ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Remove the orte notifier framework, per discussion at the devel meeting and follow-up with Jeff (who took the action item)

Просмотр исходного кода 
This commit was SVN r26637.
Этот коммит содержится в:
Ralph Castain 2012-06-22 18:09:23 +00:00
родитель 03f00c42b8
Коммит e6f3586415
76 изменённых файлов: 27 добавлений и 5886 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

20
ompi/mca/btl/openib/btl_openib_component.c
Просмотреть файл

@ -74,7 +74,6 @@ const char *ibv_get_sysfs_path(void);
#include "orte/util/show_help.h"
#include "orte/util/proc_info.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/notifier/notifier.h"
#include "ompi/constants.h"
#include "ompi/proc/proc.h"
@ -3567,13 +3566,6 @@ error:
cq_name[cq], btl_openib_component_status_to_string(wc->status),
wc->status, wc->wr_id,
wc->opcode, wc->vendor_err, qp));
orte_notifier.log_peer(ORTE_NOTIFIER_CRIT, ORTE_ERR_COMM_FAILURE,
remote_proc ? &remote_proc->proc_name : NULL,
"\n\tIB polling %s with status %s "
"status number %d for wr_id %llu opcode %d vendor error %d qp_idx %d",
cq_name[cq], btl_openib_component_status_to_string(wc->status),
wc->status, wc->wr_id,
wc->opcode, wc->vendor_err, qp);
}
if (IBV_WC_RNR_RETRY_EXC_ERR == wc->status ||
@ -3592,23 +3584,11 @@ error:
"srq rnr retry exceeded", true,
orte_process_info.nodename, device_name,
peer_hostname);
orte_notifier.show_help(ORTE_NOTIFIER_CRIT, ORTE_ERR_COMM_FAILURE,
"help-mpi-btl-openib.txt",
BTL_OPENIB_QP_TYPE_PP(qp) ?
"pp rnr retry exceeded" :
"srq rnr retry exceeded",
orte_process_info.nodename, device_name,
peer_hostname);
} else if (IBV_WC_RETRY_EXC_ERR == wc->status) {
orte_show_help("help-mpi-btl-openib.txt",
"pp retry exceeded", true,
orte_process_info.nodename,
device_name, peer_hostname);
orte_notifier.show_help(ORTE_NOTIFIER_CRIT, ORTE_ERR_COMM_FAILURE,
"help-mpi-btl-openib.txt",
"pp retry exceeded",
orte_process_info.nodename,
device_name, peer_hostname);
}
}

20
ompi/mca/btl/wv/btl_wv_component.c
Просмотреть файл

@ -48,7 +48,6 @@
#include "orte/util/show_help.h"
#include "orte/util/proc_info.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/notifier/notifier.h"
#include "ompi/constants.h"
#include "ompi/proc/proc.h"
@ -2963,13 +2962,6 @@ error:
cq_name[cq], btl_wv_component_status_to_string(wc->Status),
wc->Status, wc->WrId,
wc->Opcode, wc->VendorCode, qp));
orte_notifier.log_peer(ORTE_NOTIFIER_CRIT, ORTE_ERR_COMM_FAILURE,
remote_proc ? &remote_proc->proc_name : NULL,
"\n\tIB polling %s with status %s "
"status number %d for wr_id %llu opcode %d vendor error %d qp_idx %d",
cq_name[cq], btl_wv_component_status_to_string(wc->Status),
wc->Status, wc->WrId,
wc->Opcode, wc->VendorCode, qp);
}
if (WvWcRnrRetryError == wc->Status ||
@ -2988,23 +2980,11 @@ error:
"srq rnr retry exceeded", true,
orte_process_info.nodename, device_name,
peer_hostname);
orte_notifier.show_help(ORTE_NOTIFIER_CRIT, ORTE_ERR_COMM_FAILURE,
"help-mpi-btl-wv.txt",
BTL_WV_QP_TYPE_PP(qp) ?
"pp rnr retry exceeded" :
"srq rnr retry exceeded",
orte_process_info.nodename, device_name,
peer_hostname);
} else if (-2 == wc->Status) {
orte_show_help("help-mpi-btl-wv.txt",
"pp retry exceeded", true,
orte_process_info.nodename,
device_name, peer_hostname);
orte_notifier.show_help(ORTE_NOTIFIER_CRIT, ORTE_ERR_COMM_FAILURE,
"help-mpi-btl-wv.txt",
"pp retry exceeded",
orte_process_info.nodename,
device_name, peer_hostname);
}
}

11
ompi/mca/pml/bfo/pml_bfo_failover.c
Просмотреть файл

@ -45,7 +45,6 @@
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/grpcomm/grpcomm.h"
#include "orte/util/show_help.h"
#include "orte/mca/notifier/notifier.h"
#include "ompi/runtime/ompi_cr.h"
@ -1413,15 +1412,7 @@ void mca_pml_bfo_map_out_btl(struct mca_btl_base_module_t* btl,
if (true == remove) {
mca_bml.bml_del_proc_btl(errproc, btl);
orte_notifier.log(ORTE_NOTIFIER_ERROR, ORTE_ERR_COMM_FAILURE,
"BTL %s error: rank=%d mapping out %s "
"to rank=%d on node=%s",
btl->btl_component->btl_version.mca_component_name,
ORTE_PROC_MY_NAME->vpid,
btlname, errproc->proc_name.vpid,
errproc->proc_hostname);
opal_output_verbose(10, mca_pml_bfo_output,
opal_output_verbose(10, mca_pml_bfo_output,
"BTL %s error: rank=%d mapping out %s "
"to rank=%d on node=%s \n",
btl->btl_component->btl_version.mca_component_name,

29
ompi/mca/pml/csum/pml_csum_recvfrag.c
Просмотреть файл

@ -37,7 +37,6 @@
#include "orte/util/name_fns.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/errmgr/errmgr.h"
#include "ompi/constants.h"
@ -172,10 +171,6 @@ void mca_pml_csum_recv_frag_callback_match(mca_btl_base_module_t* btl,
if (csum_received != csum) {
opal_output(0, "%s:%s:%d: Invalid \'match header\' - received csum:0x%04x != computed csum:0x%04x\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), __FILE__, __LINE__, csum_received, csum);
orte_notifier.log(ORTE_NOTIFIER_CRIT, 1,
"Checksum header violation: job %s file %s line %d",
(NULL == orte_job_ident) ? "UNKNOWN" : orte_job_ident,
__FILE__, __LINE__);
dump_csum_error_data(segments, 1);
orte_errmgr.abort(-1,NULL);
}
@ -303,10 +298,6 @@ void mca_pml_csum_recv_frag_callback_match(mca_btl_base_module_t* btl,
if (csum_data != hdr->hdr_csum) {
opal_output(0, "%s:%s:%d: Invalid \'match data\' - received csum:0x%x != computed csum:0x%x\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), __FILE__, __LINE__, hdr->hdr_csum, csum_data);
orte_notifier.log(ORTE_NOTIFIER_CRIT, 1,
"Checksum data violation: job %s file %s line %d",
(NULL == orte_job_ident) ? "UNKNOWN" : orte_job_ident,
__FILE__, __LINE__);
dump_csum_error_data(segments, num_segments);
orte_errmgr.abort(-1,NULL);
}
@ -349,10 +340,6 @@ void mca_pml_csum_recv_frag_callback_rndv(mca_btl_base_module_t* btl,
if (csum_received != csum) {
opal_output(0, "%s:%s:%d: Invalid \'rndv header\' - received csum:0x%04x != computed csum:0x%04x\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), __FILE__, __LINE__, csum_received, csum);
orte_notifier.log(ORTE_NOTIFIER_CRIT, 1,
"Checksum header violation: job %s file %s line %d",
(NULL == orte_job_ident) ? "UNKNOWN" : orte_job_ident,
__FILE__, __LINE__);
dump_csum_error_data(segments, 1);
orte_errmgr.abort(-1,NULL);
}
@ -409,10 +396,6 @@ void mca_pml_csum_recv_frag_callback_ack(mca_btl_base_module_t* btl,
if (csum_received != csum) {
opal_output(0, "%s:%s:%d: Invalid \'ACK header\' - received csum:0x%04x != computed csum:0x%04x\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), __FILE__, __LINE__, csum_received, csum);
orte_notifier.log(ORTE_NOTIFIER_CRIT, 1,
"Checksum header violation: job %s file %s line %d",
(NULL == orte_job_ident) ? "UNKNOWN" : orte_job_ident,
__FILE__, __LINE__);
dump_csum_error_data(segments, 1);
orte_errmgr.abort(-1,NULL);
}
@ -473,10 +456,6 @@ void mca_pml_csum_recv_frag_callback_frag(mca_btl_base_module_t* btl,
if(csum_received != csum) {
opal_output(0, "%s:%s:%d: Invalid \'frag header\' - received csum:0x%04x != computed csum:0x%04x\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), __FILE__, __LINE__, csum_received, csum);
orte_notifier.log(ORTE_NOTIFIER_CRIT, 1,
"Checksum header violation: job %s file %s line %d",
(NULL == orte_job_ident) ? "UNKNOWN" : orte_job_ident,
__FILE__, __LINE__);
dump_csum_error_data(segments, 1);
orte_errmgr.abort(-1,NULL);
}
@ -515,10 +494,6 @@ void mca_pml_csum_recv_frag_callback_put(mca_btl_base_module_t* btl,
if(csum_received != csum) {
opal_output(0, "%s:%s:%d: Invalid \'PUT header\' - received csum:0x%04x != computed csum:0x%04x\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), __FILE__, __LINE__, csum_received, csum);
orte_notifier.log(ORTE_NOTIFIER_CRIT, 1,
"Checksum header violation: job %s file %s line %d",
(NULL == orte_job_ident) ? "UNKNOWN" : orte_job_ident,
__FILE__, __LINE__);
dump_csum_error_data(segments, 1);
orte_errmgr.abort(-1,NULL);
}
@ -557,10 +532,6 @@ void mca_pml_csum_recv_frag_callback_fin(mca_btl_base_module_t* btl,
if(csum_received != csum) {
opal_output(0, "%s:%s:%d: Invalid \'FIN header\' - received csum:0x%04x != computed csum:0x%04x\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), __FILE__, __LINE__, csum_received, csum);
orte_notifier.log(ORTE_NOTIFIER_CRIT, 1,
"Checksum header violation: job %s file %s line %d",
(NULL == orte_job_ident) ? "UNKNOWN" : orte_job_ident,
__FILE__, __LINE__);
dump_csum_error_data(segments, 1);
orte_errmgr.abort(-1,NULL);
}

13
ompi/mca/pml/csum/pml_csum_recvreq.c
Просмотреть файл

@ -32,7 +32,6 @@
#include "orte/mca/errmgr/errmgr.h"
#include "orte/util/name_fns.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/notifier/notifier.h"
#include "ompi/mca/pml/pml.h"
#include "ompi/mca/bml/bml.h"
@ -498,10 +497,6 @@ void mca_pml_csum_recv_request_progress_frag( mca_pml_csum_recv_request_t* recvr
if(csum != hdr->hdr_frag.hdr_csum) {
opal_output(0, "%s:%s:%d: Invalid \'frag data\' - received csum:0x%x != computed csum:0x%x\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), __FILE__, __LINE__, hdr->hdr_frag.hdr_csum, csum);
orte_notifier.log(ORTE_NOTIFIER_CRIT, 1,
"Checksum data violation: job %s file %s line %d",
(NULL == orte_job_ident) ? "UNKNOWN" : orte_job_ident,
__FILE__, __LINE__);
dump_csum_error_data(segments, num_segments);
orte_errmgr.abort(-1,NULL);
}
@ -645,10 +640,6 @@ void mca_pml_csum_recv_request_progress_rndv( mca_pml_csum_recv_request_t* recvr
if (csum != hdr->hdr_match.hdr_csum) {
opal_output(0, "%s:%s:%d: Invalid \'rndv data\' - received csum:0x%x != computed csum:0x%x\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), __FILE__, __LINE__, hdr->hdr_match.hdr_csum, csum);
orte_notifier.log(ORTE_NOTIFIER_CRIT, 1,
"Checksum data violation: job %s file %s line %d",
(NULL == orte_job_ident) ? "UNKNOWN" : orte_job_ident,
__FILE__, __LINE__);
dump_csum_error_data(segments, num_segments);
orte_errmgr.abort(-1,NULL);
}
@ -706,10 +697,6 @@ void mca_pml_csum_recv_request_progress_match( mca_pml_csum_recv_request_t* recv
if (csum != hdr->hdr_match.hdr_csum) {
opal_output(0, "%s:%s:%d: Invalid \'match data\' - received csum:0x%x != computed csum:0x%x\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), __FILE__, __LINE__, hdr->hdr_match.hdr_csum, csum);
orte_notifier.log(ORTE_NOTIFIER_CRIT, 1,
"Checksum data violation: job %s file %s line %d",
(NULL == orte_job_ident) ? "UNKNOWN" : orte_job_ident,
__FILE__, __LINE__);
dump_csum_error_data(segments, num_segments);
orte_errmgr.abort(-1,NULL);
}

22
ompi/runtime/ompi_mpi_finalize.c
Просмотреть файл

@ -49,7 +49,6 @@
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/grpcomm/grpcomm.h"
#include "orte/mca/notifier/notifier.h"
#include "orte/runtime/runtime.h"
#include "orte/runtime/orte_globals.h"
@ -122,17 +121,6 @@ int ompi_mpi_finalize(void)
*/
(void)mca_pml_base_bsend_detach(NULL, NULL);
#if !ORTE_DISABLE_FULL_SUPPORT
/* If desired, send a notify message */
if (ompi_notify_init_finalize) {
orte_notifier.log(ORTE_NOTIFIER_NOTICE,
ORTE_SUCCESS,
"MPI_FINALIZE:Starting on host %s, pid %d",
orte_process_info.nodename,
orte_process_info.pid);
}
#endif
/* Per MPI-2:4.8, we have to free MPI_COMM_SELF before doing
anything else in MPI_FINALIZE (to include setting up such that
MPI_FINALIZED will return true). */
@ -424,16 +412,6 @@ int ompi_mpi_finalize(void)
free(ompi_mpi_show_mca_params_file);
}
#if !ORTE_DISABLE_FULL_SUPPORT
/* If desired, send a notify message */
if (ompi_notify_init_finalize) {
orte_notifier.log(ORTE_NOTIFIER_NOTICE,
ORTE_SUCCESS,
"MPI_FINALIZE:Finishing on host %s, pid %d",
orte_process_info.nodename,
orte_process_info.pid);
}
#endif
/* Leave the RTE */

23
ompi/runtime/ompi_mpi_init.c
Просмотреть файл

@ -57,7 +57,6 @@
#include "orte/mca/odls/base/base.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/util/name_fns.h"
#include "orte/mca/notifier/notifier.h"
#include "ompi/constants.h"
#include "ompi/mpi/fortran/base/constants.h"
@ -467,17 +466,6 @@ int ompi_mpi_init(int argc, char **argv, int requested, int *provided)
goto error;
}
#if !ORTE_DISABLE_FULL_SUPPORT
/* If desired, send a notify message */
if (ompi_notify_init_finalize) {
orte_notifier.log(ORTE_NOTIFIER_NOTICE,
ORTE_SUCCESS,
"MPI_INIT:Starting on host %s, pid %d",
orte_process_info.nodename,
orte_process_info.pid);
}
#endif
/* initialize datatypes. This step should be done early as it will
* create the local convertor and local arch used in the proc
* init.
@ -966,16 +954,5 @@ int ompi_mpi_init(int argc, char **argv, int requested, int *provided)
(ompistop.tv_usec - ompistart.tv_usec)));
}
#if !ORTE_DISABLE_FULL_SUPPORT
/* If desired, send a notifier message that we've finished MPI_INIT */
if (ompi_notify_init_finalize) {
orte_notifier.log(ORTE_NOTIFIER_NOTICE,
ORTE_SUCCESS,
"MPI_INIT:Finishing on host %s, pid %d",
orte_process_info.nodename,
orte_process_info.pid);
}
#endif
return MPI_SUCCESS;
}

