/tools/orted/orted.c
index 347aadfe95..2105207d79 100644
--- a/orte/tools/orted/orted.c
+++ b/orte/tools/orted/orted.c
@@ -379,12 +379,12 @@ int main(int argc, char *argv[])
OBJ_CONSTRUCT(&orted_globals.condition, opal_condition_t);
/* register the daemon main receive functions */
- ret = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY, ORTE_RML_TAG_PLS_ORTED, ORTE_RML_NON_PERSISTENT, orte_daemon_recv_pls, NULL);
+ ret = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_PLS_ORTED, ORTE_RML_NON_PERSISTENT, orte_daemon_recv_pls, NULL);
if (ret != ORTE_SUCCESS && ret != ORTE_ERR_NOT_IMPLEMENTED) {
ORTE_ERROR_LOG(ret);
return ret;
}
- ret = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY, ORTE_RML_TAG_DAEMON, ORTE_RML_NON_PERSISTENT, orte_daemon_recv, NULL);
+ ret = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DAEMON, ORTE_RML_NON_PERSISTENT, orte_daemon_recv, NULL);
if (ret != ORTE_SUCCESS && ret != ORTE_ERR_NOT_IMPLEMENTED) {
ORTE_ERROR_LOG(ret);
return ret;
@@ -693,7 +693,7 @@ static void orte_daemon_recv_pls(int status, orte_process_name_t* sender,
OPAL_THREAD_UNLOCK(&orted_globals.mutex);
/* reissue the non-blocking receive */
- ret = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY, ORTE_RML_TAG_PLS_ORTED, ORTE_RML_NON_PERSISTENT, orte_daemon_recv_pls, NULL);
+ ret = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_PLS_ORTED, ORTE_RML_NON_PERSISTENT, orte_daemon_recv_pls, NULL);
if (ret != ORTE_SUCCESS && ret != ORTE_ERR_NOT_IMPLEMENTED) {
ORTE_ERROR_LOG(ret);
}
@@ -789,7 +789,7 @@ DONE:
OPAL_THREAD_UNLOCK(&orted_globals.mutex);
/* reissue the non-blocking receive */
- ret = orte_rml.recv_buffer_nb(ORTE_RML_NAME_ANY, ORTE_RML_TAG_DAEMON, ORTE_RML_NON_PERSISTENT, orte_daemon_recv, NULL);
+ ret = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DAEMON, ORTE_RML_NON_PERSISTENT, orte_daemon_recv, NULL);
if (ret != ORTE_SUCCESS && ret != ORTE_ERR_NOT_IMPLEMENTED) {
ORTE_ERROR_LOG(ret);
}
diff --git a/orte/tools/orteprobe/orteprobe.c b/orte/tools/orteprobe/orteprobe.c
index 09576826c6..920af1bf92 100644
--- a/orte/tools/orteprobe/orteprobe.c
+++ b/orte/tools/orteprobe/orteprobe.c
@@ -175,7 +175,7 @@ int main(int argc, char *argv[])
* Attempt to parse the probe's name and save in proc_info
*/
if (orteprobe_globals.name_string) {
- ret = orte_ns_base_convert_string_to_process_name(
+ ret = orte_ns.convert_string_to_process_name(
&orte_process_info.my_name, orteprobe_globals.name_string);
if(ORTE_SUCCESS != ret) {
fprintf(stderr, "orteprobe: Couldn't convert environmental string to probe's process name\n");
diff --git a/orte/tools/orterun/orterun.c b/orte/tools/orterun/orterun.c
index b031f802c2..3adde8e905 100644
--- a/orte/tools/orterun/orterun.c
+++ b/orte/tools/orterun/orterun.c
@@ -301,6 +301,7 @@ int orterun(int argc, char *argv[])
orte_proc_state_t cb_states;
orte_job_state_t exit_state;
opal_list_t attributes;
+ opal_list_item_t *item;
/* Setup MCA params */
@@ -458,6 +459,9 @@ int orterun(int argc, char *argv[])
cb_states = ORTE_PROC_STATE_TERMINATED | ORTE_PROC_STATE_AT_STG1;