9
ompi/runtime/ompi_mpi_params.c
Просмотреть файл

@ -60,7 +60,6 @@ int ompi_mpi_leave_pinned = -1;
bool ompi_mpi_leave_pinned_pipeline = false;
bool ompi_have_sparse_group_storage = OPAL_INT_TO_BOOL(OMPI_GROUP_SPARSE);
bool ompi_use_sparse_group_storage = OPAL_INT_TO_BOOL(OMPI_GROUP_SPARSE);
bool ompi_notify_init_finalize = true;
static bool show_default_mca_params = false;
static bool show_file_mca_params = false;
@ -287,14 +286,6 @@ int ompi_mpi_register_params(void)
}
}
/* Do we want notifier messages upon MPI_INIT and MPI_FINALIZE? */
mca_base_param_reg_int_name("mpi", "notify_init_finalize",
"If nonzero, send two notifications during MPI_INIT: one near when MPI_INIT starts, and another right before MPI_INIT finishes, and send 2 notifications during MPI_FINALIZE: one right when MPI_FINALIZE starts, and another near when MPI_FINALIZE finishes.",
false, false,
(int) ompi_notify_init_finalize, &value);
ompi_notify_init_finalize = OPAL_INT_TO_BOOL(value);
return OMPI_SUCCESS;
}

6
ompi/runtime/params.h
Просмотреть файл

@ -138,12 +138,6 @@ OMPI_DECLSPEC extern bool ompi_have_sparse_group_storage;
*/
OMPI_DECLSPEC extern bool ompi_use_sparse_group_storage;
/**
* Whether we should send notifications during MPI_INIT and
* MPI_FINALIZE or not.
*/
OMPI_DECLSPEC extern bool ompi_notify_init_finalize;
/**
* Register MCA parameters used by the MPI layer.
*

11
ompi/tools/ompi_info/components.c
Просмотреть файл

@ -106,8 +106,6 @@
#include "orte/util/show_help.h"
#include "orte/util/proc_info.h"
#if !ORTE_DISABLE_FULL_SUPPORT
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/notifier/base/base.h"
#include "orte/mca/iof/iof.h"
#include "orte/mca/iof/base/base.h"
#include "orte/mca/oob/oob.h"
@ -411,14 +409,6 @@ void ompi_info_open_components(void)
opal_pointer_array_add(&component_map, map);
#if !ORTE_DISABLE_FULL_SUPPORT
if (ORTE_SUCCESS != orte_notifier_base_open()) {
goto error;
}
map = OBJ_NEW(ompi_info_component_map_t);
map->type = strdup("notifier");
map->components = &orte_notifier_base_components_available;
opal_pointer_array_add(&component_map, map);
if (ORTE_SUCCESS != mca_oob_base_open()) {
goto error;
}
@ -767,7 +757,6 @@ void ompi_info_close_components()
(void) orte_snapc_base_close();
(void) orte_sstore_base_close();
#endif
(void) orte_notifier_base_close();
(void) orte_filem_base_close();
(void) orte_iof_base_close();
(void) orte_plm_base_close();

1
ompi/tools/ompi_info/ompi_info.c
Просмотреть файл

@ -270,7 +270,6 @@ int main(int argc, char *argv[])
opal_pointer_array_add(&mca_types, "ess");
opal_pointer_array_add(&mca_types, "grpcomm");
opal_pointer_array_add(&mca_types, "db");
opal_pointer_array_add(&mca_types, "notifier");
/* Execute the desired action(s) */

3
ompi/tools/ompi_info/param.c
Просмотреть файл

@ -532,7 +532,6 @@ void ompi_info_do_config(bool want_all)
char *memprofile;
char *memdebug;
char *debug;
char *log_event;
char *mpi_interface_warning;
char *cprofiling;
char *cxxprofiling;
@ -649,7 +648,6 @@ void ompi_info_do_config(bool want_all)
memprofile = OPAL_ENABLE_MEM_PROFILE ? "yes" : "no";
memdebug = OPAL_ENABLE_MEM_DEBUG ? "yes" : "no";
debug = OPAL_ENABLE_DEBUG ? "yes" : "no";
log_event = ORTE_WANT_NOTIFIER_LOG_EVENT ? "yes" : "no";
mpi_interface_warning = OMPI_WANT_MPI_INTERFACE_WARNING ? "yes" : "no";
cprofiling = OMPI_ENABLE_MPI_PROFILING ? "yes" : "no";
cxxprofiling = (OMPI_BUILD_CXX_BINDINGS && OMPI_ENABLE_MPI_PROFILING) ? "yes" : "no";
@ -982,7 +980,6 @@ void ompi_info_do_config(bool want_all)
ompi_info_out("MPI parameter check", "option:mpi-param-check", paramcheck);
ompi_info_out("Memory profiling support", "option:mem-profile", memprofile);
ompi_info_out("Memory debugging support", "option:mem-debug", memdebug);
ompi_info_out("Unusual events notif.", "option:log-event", log_event);
ompi_info_out("libltdl support", "option:dlopen", want_libltdl);
ompi_info_out("Heterogeneous support", "options:heterogeneous", heterogeneous);
ompi_info_out("mpirun default --prefix", "mpirun:prefix_by_default",

32
orte/mca/errmgr/base/errmgr_base_fns.c
Просмотреть файл

@ -11,7 +11,7 @@
* All rights reserved.
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2010-2011 Oak Ridge National Labs. All rights reserved.
* Copyright (c) 2011 Los Alamos National Security, LLC.
* Copyright (c) 2011-2012 Los Alamos National Security, LLC.
* All rights reserved.
* $COPYRIGHT$
*
@ -81,7 +81,6 @@
#include "orte/mca/snapc/base/base.h"
#include "orte/mca/sstore/sstore.h"
#include "orte/mca/sstore/base/base.h"
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/errmgr/base/base.h"
@ -300,14 +299,12 @@ void orte_errmgr_base_migrate_state_notify(int state)
switch(state) {
case ORTE_ERRMGR_MIGRATE_STATE_ERROR:
case ORTE_ERRMGR_MIGRATE_STATE_ERR_INPROGRESS:
orte_notifier.log(ORTE_NOTIFIER_ERROR, state,
"%d: Migration failed for process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
opal_output(0, "%d: Migration failed for process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
break;
case ORTE_ERRMGR_MIGRATE_STATE_FINISH:
orte_notifier.log(ORTE_NOTIFIER_INFO, state,
"%d: Migration successful for process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
opal_output(0, "%d: Migration successful for process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
break;
case ORTE_ERRMGR_MIGRATE_STATE_NONE:
@ -331,26 +328,23 @@ void orte_errmgr_base_proc_state_notify(orte_proc_state_t state, orte_process_na
case ORTE_PROC_STATE_TERMINATED:
case ORTE_PROC_STATE_KILLED_BY_CMD:
case ORTE_PROC_STATE_SENSOR_BOUND_EXCEEDED:
orte_notifier.log(ORTE_NOTIFIER_ERROR, state, "%d: Process %s is dead.",
orte_process_info.pid, ORTE_JOBID_PRINT(proc->jobid));
opal_output(0,, "%d: Process %s is dead.",
orte_process_info.pid, ORTE_JOBID_PRINT(proc->jobid));
break;
case ORTE_PROC_STATE_HEARTBEAT_FAILED:
orte_notifier.log(ORTE_NOTIFIER_ERROR, state,
"%d: Process %s is unreachable.",
orte_process_info.pid, ORTE_JOBID_PRINT(proc->jobid));
opal_output(0, "%d: Process %s is unreachable.",
orte_process_info.pid, ORTE_JOBID_PRINT(proc->jobid));
case ORTE_PROC_STATE_COMM_FAILED:
orte_notifier.log(ORTE_NOTIFIER_WARN, state,
"%d: Failed to communicate with process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(proc->jobid));
opal_output(0, "%d: Failed to communicate with process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(proc->jobid));
break;
case ORTE_PROC_STATE_CALLED_ABORT:
case ORTE_PROC_STATE_FAILED_TO_START:
orte_notifier.log(ORTE_NOTIFIER_ERROR, state,
"%d: Process %s has called abort.",
orte_process_info.pid, ORTE_JOBID_PRINT(proc->jobid));
opal_output(0, "%d: Process %s has called abort.",
orte_process_info.pid, ORTE_JOBID_PRINT(proc->jobid));
break;
case ORTE_PROC_STATE_MIGRATING:
default:

1
orte/mca/errmgr/default_hnp/errmgr_default_hnp.c
Просмотреть файл

@ -41,7 +41,6 @@
#include "orte/mca/rmaps/rmaps_types.h"
#include "orte/mca/sensor/sensor.h"
#include "orte/mca/routed/routed.h"
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/grpcomm/grpcomm.h"
#include "orte/mca/ess/ess.h"
#include "orte/mca/state/state.h"

17
orte/mca/ess/base/ess_base_std_app.c
Просмотреть файл

@ -56,7 +56,6 @@
#include "orte/util/session_dir.h"
#include "orte/util/name_fns.h"
#include "orte/util/show_help.h"
#include "orte/mca/notifier/base/base.h"
#include "orte/runtime/orte_cr.h"
#include "orte/runtime/orte_globals.h"
@ -251,18 +250,6 @@ int orte_ess_base_app_setup(void)
goto error;
}
/* setup the notifier system */
if (ORTE_SUCCESS != (ret = orte_notifier_base_open())) {
ORTE_ERROR_LOG(ret);
error = "orte_notifer_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_notifier_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orte_notifer_select";
goto error;
}
/* if we are an ORTE app - and not an MPI app - then
* we need to barrier here. MPI_Init has its own barrier,
* so we don't need to do two of them. However, if we
@ -299,9 +286,7 @@ error:
}
int orte_ess_base_app_finalize(void)
{
orte_notifier_base_close();
{
orte_cr_finalize();
#if OPAL_ENABLE_FT_CR == 1

13
orte/mca/ess/base/ess_base_std_orted.c
Просмотреть файл

@ -62,7 +62,6 @@
#include "orte/util/regex.h"
#include "orte/util/show_help.h"
#include "orte/mca/errmgr/base/base.h"
#include "orte/mca/notifier/base/base.h"
#include "orte/mca/sensor/base/base.h"
#include "orte/mca/sensor/sensor.h"
#include "orte/mca/state/base/base.h"
@ -570,18 +569,6 @@ int orte_ess_base_orted_setup(char **hosts)
goto error;
}
/* setup the notifier system */
if (ORTE_SUCCESS != (ret = orte_notifier_base_open())) {
ORTE_ERROR_LOG(ret);
error = "orte_notifer_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_notifier_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orte_notifer_select";
goto error;
}
/* setup the SENSOR framework */
if (ORTE_SUCCESS != (ret = orte_sensor_base_open())) {
ORTE_ERROR_LOG(ret);

13
orte/mca/ess/hnp/ess_hnp_module.c
Просмотреть файл

@ -58,7 +58,6 @@
#include "orte/mca/plm/base/base.h"
#include "orte/mca/plm/plm.h"
#include "orte/mca/odls/base/base.h"
#include "orte/mca/notifier/base/base.h"
#include "orte/mca/sensor/base/base.h"
#include "orte/mca/sensor/sensor.h"
#include "orte/mca/rmaps/base/base.h"
@ -613,18 +612,6 @@ static int rte_init(void)
goto error;
}
/* setup the notifier system */
if (ORTE_SUCCESS != (ret = orte_notifier_base_open())) {
ORTE_ERROR_LOG(ret);
error = "orte_notifer_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_notifier_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orte_notifer_select";
goto error;
}
/* setup the SENSOR framework */
if (ORTE_SUCCESS != (ret = orte_sensor_base_open())) {
ORTE_ERROR_LOG(ret);

38
orte/mca/notifier/Makefile.am
Просмотреть файл

@ -1,38 +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 (c) 2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# main library setup
noinst_LTLIBRARIES = libmca_notifier.la
libmca_notifier_la_SOURCES =
# local files
headers = notifier.h notifier_event_types.h notifier_event_calls.h
libmca_notifier_la_SOURCES += $(headers)
# Conditionally install the header files
if WANT_INSTALL_HEADERS
ortedir = $(includedir)/openmpi/$(subdir)
nobase_orte_HEADERS = $(headers)
endif
include base/Makefile.am
distclean-local:
rm -f base/static-components.h

33
orte/mca/notifier/base/Makefile.am
Просмотреть файл

@ -1,33 +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 (c) 2009 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
headers += \
base/base.h
libmca_notifier_la_SOURCES += \
base/notifier_base_close.c \
base/notifier_base_select.c \
base/notifier_base_open.c
if !ORTE_DISABLE_FULL_SUPPORT
libmca_notifier_la_SOURCES += \
base/notifier_base_wrappers.c \
base/notifier_base_events.c
endif

121
orte/mca/notifier/base/base.h
Просмотреть файл

@ -1,121 +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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/** @file:
*/
#ifndef MCA_NOTIFIER_BASE_H
#define MCA_NOTIFIER_BASE_H
/*
* includes
*/
#include "orte_config.h"
#include "opal/mca/mca.h"
#include "opal/class/opal_object.h"
#include "opal/class/opal_list.h"
#include "orte/mca/ess/ess.h"
#include "orte/mca/notifier/notifier.h"
BEGIN_C_DECLS
/*
* Type for holding selected module / component pairs
*/
typedef struct {
opal_list_item_t super;
/* Component */
orte_notifier_base_component_t *onbsp_component;
/* Module */
orte_notifier_base_module_t *onbsp_module;
/* Priority */
int onbsp_priority;
} orte_notifier_base_selected_pair_t;
OBJ_CLASS_DECLARATION(orte_notifier_base_selected_pair_t);
/*
* function definitions
*/
ORTE_DECLSPEC int orte_notifier_base_open(void);
ORTE_DECLSPEC int orte_notifier_base_close(void);
ORTE_DECLSPEC int orte_notifier_base_select(void);
#if !ORTE_DISABLE_FULL_SUPPORT
ORTE_DECLSPEC void orte_notifier_log(orte_notifier_base_severity_t severity,
int errcode,
const char *msg, ...);
ORTE_DECLSPEC void orte_notifier_show_help(orte_notifier_base_severity_t severity,
int errcode,
const char *file,
const char *topic, ...);
ORTE_DECLSPEC void orte_notifier_log_peer(orte_notifier_base_severity_t severity,
int errcode,
orte_process_name_t *peer_proc,
const char *msg, ...);
ORTE_DECLSPEC const char* orte_notifier_base_sev2str(orte_notifier_base_severity_t severity);
ORTE_DECLSPEC char *orte_notifier_base_peer_log(int errcode,
orte_process_name_t *peer_proc,
const char *msg, va_list ap);
#if ORTE_WANT_NOTIFIER_LOG_EVENT
ORTE_DECLSPEC int orte_notifier_base_events_init(void);
ORTE_DECLSPEC void orte_notifier_base_events_finalize(void);
#else /* ORTE_WANT_NOTIFIER_LOG_EVENT */
#define orte_notifier_base_events_init() do {} while (0)
#define orte_notifier_base_events_finalize() do {} while (0)
#endif /* ORTE_WANT_NOTIFIER_LOG_EVENT */
/*
* global variables in the base
* Needs to be declspec'ed for ompi_info and others
*/
/*
* Indication of whether a component was successfully selected or not
* (1 component per interface)
*/
ORTE_DECLSPEC extern bool orte_notifier_base_log_selected;
ORTE_DECLSPEC extern bool orte_notifier_base_help_selected;
ORTE_DECLSPEC extern bool orte_notifier_base_log_peer_selected;
ORTE_DECLSPEC extern bool orte_notifier_base_log_event_selected;
/*
* Lists of selected modules (1 per interface)
*/
ORTE_DECLSPEC extern opal_list_t orte_notifier_log_selected_modules;
ORTE_DECLSPEC extern opal_list_t orte_notifier_help_selected_modules;
ORTE_DECLSPEC extern opal_list_t orte_notifier_log_peer_selected_modules;
ORTE_DECLSPEC extern opal_list_t orte_notifier_log_event_selected_modules;
/*
* That one is a merge of the per interface lists
* It is used during finalize phase to finalize only once each selected module
*/
ORTE_DECLSPEC extern opal_list_t orte_notifier_base_selected_modules;
ORTE_DECLSPEC extern int orte_notifier_base_output;
ORTE_DECLSPEC extern orte_notifier_base_severity_t orte_notifier_threshold_severity;
ORTE_DECLSPEC extern opal_list_t orte_notifier_base_components_available;
#endif /* !ORTE_DISABLE_FULL_SUPPORT */
END_C_DECLS
#endif

66
orte/mca/notifier/base/notifier_base_close.c
Просмотреть файл

@ -1,66 +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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include <stdio.h>
#include "opal/mca/mca.h"
#include "opal/mca/base/base.h"
#include "orte/mca/notifier/base/base.h"
int orte_notifier_base_close(void)
{
#if !ORTE_DISABLE_FULL_SUPPORT
opal_list_item_t *item;
orte_notifier_base_selected_pair_t *pair;
if (orte_notifier_base_log_event_selected) {
orte_notifier_base_events_finalize();
}
/* Finalize all the selected modules
* orte_notifier_base_selected_modules has been built as a merge of the
* per interface selected modules lists, so only going through that list
* to invoke the finalize routines is enough.
*/
for (item = opal_list_remove_first(&orte_notifier_base_selected_modules);
NULL != item;
item = opal_list_remove_first(&orte_notifier_base_selected_modules)) {
pair = (orte_notifier_base_selected_pair_t*) item;
if (NULL != pair->onbsp_module->finalize) {
pair->onbsp_module->finalize();
}
free(pair);
}
OBJ_DESTRUCT(&orte_notifier_base_selected_modules);
OBJ_DESTRUCT(&orte_notifier_log_selected_modules);
OBJ_DESTRUCT(&orte_notifier_help_selected_modules);
OBJ_DESTRUCT(&orte_notifier_log_peer_selected_modules);
OBJ_DESTRUCT(&orte_notifier_log_event_selected_modules);
/* Close all remaining available components */
mca_base_components_close(orte_notifier_base_output,
&orte_notifier_base_components_available, NULL);
#endif
/* All done */
return ORTE_SUCCESS;
}

197
orte/mca/notifier/base/notifier_base_events.c
Просмотреть файл

@ -1,197 +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 (c) 2009 Bull SAS. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/types.h"
#include "orte/constants.h"
#ifdef HAVE_STDLIB_H
#include <stdlib.h>
#endif /* HAVE_STDLIB_H */
#ifdef HAVE_STDIO_H
#include <stdio.h>
#endif /* HAVE_STDIO_H */
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif /* HAVE_SYS_TIME_H */
#include "opal/class/opal_object.h"
#include "opal/class/opal_list.h"
#include "orte/util/proc_info.h"
#include "orte/util/name_fns.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/notifier/base/base.h"
#if ORTE_WANT_NOTIFIER_LOG_EVENT
/*
* Definitions for the events that are accounted for before being logged.
* They are stored in a list to ensure they are all unconditionally traced
* out during finalize.
*/
opal_list_t orte_notifier_events_list;
/*
* Log format differs depending on the phase we are in.
*/
#define ORTE_NOTIFIER_LOG_FORMAT_0 "TIME=%ld MPI_NOTIFIER_EVENT FAMILY=%u JOB=%u VPID=%u HOST=%s EVENT=%d COUNT=%u: %s"
#define ORTE_NOTIFIER_LOG_FORMAT_1 "TIME=%ld MPI_NOTIFIER_EVENT FAMILY=%u JOB=%u VPID=%u HOST=%s EVENT=%d COUNT=%u (in %ld seconds): %s"
#define ORTE_NOTIFIER_LOG_FORMAT_2 "TIME=%ld MPI_NOTIFIER_EVENT FAMILY=%u JOB=%u VPID=%u HOST=%s EVENT=%d COUNT=%u (Finalize): %s"
static void orte_notifier_event_construct(orte_notifier_event_t *ev)
{
ev->ev_cnt = 0;
ev->ev_already_traced = 0;
ev->ev_msg = NULL;
}
static void orte_notifier_event_destruct(orte_notifier_event_t *ev)
{
if (NULL != ev->ev_msg) {
free(ev->ev_msg);
}
}
OBJ_CLASS_INSTANCE(orte_notifier_event_t,
opal_list_item_t,
orte_notifier_event_construct,
orte_notifier_event_destruct);
int orte_notifier_base_events_init(void)
{
if (!ORTE_PROC_IS_HNP) {
OBJ_CONSTRUCT(&orte_notifier_events_list, opal_list_t);
}
return ORTE_SUCCESS;
}
void orte_notifier_base_events_finalize(void)
{
orte_notifier_event_t *ev;
opal_list_item_t *item;
int32_t count;
if (ORTE_PROC_IS_HNP) {
return;
}
/*
* Unconditionally trace any event that has been accounted for
*/
for (item = opal_list_remove_first(&orte_notifier_events_list);
NULL != item;
item = opal_list_remove_first(&orte_notifier_events_list)) {
ev = (orte_notifier_event_t *) item;
if ((count = ev->ev_cnt) && notifier_log_event_enabled()) {
notifier_trace_event(ORTE_NOTIFIER_LOG_2, ev->ev_id, count,
time(NULL), 0, ev->ev_msg);
}
OBJ_RELEASE(ev);
}
OBJ_DESTRUCT(&orte_notifier_events_list);
}
/*
* log_type indicates whether we are tracing one of the following:
* . ORTE_NOTIFIER_LOG_0 --> Very first trace
* . ORTE_NOTIFIER_LOG_1 --> Intermediate trace
* . ORTE_NOTIFIER_LOG_2 --> during finalize
* Depending on the log_type the output format is different.
*/
void notifier_trace_event(int log_type, int ev_id, int32_t count, time_t t,
time_t delay, const char *msg)
{
opal_list_item_t *item;
orte_notifier_base_selected_pair_t *pair;
orte_process_name_t *pname = ORTE_PROC_MY_NAME;
char *out = NULL;
switch (log_type) {
case ORTE_NOTIFIER_LOG_0:
asprintf(&out, ORTE_NOTIFIER_LOG_FORMAT_0, t,
ORTE_JOB_FAMILY(pname->jobid),
ORTE_LOCAL_JOBID(pname->jobid),
pname->vpid,
orte_process_info.nodename,
ev_id,
count,
msg);
break;
case ORTE_NOTIFIER_LOG_1:
asprintf(&out, ORTE_NOTIFIER_LOG_FORMAT_1, t,
ORTE_JOB_FAMILY(pname->jobid),
ORTE_LOCAL_JOBID(pname->jobid),
pname->vpid,
orte_process_info.nodename,
ev_id,
count,
delay,
msg);
break;
case ORTE_NOTIFIER_LOG_2:
asprintf(&out, ORTE_NOTIFIER_LOG_FORMAT_2, t,
ORTE_JOB_FAMILY(pname->jobid),
ORTE_LOCAL_JOBID(pname->jobid),
pname->vpid,
orte_process_info.nodename,
ev_id,
count,
msg);
break;
default:
asprintf(&out, "UNKNOWN!!!!!!!!!");
break;
}
if (NULL == out) {
return;
}
for (item = opal_list_get_first(&orte_notifier_log_event_selected_modules);
opal_list_get_end(&orte_notifier_log_event_selected_modules) != item;
item = opal_list_get_next(item)) {
pair = (orte_notifier_base_selected_pair_t*) item;
if (NULL != pair->onbsp_module->log_event) {
pair->onbsp_module->log_event(out);
}
}
}
void notifier_event_store(orte_notifier_event_t *ev)
{
opal_list_append(&orte_notifier_events_list, &ev->super);
}
bool notifier_log_event_enabled(void)
{
return orte_notifier_base_log_event_selected &&
(ORTE_NOTIFIER_NOTICE <= orte_notifier_threshold_severity);
}
#endif /* ORTE_WANT_NOTIFIER_LOG_EVENT */

123
orte/mca/notifier/base/notifier_base_open.c
Просмотреть файл

@ -1,123 +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 (c) 2008-2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "opal/mca/mca.h"
#include "opal/util/output.h"
#include "opal/mca/base/base.h"
#include "opal/mca/base/mca_base_param.h"
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#include "orte/mca/notifier/base/base.h"
/*
* The following file was created by configure. It contains extern
* statements and the definition of an array of pointers to each
* component's public mca_base_component_t struct.
*/
#include "orte/mca/notifier/base/static-components.h"
#if !ORTE_DISABLE_FULL_SUPPORT
/*
* Global variables
*/
int orte_notifier_base_output = -1;
orte_notifier_base_severity_t orte_notifier_threshold_severity =
ORTE_NOTIFIER_ERROR;
opal_list_t orte_notifier_base_components_available;
opal_list_t orte_notifier_base_selected_modules;
opal_list_t orte_notifier_log_selected_modules;
opal_list_t orte_notifier_help_selected_modules;
opal_list_t orte_notifier_log_peer_selected_modules;
opal_list_t orte_notifier_log_event_selected_modules;
orte_notifier_API_module_t orte_notifier = {
orte_notifier_log,
orte_notifier_show_help,
orte_notifier_log_peer,
};
#endif
/**
* Function for finding and opening either all MCA components, or the one
* that was specifically requested via a MCA parameter.
*/
int orte_notifier_base_open(void)
{
#if !ORTE_DISABLE_FULL_SUPPORT
char *level;
/* Debugging / verbose output. Always have stream open, with
verbose set by the mca open system... */
orte_notifier_base_output = opal_output_open(NULL);
/* let the user define a base level of severity to report */
mca_base_param_reg_string_name("notifier", "threshold_severity",
"Report all events at or above this severity [default: error]",
false, false, "error", &level);
if (0 == strncasecmp(level, "emerg", strlen("emerg"))) {
orte_notifier_threshold_severity = ORTE_NOTIFIER_EMERG;
} else if (0 == strncasecmp(level, "alert", strlen("alert"))) {
orte_notifier_threshold_severity = ORTE_NOTIFIER_ALERT;
} else if (0 == strncasecmp(level, "crit", strlen("crit"))) {
orte_notifier_threshold_severity = ORTE_NOTIFIER_CRIT;
} else if (0 == strncasecmp(level, "warn", strlen("warn"))) {
orte_notifier_threshold_severity = ORTE_NOTIFIER_WARN;
} else if (0 == strncasecmp(level, "notice", strlen("notice"))) {
orte_notifier_threshold_severity = ORTE_NOTIFIER_NOTICE;
} else if (0 == strncasecmp(level, "info", strlen("info"))) {
orte_notifier_threshold_severity = ORTE_NOTIFIER_INFO;
} else if (0 == strncasecmp(level, "debug", strlen("debug"))) {
orte_notifier_threshold_severity = ORTE_NOTIFIER_DEBUG;
} else if (0 != strncasecmp(level, "error", strlen("error"))) {
opal_output(0, "Unknown notifier level");
return ORTE_ERROR;
}
free(level);
OBJ_CONSTRUCT(&orte_notifier_base_selected_modules, opal_list_t);
OBJ_CONSTRUCT(&orte_notifier_log_selected_modules, opal_list_t);
OBJ_CONSTRUCT(&orte_notifier_help_selected_modules, opal_list_t);
OBJ_CONSTRUCT(&orte_notifier_log_peer_selected_modules, opal_list_t);
OBJ_CONSTRUCT(&orte_notifier_log_event_selected_modules, opal_list_t);
/* Open up all available components */
if (ORTE_SUCCESS !=
mca_base_components_open("notifier", orte_notifier_base_output,
mca_notifier_base_static_components,
&orte_notifier_base_components_available,
true)) {
return ORTE_ERROR;
}
/* All done */
#endif
return ORTE_SUCCESS;
}