rc = orte_rmgr.spawn_job(apps, num_apps, &jobid, 0, NULL, job_state_callback, cb_states, &attributes);
+ while (NULL != (item = opal_list_remove_first(&attributes))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attributes);
+
if (ORTE_SUCCESS != rc) {
/* JMS show_help */
opal_output(0, "%s: spawn failed with errno=%d\n", orterun_basename, rc);
@@ -505,11 +509,17 @@ int orterun(int argc, char *argv[])
/* the job is complete - now tell the orteds that it is
* okay to finalize and exit, we are done with them
+ * be sure to include any descendants so nothing is
+ * left hanging
*/
- if (ORTE_SUCCESS != (ret = orte_pls.terminate_orteds(jobid))) {
+ OBJ_CONSTRUCT(&attributes, opal_list_t);
+ orte_rmgr.add_attribute(&attributes, ORTE_NS_INCLUDE_DESCENDANTS, ORTE_UNDEF, NULL, ORTE_RMGR_ATTR_OVERRIDE);
+ if (ORTE_SUCCESS != (ret = orte_pls.terminate_orteds(jobid, &attributes))) {
opal_show_help("help-orterun.txt", "orterun:daemon-die", false,
orterun_basename, NULL, NULL, ret);
}
+ while (NULL != (item = opal_list_remove_first(&attributes))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attributes);
OPAL_THREAD_UNLOCK(&orterun_globals.lock);
}
@@ -744,6 +754,8 @@ static void abort_signal_callback(int fd, short flags, void *arg)
int ret;
struct timeval tv = { 1, 0 };
opal_event_t* event;
+ opal_list_t attrs;
+ opal_list_item_t *item;
static int signalled = 0;
@@ -757,10 +769,15 @@ static void abort_signal_callback(int fd, short flags, void *arg)
}
/* terminate the job - this will also wakeup orterun so
- * it can kill all the orteds
+ * it can kill all the orteds. Be sure to kill all the job's
+ * descendants, if any, so nothing is left hanging
*/
if (jobid != ORTE_JOBID_INVALID) {
- ret = orte_pls.terminate_job(jobid);
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ orte_rmgr.add_attribute(&attrs, ORTE_NS_INCLUDE_DESCENDANTS, ORTE_UNDEF, NULL, ORTE_RMGR_ATTR_OVERRIDE);
+ ret = orte_pls.terminate_job(jobid, &attrs);
+ while (NULL != (item = opal_list_remove_first(&attrs))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attrs);
if (ORTE_SUCCESS != ret) {
jobid = ORTE_JOBID_INVALID;
}
@@ -787,6 +804,8 @@ static void signal_forward_callback(int fd, short event, void *arg)
{
struct opal_event *signal = (struct opal_event*)arg;
int signum, ret;
+ opal_list_t attrs;
+ opal_list_item_t *item;
OPAL_TRACE(1);
@@ -796,11 +815,15 @@ static void signal_forward_callback(int fd, short event, void *arg)
orterun_basename, signum);
}
- /** send the signal out to the processes */
- if (ORTE_SUCCESS != (ret = orte_pls.signal_job(jobid, signum))) {
+ /** send the signal out to the processes, including any descendants */
+ OBJ_CONSTRUCT(&attrs, opal_list_t);
+ orte_rmgr.add_attribute(&attrs, ORTE_NS_INCLUDE_DESCENDANTS, ORTE_UNDEF, NULL, ORTE_RMGR_ATTR_OVERRIDE);
+ if (ORTE_SUCCESS != (ret = orte_pls.signal_job(jobid, signum, &attrs))) {
fprintf(stderr, "Signal %d could not be sent to the job (returned %d)",
signum, ret);
}
+ while (NULL != (item = opal_list_remove_first(&attrs))) OBJ_RELEASE(item);
+ OBJ_DESTRUCT(&attrs);
}