392
orte/mca/notifier/base/notifier_base_select.c
Просмотреть файл

@ -1,392 +0,0 @@
/*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#include "opal/mca/mca.h"
#include "opal/util/argv.h"
#include "opal/mca/base/base.h"
#include "opal/util/output.h"
#include "orte/mca/notifier/base/base.h"
#if !ORTE_DISABLE_FULL_SUPPORT
/* Global variables */
/*
* orte_notifier_base_XXX_selected is set to true if at least 1 module has
* been selected for the notifier XXX API interface.
*/
bool orte_notifier_base_log_selected = false;
bool orte_notifier_base_help_selected = false;
bool orte_notifier_base_log_peer_selected = false;
bool orte_notifier_base_log_event_selected = false;
static inline char **orte_notifier_get_include_list(const char *,
const char *,
char **);
static bool orte_notifier_add_module(mca_base_component_t *component,
orte_notifier_base_module_t *module,
int priority,
char **include_list,
opal_list_t *selected_modules);
static void onbsp_construct(orte_notifier_base_selected_pair_t *obj)
{
obj->onbsp_component = NULL;
obj->onbsp_module = NULL;
obj->onbsp_priority = -1;
}
static void onbsp_destruct(orte_notifier_base_selected_pair_t *obj)
{
onbsp_construct(obj);
}
OBJ_CLASS_INSTANCE(orte_notifier_base_selected_pair_t,
opal_list_item_t,
onbsp_construct,
onbsp_destruct);
/**
* Function for selecting a set of components from all those that are
* available.
*
* It is possible to select a subset of these components for any interface.
* The syntax is the following:
* [ -mca notifier <list0> ] [ -mca notifier_log <list1> ]
* [ -mca notifier_help <list2> ]
* [ -mca notifier_log_peer <list3> ]
* [ -mca notifier_log_event <list4> ]
* Rules:
* . <list0> empty means nothing selected
* . <list0> to <list4> = comma separated lists of component names
* . <list1> to <list4> may be one of:
* . subsets of <list0>
* . "none" keyword (means empty)
* . 1 of <list1> to <list4> empty means = <list0>
* Last point makes it possible to preserve the way it works today
*
* Examples:
* 1)
* -mca notifier syslog,smtp
* --> syslog and smtp are selected for the log, show_help, log_peer and
* log_event interfaces.
* 2)
* -mca notifier_log syslog
* --> no interface is activated, no component is selected
* 3)
* -mca notifier syslog -mca notifier_help none
* -mca notifier_log_peer none
* -mca notifier_log_event none
* --> only the log interface is activated, with the syslog component
* 4)
* -mca notifier syslog,smtp,hnp -mca notifier_help syslog
* -mca notifier_log_peer smtp
* -mca notifier_log_event none
* --> the log interface is activated, with the syslog, smtp and hnp
* components
* the log_help interface is activated, with the syslog component
* the log_peer interface is activated, with the smtp component
* the log_event interface is not activated
*/
int orte_notifier_base_select(void)
{
mca_base_component_list_item_t *cli = NULL;
mca_base_component_t *component = NULL;
mca_base_module_t *module = NULL;
int i, ret, priority, exit_status = ORTE_SUCCESS;
opal_list_item_t *item;
orte_notifier_base_module_t *nmodule;
char **imodules;
char **imodules_log, **imodules_help, **imodules_log_peer;
char **imodules_log_event = NULL;
bool module_needed;
/*
* Register the framework MCA param and look up include list
*/
imodules = orte_notifier_get_include_list("notifier",
"Comma-delimisted list of notifier components to use "
"(empty = none)", NULL);
if (NULL == imodules) {
return ORTE_SUCCESS;
}
/*
* Also get the include lists for each interface
*/
imodules_log = orte_notifier_get_include_list("notifier_log",
"Comma-delimisted list of notifier components to use "
"for orte_notifier_log (empty = all selected)",
imodules);
imodules_help = orte_notifier_get_include_list("notifier_help",
"Comma-delimisted list of notifier components to use "
"for orte_notifier_show_help (empty = all selected)",
imodules);
imodules_log_peer = orte_notifier_get_include_list("notifier_log_peer",
"Comma-delimisted list of notifier components to "
"use for orte_notifier_log_peer (empty = all "
"selected)", imodules);
#if ORTE_WANT_NOTIFIER_LOG_EVENT
imodules_log_event = orte_notifier_get_include_list("notifier_log_event",
"Comma-delimisted list of notifier components to "
"use for ORTE_NOTIFIER_LOG_EVENT (empty = all "
"selected)",
imodules);
#endif /* ORTE_WANT_NOTIFIER_LOG_EVENT */
/* Query all available components and ask if they have a module */
for (item = opal_list_get_first(&orte_notifier_base_components_available);
opal_list_get_end(&orte_notifier_base_components_available) != item;
item = opal_list_get_next(item)) {
cli = (mca_base_component_list_item_t *) item;
component = (mca_base_component_t *) cli->cli_component;
/* If this component is not in the include list, skip it */
for (i = 0; NULL != imodules[i]; ++i) {
if (0 == strcmp(imodules[i], component->mca_component_name)) {
break;
}
}
if (NULL == imodules[i]) {
continue;
}
/* If there's no query function, skip it */
if (NULL == component->mca_query_component) {
opal_output_verbose(5, orte_notifier_base_output,
"mca:notify:select: Skipping component [%s]. It does not implement a query function",
component->mca_component_name );
continue;
}
/* Query the component */
opal_output_verbose(5, orte_notifier_base_output,
"mca:notify:select: Querying component [%s]",
component->mca_component_name);
ret = component->mca_query_component(&module, &priority);
/* If no module was returned, then skip component */
if (ORTE_SUCCESS != ret || NULL == module) {
opal_output_verbose(5, orte_notifier_base_output,
"mca:notify:select: Skipping component [%s]. Query failed to return a module",
component->mca_component_name );
continue;
}
/* If we got a module, initialize it */
nmodule = (orte_notifier_base_module_t*) module;
if (NULL != nmodule->init) {
/* If the module doesn't want to be used, skip it */
if (ORTE_SUCCESS != (ret = nmodule->init()) ) {
if (ORTE_ERR_NOT_SUPPORTED != ret &&
ORTE_ERR_NOT_IMPLEMENTED != ret) {
exit_status = ret;
goto cleanup;
}
if (NULL != nmodule->finalize) {
nmodule->finalize();
}
continue;
}
}
/*
* OK, one module has been selected for the notifier framework, and
* successfully initialized.
* Now we have to include it in the per interface selected modules
* lists if needed.
*/
ret = orte_notifier_add_module(component,
nmodule,
priority,
imodules_log,
&orte_notifier_log_selected_modules);
orte_notifier_base_log_selected = orte_notifier_base_log_selected
|| ret;
/*
* This variable is set to check if the module is needed by at least
* one interface.
*/
module_needed = ret;
ret = orte_notifier_add_module(component,
nmodule,
priority,
imodules_help,
&orte_notifier_help_selected_modules);
orte_notifier_base_help_selected = orte_notifier_base_help_selected
|| ret;
module_needed = module_needed || ret;
ret = orte_notifier_add_module(component,
nmodule,
priority,
imodules_log_peer,
&orte_notifier_log_peer_selected_modules);
orte_notifier_base_log_peer_selected =
orte_notifier_base_log_peer_selected || ret;
module_needed = module_needed || ret;
ret = orte_notifier_add_module(component,
nmodule,
priority,
imodules_log_event,
&orte_notifier_log_event_selected_modules);
orte_notifier_base_log_event_selected =
orte_notifier_base_log_event_selected || ret;
module_needed = module_needed || ret;
/*
* If the module is needed by at least one interface:
* Unconditionally update the global list that will be used during
* the close step. Else unload it.
*/
if (module_needed) {
orte_notifier_add_module(component,
nmodule,
priority,
imodules,
&orte_notifier_base_selected_modules);
} else {
if (NULL != nmodule->finalize) {
nmodule->finalize();
}
}
}
if (orte_notifier_base_log_event_selected) {
/*
* This has to be done whatever the selected module. That's why it's
* done here.
*/
orte_notifier_base_events_init();
}
cleanup:
return exit_status;
}
/**
* Register an mca param that represents an include list and build that list.
*
* @param param_name (IN) param name to be registered
* @param help_message (IN) help message for that param
* @param default_modules (IN) list of module names to be inherited if an
* empty include list is provided
* @return list of modules names
*/
static inline char **orte_notifier_get_include_list(const char *param_name,
const char *help_message,
char **default_modules)
{
char *include_list = NULL;
char **imodules;
mca_base_param_reg_string_name(param_name, NULL, help_message,
false, false, NULL, &include_list);
imodules = opal_argv_split(include_list, ',');
if (NULL == imodules) {
/*
* Inherit the default list if nothing specified
*/
return default_modules;
}
if (!strcmp(include_list, "none")) {
return NULL;
}
return imodules;
}
/**
* Check if a component name belongs to an include list and add it to the
* list of selected modules.
*
* @param component (IN) component to be included
* @param module (IN) module to be included
* @param priority (IN) module priority
* @param include_list (IN) list of module names to go through
* @param selected_modules (OUT) list of selected modules to be updated
* @return true/false depending on whether the module
* has been added or not
*/
static bool orte_notifier_add_module(mca_base_component_t *component,
orte_notifier_base_module_t *module,
int priority,
char **include_list,
opal_list_t *selected_modules)
{
orte_notifier_base_selected_pair_t *pair, *pair2;
char *module_name;
opal_list_item_t *item;
int i;
if (NULL == include_list) {
return false;
}
module_name = component->mca_component_name;
/* If this component is not in the include list, skip it */
for (i = 0; NULL != include_list[i]; i++) {
if (!strcmp(include_list[i], module_name)) {
break;
}
}
if (NULL == include_list[i]) {
return false;
}
/* Make an item for the list */
pair = OBJ_NEW(orte_notifier_base_selected_pair_t);
pair->onbsp_component = (orte_notifier_base_component_t*) component;
pair->onbsp_module = module;
pair->onbsp_priority = priority;
/* Put it in the list in priority order */
for (item = opal_list_get_first(selected_modules);
opal_list_get_end(selected_modules) != item;
item = opal_list_get_next(item)) {
pair2 = (orte_notifier_base_selected_pair_t*) item;
if (priority > pair2->onbsp_priority) {
opal_list_insert_pos(selected_modules, item, &(pair->super));
break;
}
}
if (opal_list_get_end(selected_modules) == item) {
opal_list_append(selected_modules, &(pair->super));
}
return true;
}
#endif

193
orte/mca/notifier/base/notifier_base_wrappers.c
Просмотреть файл

@ -1,193 +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 (c) 2008-2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "orte/mca/ess/ess.h"
#include "orte/util/error_strings.h"
#include "orte/util/name_fns.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/notifier/base/base.h"
#if !ORTE_DISABLE_FULL_SUPPORT
void orte_notifier_log(orte_notifier_base_severity_t severity,
int errcode, const char *msg, ...)
{
va_list ap;
opal_list_item_t *item;
orte_notifier_base_selected_pair_t *pair;
if (!orte_notifier_base_log_selected) {
return;
}
/* is the severity value above the threshold - I know
* this seems backward, but lower severity values are
* considered "more severe"
*/
if (severity > orte_notifier_threshold_severity) {
return;
}
for (item = opal_list_get_first(&orte_notifier_log_selected_modules);
opal_list_get_end(&orte_notifier_log_selected_modules) != item;
item = opal_list_get_next(item)) {
pair = (orte_notifier_base_selected_pair_t*) item;
if (NULL != pair->onbsp_module->log) {
va_start(ap, msg);
pair->onbsp_module->log(severity, errcode, msg, ap);
va_end(ap);
}
}
}
void orte_notifier_show_help(orte_notifier_base_severity_t severity,
int errcode, const char *file,
const char *topic, ...)
{
va_list ap;
opal_list_item_t *item;
orte_notifier_base_selected_pair_t *pair;
if (!orte_notifier_base_help_selected) {
return;
}
/* is the severity value above the threshold - I know
* this seems backward, but lower severity values are
* considered "more severe"
*/
if (severity > orte_notifier_threshold_severity) {
return;
}
for (item = opal_list_get_first(&orte_notifier_help_selected_modules);
opal_list_get_end(&orte_notifier_help_selected_modules) != item;
item = opal_list_get_next(item)) {
pair = (orte_notifier_base_selected_pair_t*) item;
if (NULL != pair->onbsp_module->help) {
va_start(ap, topic);
pair->onbsp_module->help(severity, errcode, file, topic, ap);
va_end(ap);
}
}
}
void orte_notifier_log_peer(orte_notifier_base_severity_t severity,
int errcode,
orte_process_name_t *peer_proc,
const char *msg, ...)
{
va_list ap;
opal_list_item_t *item;
orte_notifier_base_selected_pair_t *pair;
if (!orte_notifier_base_log_peer_selected) {
return;
}
/* is the severity value above the threshold - I know
* this seems backward, but lower severity values are
* considered "more severe"
*/
if (severity > orte_notifier_threshold_severity) {
return;
}
for (item = opal_list_get_first(&orte_notifier_log_peer_selected_modules);
opal_list_get_end(&orte_notifier_log_peer_selected_modules) != item;
item = opal_list_get_next(item)) {
pair = (orte_notifier_base_selected_pair_t*) item;
if (NULL != pair->onbsp_module->peer) {
va_start(ap, msg);
pair->onbsp_module->peer(severity, errcode, peer_proc, msg, ap);
va_end(ap);
}
}
}
const char* orte_notifier_base_sev2str(orte_notifier_base_severity_t severity)
{
switch (severity) {
case ORTE_NOTIFIER_EMERG: return "EMERG"; break;
case ORTE_NOTIFIER_ALERT: return "ALERT"; break;
case ORTE_NOTIFIER_CRIT: return "CRIT"; break;
case ORTE_NOTIFIER_ERROR: return "ERROR"; break;
case ORTE_NOTIFIER_WARN: return "WARN"; break;
case ORTE_NOTIFIER_NOTICE: return "NOTICE"; break;
case ORTE_NOTIFIER_INFO: return "INFO"; break;
case ORTE_NOTIFIER_DEBUG: return "DEBUG"; break;
default: return "UNKNOWN"; break;
}
}
char *orte_notifier_base_peer_log(int errcode, orte_process_name_t *peer_proc,
const char *msg, va_list ap)
{
char *buf = (char *) malloc(ORTE_NOTIFIER_MAX_BUF + 1);
char *peer_host = NULL, *peer_name = NULL;
char *pos = buf;
char *errstr;
int ret, len, space = ORTE_NOTIFIER_MAX_BUF;
if (NULL == buf) {
return NULL;
}
if (peer_proc) {
peer_host = orte_ess.proc_get_hostname(peer_proc);
peer_name = ORTE_NAME_PRINT(peer_proc);
}
len = snprintf(pos, space,
"While communicating to proc %s on node %s,"
" proc %s on node %s encountered an error ",
peer_name ? peer_name : "UNKNOWN",
peer_host ? peer_host : "UNKNOWN",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
orte_process_info.nodename);
space -= len;
pos += len;
if (0 < space) {
ret = orte_err2str(errcode, (const char **)&errstr);
if (ORTE_SUCCESS == ret) {
len = snprintf(pos, space, "'%s':", errstr);
free(errstr);
} else {
len = snprintf(pos, space, "(%d):", errcode);
}
space -= len;
pos += len;
}
if (0 < space) {
vsnprintf(pos, space, msg, ap);
}
buf[ORTE_NOTIFIER_MAX_BUF] = '\0';
return buf;
}
#endif

49
orte/mca/notifier/command/Makefile.am
Просмотреть файл

@ -1,49 +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 (c) 2009-2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
dist_pkgdata_DATA = \
help-orte-notifier-command.txt
sources = \
notifier_command.h \
notifier_command_fd.c \
notifier_command_child.c \
notifier_command_module.c \
notifier_command_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 MCA_BUILD_orte_notifier_command_DSO
component_noinst =
component_install = mca_notifier_command.la
else
component_noinst = libmca_notifier_command.la
component_install =
endif
mcacomponentdir = $(pkglibdir)
mcacomponent_LTLIBRARIES = $(component_install)
mca_notifier_command_la_SOURCES = $(sources)
mca_notifier_command_la_LDFLAGS = -module -avoid-version
noinst_LTLIBRARIES = $(component_noinst)
libmca_notifier_command_la_SOURCES =$(sources)
libmca_notifier_command_la_LDFLAGS = -module -avoid-version

36
orte/mca/notifier/command/configure.m4
Просмотреть файл

@ -1,36 +0,0 @@
# -*- command-script -*-
#
# Copyright (c) 2007 Sandia National Laboratories. All rights reserved.
# Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved.
# Copyright (c) 2011 Los Alamos National Security, LLC.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# MCA_notifier_command_CONFIG([action-if-found], [action-if-not-found])
# -----------------------------------------------------------
AC_DEFUN([MCA_orte_notifier_command_CONFIG], [
AC_CONFIG_FILES([orte/mca/notifier/command/Makefile])
OPAL_VAR_SCOPE_PUSH(notifier_happy)
notifier_happy=no
# We need fork() and pipe()
AC_CHECK_FUNC([fork],
[AC_CHECK_FUNC([pipe], [notifier_happy=yes])])
# We also need thread support
AS_IF([test "$notifier_happy" = "yes"],
[AC_MSG_CHECKING([for thread support])
AC_MSG_RESULT([$THREAD_TYPE])
AS_IF([test "$THREAD_TYPE" != "none"],
[notifier_happy=yes])])
AS_IF([test "$notifier_happy" = "yes" -a "$orte_without_full_support" = 0], [$1], [$2])
OPAL_VAR_SCOPE_POP
])

66
orte/mca/notifier/command/help-orte-notifier-command.txt
Просмотреть файл

@ -1,66 +0,0 @@
# -*- text -*-
#
# Copyright (c) 2009 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# This is the US/English help file for Open MPI's SMTP notifier support
#
[command not specified]
Error: the Open MPI command notifier component had no command specified.
#
[bad command]
Error: the command notifier component received a bad command in the
notifier_command_cmd MCA parameter. This usually means that there
are mismatched quotes in the command string. Your MPI job may
continue, but the command notifier has been disabled.
Local host: %s
Command: %s
#
[system call fail]
Error: a system call failed during the setup of the command notifier
component. Open MPI is now going to abort your job.
Local host: %s
System call: %s
Errno: %s (%d)
#
[grandchild fail]
The command notifier process died with a non-zero exit status. This
should not happen. Your MPI job will continue, however, and
notifications will attempt to continue. But you may only see this
message once, even if notifications continue to fail.
Local host: %s
Command: %s
Exit status: %s %d
#
[grandchild did not exit]
ERROR: The command notifier process took too long, but was unable to be
killed by Open MPI (Open MPI tried killing it with SIGTERM and
SIGKILL). This should not happen; you should both check the host
where this occurred to see if there are any notifier processes still
running, and check your notifier command and ensure that it is
functioning properly. Your MPI job will continue, however, and
notifications will attempt to continue. But you may only see this
message once, even if notifications continue to fail.
Local host: %s
Command: %s
Timeout (sec): %d
#
[grandchild timeout]
The command notifier process took too long and was killed by Open MPI.
This should not happen; you should check your notifier command and
ensure that it is functioning properly. Your MPI job will continue,
however, and notifications will attempt to continue. But you may only
see this message once, even if notifications continue to fail.
Local host: %s
Command: %s
Timeout (sec): %d
Exit status: %s %d

107
orte/mca/notifier/command/notifier_command.h
Просмотреть файл

@ -1,107 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
*/
#ifndef NOTIFIER_COMMAND_H
#define NOTIFIER_COMMAND_H
#include "orte_config.h"
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#include "orte/mca/notifier/notifier.h"
BEGIN_C_DECLS
typedef struct {
orte_notifier_base_component_t super;
/* Command to execute */
char *cmd;
/* Timeout of the command (seconds) */
int timeout;
/* Priority of this component */
int priority;
/* Child PID */
pid_t child_pid;
/* Pipe to the child */
int to_child[2];
/* Pipe to the parent */
int to_parent[2];
/* Do we want data sent to child via stdin? */
bool pass_via_stdin;
} orte_notifier_command_component_t;
/*
* Notifier interfaces
*/
ORTE_MODULE_DECLSPEC extern orte_notifier_command_component_t
mca_notifier_command_component;
extern orte_notifier_base_module_t orte_notifier_command_module;
/*
* Pipe commands
*/
typedef enum {
/* Fork/exec a command */
CMD_EXEC,
/* Time to quit */
CMD_TIME_TO_QUIT,
/* Sentinel value */
CMD_MAX
} orte_notifier_command_pipe_cmd_t;
/**
* Simple blocking function to read a specific number of bytes from an
* fd.
*/
int orte_notifier_command_read_fd(int fd, int len, void *buffer);
/**
* Simple blocking function to write a specific number of bytes to an
* fd.
*/
int orte_notifier_command_write_fd(int fd, int len, void *buffer);
/**
* Main entry point for child
*/
void orte_notifier_command_child_main(void) __opal_attribute_noreturn__;
/**
* Function to split a spint into argv, honoring quoting, etc. (and do
* some error checking of the string)
*/
int orte_notifier_command_split(const char *cmd, char ***argv);
END_C_DECLS
#endif

446
orte/mca/notifier/command/notifier_command_child.c
Просмотреть файл

@ -1,446 +0,0 @@
/*
* Copyright (c) 2008-2009 Cisco Systems, Inc. All rights reserved.
*
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* Note: this file is a little fast-n-loose with OPAL_HAVE_THREADS --
* it uses this value in run-time "if" conditionals (vs. compile-time
* #if conditionals). We also don't protect including <pthread.h>.
* That's because this component currently only compiles on Linux and
* Solaris, and both of these OS's have pthreads. Using the run-time
* conditionals gives us better compile-time checking, even of code
* that isn't activated.
*
* Note, too, that the functionality in this file does *not* require
* all the heavyweight OMPI thread infrastructure (e.g., from
* --enable-mpi-thread-multiple or --enable-progress-threads). All work that
* is done in a separate progress thread is very carefully segregated
* from that of the main thread, and communication back to the main
* thread
*/
#include "orte_config.h"
#include <string.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#include <errno.h>
#ifdef HAVE_SYS_WAIT_H
#include <sys/wait.h>
#endif
#ifdef HAVE_TIME_H
#include <time.h>
#endif
#include <ctype.h>
#include <signal.h>
#include "opal/util/argv.h"
#include "opal/threads/threads.h"
#include "orte/constants.h"
#include "orte/mca/notifier/base/base.h"
#include "notifier_command.h"
static void diediedie(int status) __opal_attribute_noreturn__;
/* Structre for holding the argument to stdin_main() */
typedef struct {
int sat_pipe_fd;
int sat_severity;
int sat_errcode;
char *sat_msg;
} stdin_arg_t;
int orte_notifier_command_split(const char *cmd_arg, char ***argv_arg)
{
int i;
char *cmd, *p, *q, *token_start, **argv = NULL;
bool in_space, in_quote, in_2quote;
*argv_arg = NULL;
cmd = strdup(cmd_arg);
if (NULL == cmd) {
return ORTE_ERR_IN_ERRNO;
}
in_space = in_quote = in_2quote = false;
for (token_start = p = cmd; '\0' != *p; ++p) {
/* If we're in a quoted string, all we're doing it looking for
the matching end quote. Note that finding the end quote
does not necessarily mean the end of the token! So use the
normal "I found a space [outside of a quote]" processing to
find the end of the token. */
if (in_quote &&
('\'' == *p && p > token_start && '\\' != *(p - 1))) {
in_quote = false;
} else if (in_2quote &&
('\"' == *p && p > token_start && '\\' != *(p - 1))) {
in_2quote = false;
}
/* If we hit a space, it could be the end of a token -- unless
we're already in a series of spaces. */
else if (!in_quote && !in_2quote && isspace(*p)) {
if (!in_space) {
/* We weren't in a series of spaces, so this was the
end of a token. Save it. */
in_space = true;
*p = '\0';
opal_argv_append_nosize(&argv, token_start);
token_start = p + 1;
} else {
/* We're in a series of spaces, so just move
token_start up to the next character. */
token_start = p + 1;
}
} else {
/* We're not in a series of spaces. We only need to check
if we find ' or " to start a quoted string (in which
case spaces no longer mark the end of a string). */
in_space = false;
if ('\'' == *p) {
in_quote = true;
} else if ('"' == *p) {
in_2quote = true;
}
}
}
if (in_quote || in_2quote) {
free(cmd);
opal_argv_free(argv);
return ORTE_ERR_BAD_PARAM;
}
/* Get the last token, if there is one */
if (!in_space) {
opal_argv_append_nosize(&argv, token_start);
}
/* Replace escapes and non-escaped quotes */
for (i = 0; NULL != argv[i]; ++i) {
for (p = q = argv[i]; '\0' != *p; ++p) {
if ('\\' == *p) {
switch (*(p + 1)) {
/* For quotes, just copy them over and
double-increment p */
case '\'': *q = *(p + 1); ++p; break;
case '"': *q = *(p + 1); ++p; break;
/* For other normal escapes, insert the right code
and double-increment p */
case 'a': *q = '\a'; ++p; break;
case 'b': *q = '\b'; ++p; break;
case 'f': *q = '\f'; ++p; break;
case 'n': *q = '\n'; ++p; break;
case 'r': *q = '\r'; ++p; break;
case 't': *q = '\t'; ++p; break;
case 'v': *q = '\v'; ++p; break;
/* For un-terminated escape, just put in a \. Do
*not* double increment p; it's the end of the
string! */
case '\0': *q = '\\'; break;
/* Otherwise, just copy and double increment */
default: *q = *p; ++p; break;
}
++q;
} else {
/* Don't copy un-escaped quotes */
if ('\'' != *p && '"' != *p) {
*q = *p;
++q;
}
}
}
*q = '\0';
}
*argv_arg = argv;
free(cmd);
return ORTE_SUCCESS;
}
/*
* Die nicely
*/
static void diediedie(int status)
{
/* We don't really have any way to report anything, so just close
the pipe fd and die */
close(mca_notifier_command_component.to_child[0]);
close(mca_notifier_command_component.to_parent[1]);
_exit(status);
}
/*
* Main entry point for stdin thread
*/
static void *stdin_main(opal_object_t *obj)
{
char *data;
opal_thread_t *t = (opal_thread_t*) obj;
stdin_arg_t *arg = (stdin_arg_t*) t->t_arg;
asprintf(&data, "<stdin>\n<notifier severity_int=\"%d\" severity_str=\"%s\" errcode=\"%d\">\n<message>%s</message>\n</notifier>\n</stdin>\n",
arg->sat_severity,
orte_notifier_base_sev2str((orte_notifier_base_severity_t)arg->sat_severity),
arg->sat_errcode,
arg->sat_msg);
if (NULL != data) {
orte_notifier_command_write_fd(arg->sat_pipe_fd,
strlen(data) + 1, data);
free(data);
close(arg->sat_pipe_fd);
}
return NULL;
}
/*
* Loop over waiting for a child to die
*/
static int do_wait(pid_t pid, int timeout, int *status, bool *exited)
{
pid_t pid2;
time_t t1, t2;
t2 = t1 = time(NULL);
*exited = false;
while (timeout <= 0 || t2 - t1 < timeout) {
pid2 = waitpid(pid, status, WNOHANG);
if (pid2 == pid) {
*exited = true;
return ORTE_SUCCESS;
} else if (pid2 < 0 && EINTR != errno) {
if (ECHILD == errno) {
*exited = true;
return ORTE_ERR_NOT_FOUND;
}
/* What else can we do? */
diediedie(10);
}
/* Let the child run a bit */
usleep(100);
t2 = time(NULL);
}
return ORTE_ERROR;
}
/*
* Fork/exec a command from the parent
*/
static void do_exec(void)
{
pid_t pid;
bool exited, killed;
int sel[3], status;
int pipe_to_stdin[2];
char *msg, *p, *cmd, **argv = NULL;
orte_notifier_command_component_t *c = &mca_notifier_command_component;
opal_thread_t stdin_thread;
stdin_arg_t arg;
/* First three items on the pipe are: severity, errcode, and
string length (sel = Severity, Errcode, string Length. */
if (ORTE_SUCCESS !=
orte_notifier_command_read_fd(c->to_child[0], sizeof(sel), sel)) {
diediedie(1);
}
/* Malloc out enough space for the string to read */
msg = malloc(sel[2] + 1);
if (NULL == msg) {
diediedie(2);
}
if (ORTE_SUCCESS !=
orte_notifier_command_read_fd(c->to_child[0], sel[2] + 1, msg)) {
diediedie(3);
/* What else can we do? */
}
/* We have all the info. Now build up the string command to
exec. Do the $<foo> replacements. */
cmd = strdup(c->cmd);
if ('\0' != *cmd) {
char *temp;
while (NULL != (p = strstr(cmd, "$s"))) {
*p = '\0';
asprintf(&temp, "%s%d%s", cmd, sel[0], p + 2);
free(cmd);
cmd = temp;
}
while (NULL != (p = strstr(cmd, "$S"))) {
*p = '\0';
asprintf(&temp, "%s%s%s", cmd,
orte_notifier_base_sev2str((orte_notifier_base_severity_t)sel[0]), p + 2);
free(cmd);
cmd = temp;
}
while (NULL != (p = strstr(cmd, "$e"))) {
*p = '\0';
asprintf(&temp, "%s%d%s", cmd, sel[1], p + 2);
free(cmd);
cmd = temp;
}
while (NULL != (p = strstr(cmd, "$m"))) {
*p = '\0';
asprintf(&temp, "%s%s%s", cmd, msg, p + 2);
free(cmd);
cmd = temp;
}
}
/* Now break it up into a list of argv */
if (ORTE_SUCCESS != orte_notifier_command_split(cmd, &argv)) {
diediedie(7);
/* What else can we do? */
}
/* Do we need a stdin pipe? */
if (mca_notifier_command_component.pass_via_stdin) {
if (0 != pipe(pipe_to_stdin)) {
diediedie(8);
}
}
/* Fork off the child and run the command */
pid = fork();
if (pid < 0) {
diediedie(8);
} else if (pid == 0) {
int i;
int fdmax = sysconf(_SC_OPEN_MAX);
close(0);
for (i = 3; i < fdmax; ++i) {
if (!mca_notifier_command_component.pass_via_stdin ||
pipe_to_stdin[0] != i) {
close(i);
}
}
/* If we have a pipe to stdin, dup it */
if (mca_notifier_command_component.pass_via_stdin) {
close(pipe_to_stdin[1]);
if (0 != pipe_to_stdin[0]) {
if (dup2(pipe_to_stdin[0], 0) < 0) {
diediedie(13);
}
close(pipe_to_stdin[0]);
}
}
/* Run it! */
execvp(argv[0], argv);
/* If we get here, bad */
diediedie(9);
}
/* Write down stdin. Start a thread because this has to run in
parallel to the timer to kill the grandchild if it runs too
long. */
if (mca_notifier_command_component.pass_via_stdin) {
close(pipe_to_stdin[0]);
OBJ_CONSTRUCT(&stdin_thread, opal_thread_t);
stdin_thread.t_run = stdin_main;
arg.sat_pipe_fd = pipe_to_stdin[1];
arg.sat_severity = sel[0];
arg.sat_errcode = sel[1];
arg.sat_msg = msg;
stdin_thread.t_arg = (void *) &arg;
if (OPAL_SUCCESS != opal_thread_start(&stdin_thread)) {
diediedie(9);
}
}
/* Parent: wait for / reap the child. */
do_wait(pid, mca_notifier_command_component.timeout, &status, &exited);
/* If the child didn't die, try killing it nicely. If that fails, kill
it dead. */
killed = false;
if (!exited) {
killed = true;
kill(pid, SIGTERM);
do_wait(pid, mca_notifier_command_component.timeout, &status, &exited);
if (!exited) {
kill(pid, SIGKILL);
do_wait(pid, mca_notifier_command_component.timeout, &status,
&exited);
}
}
/* Wait for the thread to complete */
if (mca_notifier_command_component.pass_via_stdin) {
void *ret;
close(pipe_to_stdin[1]);
opal_thread_join(&stdin_thread, &ret);
OBJ_DESTRUCT(&stdin_thread);
}
/* Free stuff */
free(cmd);
free(msg);
opal_argv_free(argv);
/* Handshake back up to the parent: just send the status value
back up to the parent and let all interpretation occur up
there. */
sel[0] = (int) exited;
sel[1] = (int) killed;
sel[2] = status;
if (ORTE_SUCCESS !=
orte_notifier_command_write_fd(mca_notifier_command_component.to_parent[1],
sizeof(sel), sel)) {
diediedie(11);
}
}
/*
* Main entry point for child
*/
void orte_notifier_command_child_main(void)
{
orte_notifier_command_pipe_cmd_t cmd;
orte_notifier_command_component_t *c = &mca_notifier_command_component;
while (1) {
/* Block waiting for a command */
cmd = CMD_MAX;
if (ORTE_SUCCESS !=
orte_notifier_command_read_fd(c->to_child[0], sizeof(cmd), &cmd)) {
diediedie(4);
}
switch (cmd) {
case CMD_EXEC:
do_exec();
break;
case CMD_TIME_TO_QUIT:
diediedie(0);
default:
diediedie(cmd + 50);
}
}
}

254
orte/mca/notifier/command/notifier_command_component.c
Просмотреть файл

@ -1,254 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2009 Sun Microsystems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/*
* Simple command notifier
*/
#include "orte_config.h"
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#ifdef HAVE_SYS_WAIT_H
#include <sys/wait.h>
#endif
#include <errno.h>
#include "opal/mca/base/mca_base_param.h"
#include "opal/util/argv.h"
#include "orte/constants.h"
#include "orte/runtime/orte_wait.h"
#include "orte/util/show_help.h"
#include "orte/util/proc_info.h"
#include "notifier_command.h"
static int command_component_query(mca_base_module_t **module, int *priority);
static int command_close(void);
static int command_register(void);
/*
* Struct of function pointers that need to be initialized
*/
orte_notifier_command_component_t mca_notifier_command_component = {
{
{
ORTE_NOTIFIER_BASE_VERSION_1_0_0,
"command",
ORTE_MAJOR_VERSION,
ORTE_MINOR_VERSION,
ORTE_RELEASE_VERSION,
NULL,
command_close,
command_component_query,
command_register,
},
{
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
}
},
/* Command command to run */
"/sbin/initlog -f $s -n \"Open MPI\" -s \"$S: $m (errorcode: $e)\"",
/* Timeout */
30,
/* Priority */
10,
/* PID of child */
-1,
/* To-child pipe FDs */
{ -1, -1 },
/* To-parent pipe FDs */
{ -1, -1 },
/* Pass via stdin? */
true,
};
/* Safety to ensure we don't try to write down a dead pipe */
static void child_death_cb(pid_t pid, int status, void *data)
{
if (pid == mca_notifier_command_component.child_pid) {
OPAL_OUTPUT((0, "Command notifier: child unexpectedly died! Exited, %d, exitstatus %d", WIFEXITED(status), WEXITSTATUS(status)));
mca_notifier_command_component.child_pid = 0;
mca_notifier_command_component.to_child[1] = -1;
}
}
static int command_register(void)
{
int val;
mca_base_param_reg_string(&mca_notifier_command_component.super.base_version,
"cmd",
"Command to execute, with substitution. $s = integer severity; $S = string severity; $e = integer error code; $m = string message",
false, false,
mca_notifier_command_component.cmd,
&mca_notifier_command_component.cmd);
mca_base_param_reg_int(&mca_notifier_command_component.super.base_version,
"timeout",
"Timeout (in seconds) of the command",
false, false,
mca_notifier_command_component.timeout,
&mca_notifier_command_component.timeout);
mca_base_param_reg_int(&mca_notifier_command_component.super.base_version,
"use_stdin",
"If true, pass parameters to the command via stdin, formatted with trivial XML",
false, false,
(int) mca_notifier_command_component.pass_via_stdin,
&val);
mca_notifier_command_component.pass_via_stdin = OPAL_INT_TO_BOOL(val);
/* Priority */
mca_base_param_reg_int(&mca_notifier_command_component.super.base_version,
"priority",
"Priority of this component",
false, false,
mca_notifier_command_component.priority,
&mca_notifier_command_component.priority);
return ORTE_SUCCESS;
}
static int command_close(void)
{
if (NULL != mca_notifier_command_component.cmd) {
free(mca_notifier_command_component.cmd);
}
/* Tell the child process to die */
if (0 != mca_notifier_command_component.child_pid &&
-1 != mca_notifier_command_component.to_child[1]) {
orte_notifier_command_pipe_cmd_t cmd = CMD_TIME_TO_QUIT;
orte_notifier_command_write_fd(mca_notifier_command_component.to_child[1],
sizeof(cmd), &cmd);
close(mca_notifier_command_component.to_child[1]);
mca_notifier_command_component.to_child[1] = -1;
close(mca_notifier_command_component.to_parent[0]);
mca_notifier_command_component.to_parent[0] = -1;
}
return ORTE_SUCCESS;
}
static int command_component_query(mca_base_module_t **module, int *priority)
{
char **argv = NULL;
*priority = 0;
*module = NULL;
/* If there's no command, there's no love */
if (NULL == mca_notifier_command_component.cmd ||
'\0' == mca_notifier_command_component.cmd[0]) {
orte_show_help("help-orte-notifier-command.txt",
"command not specified", true);
return ORTE_ERR_NOT_FOUND;
}
/* Attempt to parse the command into argv, just as a basic sanity
check to ensure that it seems to be ok. */
if (ORTE_SUCCESS !=
orte_notifier_command_split(mca_notifier_command_component.cmd, &argv)) {
orte_show_help("help-orte-notifier-command.txt",
"bad command", true, orte_process_info.nodename,
mca_notifier_command_component.cmd);
return ORTE_ERR_BAD_PARAM;
}
opal_argv_free(argv);
/* Create the pipe to be used (it'll be closed in component
close if we're not selected) */
if (0 != pipe(mca_notifier_command_component.to_child) ||
0 != pipe(mca_notifier_command_component.to_parent)) {
int save = errno;
orte_show_help("help-orte-notifier-command.txt",
"system call fail", true, orte_process_info.nodename,
"pipe", save, strerror(save));
errno = save;
return ORTE_ERR_IN_ERRNO;
}
/* Create the child (it'll be shut down in component close if
we're not selected). We create the child very early so that we
do it before any MPI networks are initialized that have
problems with fork(). The child sits on the other end of a
pipe and waits for commands from this main process. Commands
include telling the child to fork/exec a proces and shutting
down. */
mca_notifier_command_component.child_pid = fork();
if (mca_notifier_command_component.child_pid < 0) {
int save = errno;
orte_show_help("help-orte-notifier-command.txt",
"system call fail", true, orte_process_info.nodename,
"fork", save, strerror(save));
errno = save;
return ORTE_ERR_IN_ERRNO;
}
/* Child: close all fd's except the reading pipe and call the
child main routine */
if (0 == mca_notifier_command_component.child_pid) {
int i;
int fdmax = sysconf(_SC_OPEN_MAX);
for (i = 3; i < fdmax; ++i) {
if (i != mca_notifier_command_component.to_child[0] &&
i != mca_notifier_command_component.to_parent[1]) {
close(i);
}
}
orte_notifier_command_child_main();
/* Never returns */
}
/* Parent: close other ends of pipes */
close(mca_notifier_command_component.to_child[0]);
close(mca_notifier_command_component.to_parent[1]);
/* Let's find out if the child unexpectedly dies */
orte_wait_cb(mca_notifier_command_component.child_pid, child_death_cb, 0);
*priority = 10;
*module = (mca_base_module_t *) &orte_notifier_command_module;
return ORTE_SUCCESS;
}

82
orte/mca/notifier/command/notifier_command_fd.c
Просмотреть файл

@ -1,82 +0,0 @@
/*
* Copyright (c) 2008-2009 Cisco Systems, Inc. All rights reserved.
*
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* Note: this file is a little fast-n-loose with OPAL_HAVE_THREADS --
* it uses this value in run-time "if" conditionals (vs. compile-time
* #if conditionals). We also don't protect including <pthread.h>.
* That's because this component currently only compiles on Linux and
* Solaris, and both of these OS's have pthreads. Using the run-time
* conditionals gives us better compile-time checking, even of code
* that isn't activated.
*
* Note, too, that the functionality in this file does *not* require
* all the heavyweight OMPI thread infrastructure (e.g., from
* --enable-mpi-thread-multiple or --enable-progress-threads). All work that
* is done in a separate progress thread is very carefully segregated
* from that of the main thread, and communication back to the main
* thread
*/
#include "orte_config.h"
#include <unistd.h>
#include <errno.h>
#include "orte/constants.h"
#include "notifier_command.h"
/*
* Simple loop over reading from a fd
*/
int orte_notifier_command_read_fd(int fd, int len, void *buffer)
{
int rc;
char *b = buffer;
while (len > 0) {
rc = read(fd, b, len);
if (rc < 0 && EAGAIN == errno) {
continue;
} else if (rc > 0) {
len -= rc;
b += rc;
} else {
return ORTE_ERROR;
}
}
return ORTE_SUCCESS;
}
/*
* Simple loop over writing to an fd
*/
int orte_notifier_command_write_fd(int fd, int len, void *buffer)
{
int rc;
char *b = buffer;
while (len > 0) {
rc = write(fd, b, len);
if (rc < 0 && EAGAIN == errno) {
continue;
} else if (rc > 0) {
len -= rc;
b += rc;
} else {
return ORTE_ERROR;
}
}
return ORTE_SUCCESS;
}

196
orte/mca/notifier/command/notifier_command_module.c
Просмотреть файл

@ -1,196 +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 (c) 2007-2009 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/*
* Send an email upon notifier events.
*/
#include "orte_config.h"
#include <stdlib.h>
#include <sys/wait.h>
#include <stdio.h>
#include <string.h>
#ifdef HAVE_STDARG_H
#include <stdarg.h>
#endif
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#ifdef HAVE_SIGNAL_H
#include <signal.h>
#endif
#ifdef HAVE_SYS_WAIT_H
#include <sys/wait.h>
#endif
#include "opal/util/show_help.h"
#include "opal/util/error.h"
#include "orte/constants.h"
#include "orte/mca/ess/ess.h"
#include "orte/util/error_strings.h"
#include "orte/util/name_fns.h"
#include "orte/util/show_help.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/notifier/base/base.h"
#include "notifier_command.h"
static void command_log(orte_notifier_base_severity_t severity, int errcode,
const char *msg, va_list ap);
static void command_help(orte_notifier_base_severity_t severity, int errcode,
const char *filename,
const char *topic, va_list ap);
static void command_peer(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc,
const char *msg, va_list ap);
/* Module */
orte_notifier_base_module_t orte_notifier_command_module = {
NULL,
NULL,
command_log,
command_help,
command_peer,
NULL
};
/*
* Back-end function to actually tell the child to fork the command
*/
static int send_command(orte_notifier_base_severity_t severity, int errcode,
char *msg)
{
/* csel = Command, Severity, Errcode, string Length */
int rc, csel[4];
char *errmsg = NULL;
csel[0] = CMD_EXEC;
csel[1] = severity;
csel[2] = errcode;
csel[3] = strlen(msg);
/* Write the severity, errcode, and string length */
if (ORTE_SUCCESS !=
(rc = orte_notifier_command_write_fd(mca_notifier_command_component.to_child[1],
sizeof(csel), csel))) {
errmsg = "write";
goto error;
}
/* Now write the message itself */
if (ORTE_SUCCESS !=
(rc = orte_notifier_command_write_fd(mca_notifier_command_component.to_child[1],
csel[3] + 1, msg))) {
errmsg = "write";
goto error;
}
/* Now read back the grandchild's exit status from the child:
0 = 0/1 indicating whether the grandchild exited or not
1 = 0/1 indicating whether the grandchild timed out/was killed or not
2 = exit status returned by waitpid() (only relevant if exited==1) */
if (ORTE_SUCCESS !=
(rc = orte_notifier_command_read_fd(mca_notifier_command_component.to_parent[0],
sizeof(int) * 3, csel))) {
errmsg = "read";
goto error;
}
/* Did the grandchild exit? */
if (0 == csel[0]) {
orte_show_help("help-orte-notifier-command.txt",
"grandchild did not exit", true,
orte_process_info.nodename,
mca_notifier_command_component.cmd,
mca_notifier_command_component.timeout);
return ORTE_ERROR;
}
/* Did the grandchild timeout? */
if (1 == csel[1]) {
orte_show_help("help-orte-notifier-command.txt",
"grandchild timeout", true,
orte_process_info.nodename,
mca_notifier_command_component.cmd,
mca_notifier_command_component.timeout,
WIFEXITED(csel[0]) ? "Exit status" : "Signal",
WIFEXITED(csel[0]) ? WEXITSTATUS(csel[0]) : WTERMSIG(csel[0]));
return ORTE_ERR_TIMEOUT;
}
/* The grandchild exited in less than the timeout -- yay. Did it
exit cleanly? */
if (WIFEXITED(csel[1]) && 0 == WEXITSTATUS(csel[1])) {
return ORTE_SUCCESS;
}
/* Nope -- didn't exit cleanly, so print a warning. */
orte_show_help("help-orte-notifier-command.txt",
"grandchild fail", true, orte_process_info.nodename,
mca_notifier_command_component.cmd,
WIFEXITED(csel[0]) ? "Exit status" : "Signal",
WIFEXITED(csel[0]) ? WEXITSTATUS(csel[0]) : WTERMSIG(csel[0]));
return ORTE_ERROR;
error:
orte_show_help("help-orte-notifier-command.txt",
"system call fail", true, orte_process_info.nodename,
errmsg, opal_strerror(rc), rc);
return rc;
}
static void command_log(orte_notifier_base_severity_t severity, int errcode,
const char *msg, va_list ap)
{
char *output;
/* If there was a message, output it */
vasprintf(&output, msg, ap);
if (NULL != output) {
send_command(severity, errcode, output);
free(output);
}
}
static void command_help(orte_notifier_base_severity_t severity, int errcode,
const char *filename,
const char *topic, va_list ap)
{
char *output = opal_show_help_vstring(filename, topic, false, ap);
if (NULL != output) {
send_command(severity, errcode, output);
free(output);
}
}
static void command_peer(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc, const char *msg,
va_list ap)
{
char *buf = orte_notifier_base_peer_log(errcode, peer_proc, msg, ap);
if (NULL != buf) {
send_command(severity, errcode, buf);
free(buf);
}
}

35
orte/mca/notifier/configure.m4
Просмотреть файл

@ -1,35 +0,0 @@
dnl -*- shell-script -*-
dnl
dnl Copyright (c) 2004-2006 High Performance Computing Center Stuttgart,
dnl University of Stuttgart. All rights reserved.
dnl Copyright (c) 2008 Cisco Systems, Inc. All rights reserved.
dnl Copyright (c) 2009 Bull SAS. All rights reserved.
dnl $COPYRIGHT$
dnl
dnl Additional copyrights may follow
dnl
dnl $HEADER$
dnl
AC_DEFUN([MCA_orte_notifier_CONFIG],[
ompi_show_subsubtitle "Pre-configuring the framework notifier"
AC_MSG_CHECKING([if --enable-notifier-log-event was specified])
AC_ARG_ENABLE(notifier-log-event,
AC_HELP_STRING([--enable-notifier-log-event],
[Enable unusual events notification. (default: disabled)]))
if test "$enable_notifier_log_event" = "yes"; then
AC_MSG_RESULT([yes])
WANT_NOTIFIER_LOG_EVENT=1
else
AC_MSG_RESULT([no (disabling "notifier-log-event")])
WANT_NOTIFIER_LOG_EVENT=0
fi
AC_DEFINE_UNQUOTED([ORTE_WANT_NOTIFIER_LOG_EVENT],
[$WANT_NOTIFIER_LOG_EVENT],
[if the notifier_log_event should be enabled])
AM_CONDITIONAL([ORTE_WANT_NOTIFIER_LOG_EVENT],
[test "$WANT_NOTIFIER_LOG_EVENT" = "1"])
MCA_CONFIGURE_FRAMEWORK($1, $2, 1)
])

45
orte/mca/notifier/file/Makefile.am
Просмотреть файл

@ -1,45 +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 (c) 2009 Bull SAS. All rights reserved.
# Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
sources = \
notifier_file.h \
notifier_file_module.c \
notifier_file_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 MCA_BUILD_orte_notifier_file_DSO
component_noinst =
component_install = mca_notifier_file.la
else
component_noinst = libmca_notifier_file.la
component_install =
endif
mcacomponentdir = $(pkglibdir)
mcacomponent_LTLIBRARIES = $(component_install)
mca_notifier_file_la_SOURCES = $(sources)
mca_notifier_file_la_LDFLAGS = -module -avoid-version
noinst_LTLIBRARIES = $(component_noinst)
libmca_notifier_file_la_SOURCES =$(sources)
libmca_notifier_file_la_LDFLAGS = -module -avoid-version

19
orte/mca/notifier/file/configure.m4
Просмотреть файл

@ -1,19 +0,0 @@
# -*- shell-script -*-
#
# Copyright (c) 2011 Los Alamos National Security, LLC.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# MCA_notifier_file_CONFIG([action-if-found], [action-if-not-found])
# -----------------------------------------------------------
AC_DEFUN([MCA_orte_notifier_file_CONFIG], [
AC_CONFIG_FILES([orte/mca/notifier/file/Makefile])
AS_IF([test "$orte_without_full_support" = 0],
[$1],
[$2])
])

15
orte/mca/notifier/file/help-orte-notifier-file.txt
Просмотреть файл

@ -1,15 +0,0 @@
# -*- text -*-
#
# Copyright (c) 2009 Cisco Systems, Inc. All rights reserved.
# Copyright (c) 2009 Bull SAS. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# This is the US/English help file for Open MPI's SMTP notifier support
#
[file name not specified]
Error: the Open MPI file notifier component had no file name specified.
#

51
orte/mca/notifier/file/notifier_file.h
Просмотреть файл

@ -1,51 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2009 Bull SAS. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
*/
#ifndef NOTIFIER_FILE_H
#define NOTIFIER_FILE_H
#include "orte_config.h"
#include "orte/mca/notifier/notifier.h"
BEGIN_C_DECLS
typedef struct {
orte_notifier_base_component_t super;
/* File name the traces should be sent to */
char *fname;
/* Priority */
int priority;
} orte_notifier_file_component_t;
/*
* Notifier interfaces
*/
ORTE_MODULE_DECLSPEC extern orte_notifier_file_component_t mca_notifier_file_component;
extern orte_notifier_base_module_t orte_notifier_file_module;
END_C_DECLS
#endif /* NOTIFIER_FILE_H */

115
orte/mca/notifier/file/notifier_file_component.c
Просмотреть файл

@ -1,115 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2009 Bull SAS. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/** @file:
*
*/
/*
* includes
*/
#include "orte_config.h"
#include "orte/constants.h"
#ifdef HAVE_STRING_H
#include <string.h>
#endif /* HAVE_STRING_H */
#include "opal/mca/base/mca_base_param.h"
#include "orte/util/show_help.h"
#include "notifier_file.h"
static int orte_notifier_file_register(void);
static int orte_notifier_file_component_query(mca_base_module_t **, int *);
static int orte_notifier_file_close(void);
/*
* Struct of function pointers that need to be initialized
*/
orte_notifier_file_component_t mca_notifier_file_component = {
{
{
ORTE_NOTIFIER_BASE_VERSION_1_0_0,
"file", /* MCA module name */
ORTE_MAJOR_VERSION, /* MCA module major version */
ORTE_MINOR_VERSION, /* MCA module minor version */
ORTE_RELEASE_VERSION, /* MCA module release version */
NULL,
orte_notifier_file_close,
orte_notifier_file_component_query,
orte_notifier_file_register,
},
{
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
}
},
/* Log file name */
"wdc",
/* Priority */
10
};
static int orte_notifier_file_register(void)
{
mca_base_component_t *mcb = &mca_notifier_file_component.super.base_version;
orte_notifier_file_component_t *nfc = &mca_notifier_file_component;
mca_base_param_reg_string(mcb, "name",
"File name the traces should be redirected to",
false, false, nfc->fname, &nfc->fname);
mca_base_param_reg_int(mcb, "priority",
"Priority of the file notifier component",
false, false, nfc->priority, &nfc->priority);
return ORTE_SUCCESS;
}
static int orte_notifier_file_close(void)
{
if (NULL != mca_notifier_file_component.fname) {
free(mca_notifier_file_component.fname);
}
return ORTE_SUCCESS;
}
static int orte_notifier_file_component_query(mca_base_module_t **module,
int *priority)
{
*priority = mca_notifier_file_component.priority;
*module = (mca_base_module_t *)&orte_notifier_file_module;
if (NULL == mca_notifier_file_component.fname ||
!strlen(mca_notifier_file_component.fname)) {
orte_show_help("help-orte-notifier-file.txt",
"file name not specified", true);
return ORTE_ERR_BAD_PARAM;
}
return ORTE_SUCCESS;
}

237
orte/mca/notifier/file/notifier_file_module.c
Просмотреть файл

@ -1,237 +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 (c) 2007 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2009 Bull SAS. All rights reserved.
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif /* HAVE_UNISTD_H */
#ifdef HAVE_STDIO_H
#include <stdio.h>
#endif /* HAVE_STDIO_H */
#ifdef HAVE_STRING_H
#include <string.h>
#endif /* HAVE_STRING_H */
#ifdef HAVE_FCNTL_H
#include <fcntl.h>
#endif /* HAVE_FCNTL_H */
#ifdef HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif
#include "opal/util/os_path.h"
#include "opal/util/show_help.h"
#include "orte/mca/notifier/base/base.h"
#include "orte/util/proc_info.h"
#include "notifier_file.h"
/* Static API's */
static int init(void);
static void finalize(void);
static int open_file(void);
static void file_log(orte_notifier_base_severity_t severity, int errcode,
const char *msg, va_list ap);
static void file_helplog(orte_notifier_base_severity_t severity, int errcode,
const char *filename, const char *topic, va_list ap);
static void file_peerlog(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc, const char *msg,
va_list ap);
static void file_eventlog(const char *msg);
/* Module def */
orte_notifier_base_module_t orte_notifier_file_module = {
init,
finalize,
file_log,
file_helplog,
file_peerlog,
file_eventlog
};
static int mylogfd = -1;
static int init(void)
{
orte_notifier_file_component_t *comp = &mca_notifier_file_component;
if (!strcmp(comp->fname, "stdout")) {
mylogfd = fileno(stdout);
} else if (!strcmp(comp->fname, "stderr")) {
mylogfd = fileno(stderr);
}
/* Don't open in the case of a plain file: wait for the 1st write */
return ORTE_SUCCESS;
}
static void finalize(void)
{
if (-1 == mylogfd || fileno(stderr) == mylogfd
|| fileno(stdout) == mylogfd) {
return;
}
close(mylogfd);
}
static int open_file(void)
{
orte_notifier_file_component_t *comp = &mca_notifier_file_component;
char *full_name = NULL;
char *fname = NULL;
int rc = ORTE_SUCCESS;
if (-1 != mylogfd) {
return ORTE_SUCCESS;
}
asprintf(&fname, "output-%s", comp->fname);
if (NULL == fname) {
return ORTE_ERR_OUT_OF_RESOURCE;
}
full_name = opal_os_path(false, orte_process_info.job_session_dir,
fname, NULL);
if (NULL == full_name) {
rc = ORTE_ERR_OUT_OF_RESOURCE;
goto out_err;
}
mylogfd = open(full_name, O_CREAT | O_RDWR | O_APPEND, S_IRWXU);
if (-1 == mylogfd) {
rc = ORTE_ERR_FILE_OPEN_FAILURE;
}
free(full_name);
out_err:
free(fname);
return rc;
}
static void file_log(orte_notifier_base_severity_t severity, int errcode,
const char *msg, va_list ap)
{
char *output;
char *tmp = NULL;
/* Add a newline at the end of the format string */
asprintf(&tmp, "%s\n", msg);
if (NULL == tmp) {
return;
}
vasprintf(&output, tmp, ap);
free(tmp);
if (NULL == output) {
return;
}
/* If not done yet, open the log file */
if (-1 == mylogfd) {
if (ORTE_SUCCESS != open_file()) {
free(output);
return;
}
}
write(mylogfd, output, strlen(output));
fflush(NULL);
free(output);
}
static void file_helplog(orte_notifier_base_severity_t severity, int errcode,
const char *filename, const char *topic, va_list ap)
{
char *output = opal_show_help_vstring(filename, topic, false, ap);
if (NULL == output) {
return;
}
if (-1 == mylogfd) {
if (ORTE_SUCCESS != open_file()) {
free(output);
return;
}
}
write(mylogfd, output, strlen(output));
fflush(NULL);
free(output);
}
static void file_peerlog(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc, const char *msg,
va_list ap)
{
char *buf;
char *tmp = NULL;
/* Add a newline at the end of the format string */
asprintf(&tmp, "%s\n", msg);
if (NULL == tmp) {
return;
}
buf = orte_notifier_base_peer_log(errcode, peer_proc, tmp, ap);
free(tmp);
if (NULL == buf) {
return;
}
/* If not done yet, open the log file */
if (-1 == mylogfd) {
if (ORTE_SUCCESS != open_file()) {
free(buf);
return;
}
}
write(mylogfd, buf, strlen(buf));
fflush(NULL);
free(buf);
}
static void file_eventlog(const char *msg)
{
char *tmp = NULL;
/* Add a newline at the end of the string */
asprintf(&tmp, "%s\n", msg);
if (NULL == tmp) {
return;
}
/* If not done yet, open the log file */
if (-1 == mylogfd) {
if (ORTE_SUCCESS != open_file()) {
free(tmp);
return;
}
}
write(mylogfd, tmp, strlen(tmp));
fflush(NULL);
free(tmp);
}

52
orte/mca/notifier/ftb/Makefile.am
Просмотреть файл

@ -1,52 +0,0 @@
#
# Copyright (c) 2004-2009 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 (c) 2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
AM_CPPFLAGS = $(notifier_ftb_CPPFLAGS)
dist_pkgdata_DATA = \
help-orte-notifier-ftb.txt \
help-ftb-event-schema.txt
sources = \
notifier_ftb.h \
notifier_ftb_module.c \
notifier_ftb_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 MCA_BUILD_orte_notifier_ftb_DSO
component_noinst =
component_install = mca_notifier_ftb.la
else
component_noinst = libmca_notifier_ftb.la
component_install =
endif
mcacomponentdir = $(pkglibdir)
mcacomponent_LTLIBRARIES = $(component_install)
mca_notifier_ftb_la_SOURCES = $(sources)
mca_notifier_ftb_la_LDFLAGS = -module -avoid-version $(notifier_ftb_LDFLAGS)
mca_notifier_ftb_la_LIBADD = $(notifier_ftb_LIBS)
noinst_LTLIBRARIES = $(component_noinst)
libmca_notifier_ftb_la_SOURCES =$(sources)
libmca_notifier_ftb_la_LDFLAGS = -module -avoid-version $(notifier_ftb_LDFLAGS)
libmca_notifier_ftb_la_LIBADD = $(notifier_ftb_LIBS)

76
orte/mca/notifier/ftb/configure.m4
Просмотреть файл

@ -1,76 +0,0 @@
# -*- shell-script -*-
#
# Copyright (c) 2009 The Trustees of Indiana University and Indiana
# University Research and Technology
# Corporation. All rights reserved.
# Copyright (c) 2007 The University of Tennessee and The University
# of Tennessee Research Foundation. All rights
# reserved.
dnl Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# ORTE_CHECK_FTB(prefix, [action-if-found], [action-if-not-found])
# --------------------------------------------------------
# check if FTB (Fault Tolerance Backplane) support can be found. sets prefix_{CPPFLAGS,
# LDFLAGS, LIBS} as needed and runs action-if-found if there is
# support, otherwise executes action-if-not-found
AC_DEFUN([ORTE_CHECK_FTB],[
AC_ARG_WITH([ftb],
[AC_HELP_STRING([--with-ftb(=DIR)],
[Build FTB (Fault Tolerance Backplane) support, searching for libraries in DIR])])
AC_ARG_WITH([ftb-libdir],
[AC_HELP_STRING([--with-ftb-libdir=DIR],
[Search for FTB (Fault Tolerance Backplane) libraries in DIR])])
AS_IF([test "$with_ftb" != "no"],
[AS_IF([test ! -z "$with_ftb" -a "$with_ftb" != "yes"],
[ompi_check_ftb_dir="$with_ftb"])
AS_IF([test ! -z "$with_ftb_libdir" -a "$with_ftb_libdir" != "yes"],
[ompi_check_ftb_libdir="$with_ftb_libdir"])
OMPI_CHECK_PACKAGE([$1],
[libftb.h],
[ftb],
[FTB_Connect],
[],
[$ompi_check_ftb_dir],
[$ompi_check_ftb_libdir],
[ompi_check_ftb_happy="yes"],
[ompi_check_ftb_happy="no"])
],
[ompi_check_ftb_happy="no"])
AS_IF([test "$ompi_check_ftb_happy" = "yes"],
[$2],
[AS_IF([test ! -z "$with_ftb" -a "$with_ftb" != "no"],
[AC_MSG_ERROR([FTB (Fault Tolerance Backplane) support requested but not found. Aborting])])
$3])
])
# MCA_notifier_ftb_CONFIG([action-if-found], [action-if-not-found])
# -----------------------------------------------------------
AC_DEFUN([MCA_orte_notifier_ftb_CONFIG], [
AC_CONFIG_FILES([orte/mca/notifier/ftb/Makefile])
ORTE_CHECK_FTB([notifier_ftb],
[notifier_ftb_happy="yes"],
[notifier_ftb_happy="no"])
AS_IF([test "$notifier_ftb_happy" = "yes" -a "$orte_without_full_support" = 0],
[notifier_ftb_WRAPPER_EXTRA_LDFLAGS="$notifier_ftb_LDFLAGS"
notifier_ftb_WRAPPER_EXTRA_LIBS="$notifier_ftb_LIBS"
$1],
[$2])
# substitute in the things needed to build ftb
AC_SUBST([notifier_ftb_CFLAGS])
AC_SUBST([notifier_ftb_CPPFLAGS])
AC_SUBST([notifier_ftb_LDFLAGS])
AC_SUBST([notifier_ftb_LIBS])
])dnl

29
orte/mca/notifier/ftb/help-ftb-event-schema.txt
Просмотреть файл

@ -1,29 +0,0 @@
# -*- text -*-
#
# Copyright (c) 2010 The Trustees of Indiana University and Indiana
# University Research and Technology Corporation.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# This is the FTB event schema file for Open MPI's FTB notifier
#
start
ftb.mpi.openmpi
FTB_MPI_PROCS_DEAD ERROR
FTB_MPI_PROCS_UNREACHABLE ERROR
FTB_MPI_PROCS_COMM_ERROR WARNING
FTB_MPI_PROCS_MIGRATED INFO
FTB_MPI_PROCS_MIGRATE_FAIL ERROR
FTB_MPI_PROCS_CKPTED INFO
FTB_MPI_PROCS_CKPT_FAIL ERROR
FTB_MPI_PROCS_RESTARTED INFO
FTB_MPI_PROCS_RESTART_FAIL ERROR
FTB_MPI_PROCS_ABORTED ERROR
end

45
orte/mca/notifier/ftb/help-orte-notifier-ftb.txt
Просмотреть файл

@ -1,45 +0,0 @@
# -*- text -*-
#
# Copyright (c) 2009 Cisco Systems, Inc. All rights reserved.
# Copyright (c) 2009 The Trustees of Indiana University and Indiana
# University Research and Technology Corporation.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# This is the US/English help file for Open MPI's FTB notifier support
#
[ftb connect failed]
Open MPI's FTB notifier component failed to connect to the FTB server.
Check if the FTB bootstrap server is running or not. For further help,
refer the FTB documentation (Section 4.0: RUNNING FTB).
Reason: %s (errno: %d)
#
[declare events failed]
The Open MPI FTB notifier component failed to declare publishable events
to the FTB.
Reason: %s (errno: %d)
#
[publish failed]
Sorry, Open MPI's FTB component failed to publish the following event to
the FTB.
Reason: %s (errno: %d)
Event info: [%s] %s
Event properties: %s (errno: %d)
#
[invalid subscription style]
Error: the Open MPI FTB component tried to register with an invalid
FTB client subscription style.
Subscription style: %s
#
[invalid value]
Error: the Open MPI FTB notifier component tried to register with an
invalid value in the FTB client information.
#

61
orte/mca/notifier/ftb/notifier_ftb.h
Просмотреть файл

@ -1,61 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2009 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
*/
#ifndef NOTIFIER_FTB_H
#define NOTIFIER_FTB_H
#include "orte_config.h"
#include "orte/mca/notifier/notifier.h"
#include "libftb.h"
BEGIN_C_DECLS
typedef struct {
orte_notifier_base_component_t super;
/* FTB client subscription style */
char *subscription_style;
/* Priority of this component */
int priority;
} orte_notifier_ftb_component_t;
/* Notifier interfaces */
ORTE_MODULE_DECLSPEC extern orte_notifier_ftb_component_t mca_notifier_ftb_component;
extern orte_notifier_base_module_t orte_notifier_ftb_module;
/* FTB client information */
extern FTB_client_t ftb_client_info;
extern FTB_client_handle_t ftb_client_handle;
/* FTB event types */
typedef enum {
FTB_EVENT_NORMAL = 1,
FTB_EVENT_RESPONSE = 2
} ftb_event_type_t;
/* Returns the FTB event name (as a string) given the event code */
#define FTB_EVENT(errnum) #errnum
END_C_DECLS
#endif

160
orte/mca/notifier/ftb/notifier_ftb_component.c
Просмотреть файл

@ -1,160 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2009 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/** @file:
*
* This component proxies notification events to the Fault Tolerant
* Backplane (See http://www.mcs.anl.gov/research/cifts/).
* The ORTE notifier severity is translated to the corresponding
* FTB severity before the event is published to the FTB.
*/
/*
* includes
*/
#include "orte_config.h"
#include "orte/constants.h"
#include <string.h>
#include "orte/util/show_help.h"
#include "orte/runtime/orte_globals.h"
#include "opal/mca/base/mca_base_param.h"
#include "notifier_ftb.h"
static int orte_notifier_ftb_close(void);
static int orte_notifier_ftb_component_query(mca_base_module_t **module, int *priority);
static int orte_notifier_ftb_register(void);
/*
* Struct of function pointers that need to be initialized
*/
orte_notifier_ftb_component_t mca_notifier_ftb_component = {
{
{
ORTE_NOTIFIER_BASE_VERSION_1_0_0,
"ftb", /* MCA module name */
ORTE_MAJOR_VERSION, /* MCA module major version */
ORTE_MINOR_VERSION, /* MCA module minor version */
ORTE_RELEASE_VERSION, /* MCA module release version */
NULL,
orte_notifier_ftb_close, /* module close */
orte_notifier_ftb_component_query, /* module query */
orte_notifier_ftb_register, /* module register */
},
{
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
}
},
/* FTB client subscription style */
"FTB_SUBSCRIPTION_NONE",
/* Priority */
10,
};
static int orte_notifier_ftb_close(void)
{
if (NULL != mca_notifier_ftb_component.subscription_style) {
free(mca_notifier_ftb_component.subscription_style);
}
return ORTE_SUCCESS;
}
static int orte_notifier_ftb_component_query(mca_base_module_t **module,
int *priority)
{
int ret;
*priority = 0;
*module = NULL;
/* Fill the FTB client information structure */
memset(&ftb_client_info, 0, sizeof(ftb_client_info));
strcpy(ftb_client_info.event_space, "ftb.mpi.openmpi");
/* We represent each client with a client name of the form
openmpi/<hostname>/<PID> as a unique identifier in the
FTB client namespace */
sprintf(ftb_client_info.client_name, "ompi%u", orte_process_info.pid);
sprintf(ftb_client_info.client_jobid, "%u", ORTE_PROC_MY_NAME->jobid);
strncpy(ftb_client_info.client_subscription_style,
mca_notifier_ftb_component.subscription_style,
strlen(mca_notifier_ftb_component.subscription_style));
/* We try to connect to the FTB backplane now, and we abort
if we cannot connect for some reason. */
if (FTB_SUCCESS != (ret = FTB_Connect(&ftb_client_info, &ftb_client_handle))) {
switch (ret) {
case FTB_ERR_SUBSCRIPTION_STYLE:
orte_show_help("help-orte-notifier-ftb.txt",
"invalid subscription style",
true, ftb_client_info.client_subscription_style);
break;
case FTB_ERR_INVALID_VALUE:
orte_show_help("help-orte-notifier-ftb.txt",
"invalid value",
true);
break;
default:
orte_show_help("help-orte-notifier-ftb.txt",
"ftb connect failed",
true);
}
return ORTE_ERR_NOT_FOUND;
}
*priority = 10;
*module = (mca_base_module_t *)&orte_notifier_ftb_module;
return ORTE_SUCCESS;
}
static int orte_notifier_ftb_register(void)
{
/* FTB client subscription style */
mca_base_param_reg_string(&mca_notifier_ftb_component.super.base_version,
"subscription_style",
"FTB client subscription style. "
"Possible values are none, polling, notify and both (polling and notify).",
false, false,
mca_notifier_ftb_component.subscription_style,
&mca_notifier_ftb_component.subscription_style);
/* Priority */
mca_base_param_reg_int(&mca_notifier_ftb_component.super.base_version,
"priority",
"Priority of this component",
false, false,
mca_notifier_ftb_component.priority,
&mca_notifier_ftb_component.priority);
return ORTE_SUCCESS;
}

294
orte/mca/notifier/ftb/notifier_ftb_module.c
Просмотреть файл

@ -1,294 +0,0 @@
/*
* Copyright (c) 2004-2010 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 (c) 2007 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include <stdio.h>
#include <string.h>
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif /* HAVE_SYS_TIME_H */
#ifdef HAVE_STDARG_H
#include <stdarg.h>
#endif
#include "opal/mca/installdirs/installdirs.h"
#include "opal/util/show_help.h"
#include "opal/util/os_path.h"
#include "orte/mca/plm/base/plm_private.h"
#include "orte/mca/plm/plm.h"
#include "orte/mca/sensor/sensor.h"
#include "orte/mca/ess/ess.h"
#include "orte/util/show_help.h"
#include "orte/mca/snapc/snapc.h"
#include "orte/mca/snapc/base/base.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/errmgr/base/base.h"
#include "orte/mca/notifier/base/base.h"
#include "notifier_ftb.h"
/* Static API's */
static int init(void);
static void finalize(void);
static void ftb_log(orte_notifier_base_severity_t severity, int errcode,
const char *msg, va_list ap);
static void ftb_help(orte_notifier_base_severity_t severity, int errcode,
const char *filename, const char *topic, va_list ap);
static void ftb_peer(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc, const char *msg,
va_list ap);
/* Module def */
orte_notifier_base_module_t orte_notifier_ftb_module = {
init,
finalize,
ftb_log,
ftb_help,
ftb_peer,
NULL
};
/* FTB client information */
FTB_client_t ftb_client_info;
/* FTB client handle */
FTB_client_handle_t ftb_client_handle;
static int init(void) {
int ret;
char *schema_file;
/* Locate the FTB events schema file */
if (NULL == (schema_file = opal_os_path(false, opal_install_dirs.pkgdatadir,
"help-ftb-event-schema.txt", NULL))) {
schema_file = strdup("help-ftb-event-schema.txt");
}
/* Declare the Open MPI publishable events to the FTB */
ret = FTB_Declare_publishable_events(ftb_client_handle, schema_file, NULL, 0);
free(schema_file);
if (FTB_SUCCESS != ret) {
orte_show_help("help-orte-notifier-ftb.txt", "declare events failed", true,
"FTB_Declare_publishable_events() failed", ret);
FTB_Disconnect(ftb_client_handle);
return ORTE_ERROR;
}
return ORTE_SUCCESS;
}
static void finalize(void) {
/* If the FTB client handle is valid, disconnect the client from FTB. */
if (1 == ftb_client_handle.valid) {
FTB_Disconnect(ftb_client_handle);
}
}
static const char* get_ftb_event_severity(orte_notifier_base_severity_t severity)
{
switch (severity) {
case ORTE_NOTIFIER_EMERG:
case ORTE_NOTIFIER_ALERT:
return "ALL";
case ORTE_NOTIFIER_CRIT:
return "FATAL";
case ORTE_NOTIFIER_ERROR:
return "ERROR";
case ORTE_NOTIFIER_WARN:
case ORTE_NOTIFIER_NOTICE:
return "WARNING";
case ORTE_NOTIFIER_INFO:
case ORTE_NOTIFIER_DEBUG:
return "INFO";
default:
return "UNKNOWN";
}
}
static const char* get_ftb_event_name(int errnum)
{
/* Handle checkpoint/restart and migration events */
if ( CHECK_ORTE_SNAPC_CKPT_STATE(errnum) ) {
errnum = ORTE_SNAPC_CKPT_STATE(errnum);
switch (errnum) {
case ORTE_SNAPC_CKPT_STATE_ESTABLISHED:
return FTB_EVENT(FTB_MPI_PROCS_CKPTED);
case ORTE_SNAPC_CKPT_STATE_NO_CKPT:
case ORTE_SNAPC_CKPT_STATE_ERROR:
return FTB_EVENT(FTB_MPI_PROCS_CKPT_FAIL);
/* Restart events */
case ORTE_SNAPC_CKPT_STATE_RECOVERED:
return FTB_EVENT(FTB_MPI_PROCS_RESTARTED);
case ORTE_SNAPC_CKPT_STATE_NO_RESTART:
return FTB_EVENT(FTB_MPI_PROCS_RESTART_FAIL);
/* Process migration events */
case ORTE_ERRMGR_MIGRATE_STATE_FINISH:
return FTB_EVENT(FTB_MPI_PROCS_MIGRATED);
case ORTE_ERRMGR_MIGRATE_STATE_ERROR:
case ORTE_ERRMGR_MIGRATE_STATE_ERR_INPROGRESS:
return FTB_EVENT(FTB_MPI_PROCS_MIGRATE_FAIL);
default:
return NULL;
}
} else {
/* Handle process and communication failure events */
switch (errnum) {
case ORTE_ERR_CONNECTION_REFUSED:
case ORTE_ERR_CONNECTION_FAILED:
case ORTE_ERR_UNREACH:
case ORTE_PROC_STATE_HEARTBEAT_FAILED:
return FTB_EVENT(FTB_MPI_PROCS_UNREACHABLE);
case ORTE_ERR_COMM_FAILURE:
case ORTE_PROC_STATE_COMM_FAILED:
return FTB_EVENT(FTB_MPI_PROCS_COMM_ERROR);
case ORTE_PROC_STATE_FAILED_TO_START:
case ORTE_PROC_STATE_CALLED_ABORT:
return FTB_EVENT(FTB_MPI_PROCS_ABORTED);
case ORTE_PROC_STATE_ABORTED:
case ORTE_PROC_STATE_ABORTED_BY_SIG:
case ORTE_PROC_STATE_TERM_WO_SYNC:
case ORTE_PROC_STATE_TERMINATED:
case ORTE_PROC_STATE_KILLED_BY_CMD:
return FTB_EVENT(FTB_MPI_PROCS_DEAD);
default:
return NULL;
}
}
return NULL;
}
/* Extracts the FTB payload (inside the brackets []) from notifier
* message payload.
* For instance: "<FTB message [payload]>" would return "payload".
*/
static unsigned int extract_payload(char *dest, char *src, unsigned int size)
{
unsigned int ret;
char *lbrace, *rbrace;
rbrace = strrchr(src, ']');
lbrace = strchr(src, '[');
if (NULL == rbrace || NULL == lbrace) {
strncpy(dest, src, size);
ret = size;
} else {
ret = rbrace - lbrace + 1;
if (ret > size) {
ret = size;
}
strncpy(dest, lbrace, ret);
}
return ret;
}
static void publish_ftb_event(orte_notifier_base_severity_t severity, int errcode,
FTB_event_properties_t *eprop)
{
int ret;
const char *event_name;
FTB_event_handle_t ehandle;
/* Publish the event to the Fault Tolerant Backplane */
event_name = get_ftb_event_name(errcode);
if (NULL != event_name) {
ret = FTB_Publish(ftb_client_handle, event_name, eprop, &ehandle);
if (FTB_SUCCESS != ret) {
orte_show_help("help-orte-notifier-ftb.txt", "publish failed", true,
"FTB_Publish() failed", ret, get_ftb_event_severity(severity),
event_name, eprop->event_payload, errcode);
}
}
}
static void ftb_log(orte_notifier_base_severity_t severity, int errcode, const char *msg,
va_list ap)
{
char *payload;
FTB_event_properties_t ev_prop;
/* Only normal FTB events are supported currently. */
ev_prop.event_type = (int) FTB_EVENT_NORMAL;
/* Copy the event payload, if we have one */
vasprintf(&payload, msg, ap);
if (NULL != payload) {
extract_payload(ev_prop.event_payload, payload, FTB_MAX_PAYLOAD_DATA);
free(payload);
publish_ftb_event(severity, errcode, &ev_prop);
}
}
static void ftb_help(orte_notifier_base_severity_t severity, int errcode,
const char *filename, const char *topic, va_list ap)
{
char *payload;
FTB_event_properties_t ev_prop;
/* Only normal FTB events are supported currently. */
ev_prop.event_type = (int) FTB_EVENT_NORMAL;
payload = opal_show_help_vstring(filename, topic, false, ap);
if (NULL != payload) {
extract_payload(ev_prop.event_payload, payload, FTB_MAX_PAYLOAD_DATA);
free(payload);
publish_ftb_event(severity, errcode, &ev_prop);
}
}
static void ftb_peer(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc, const char *msg,
va_list ap)
{
char *payload, *peer_host;
FTB_event_properties_t ev_prop;
/* Only normal FTB events are supported currently. */
ev_prop.event_type = (int) FTB_EVENT_NORMAL;
peer_host = NULL;
if (peer_proc) {
peer_host = orte_ess.proc_get_hostname(peer_proc);
/* Ignore the peer_host for now. */
}
vasprintf(&payload, msg, ap);
if (NULL != payload) {
extract_payload(ev_prop.event_payload, payload, FTB_MAX_PAYLOAD_DATA);
free(payload);
publish_ftb_event(severity, errcode, &ev_prop);
}
}

49
orte/mca/notifier/hnp/Makefile.am
Просмотреть файл

@ -1,49 +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 (c) 2009-2010 Cisco Systems, Inc. All rights reserved.
# Copyright (c) 2012 Los Alamos National Security, LLC.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
EXTRA_DIST = orte_notifier_hnp.txt
sources = \
notifier_hnp.h \
notifier_hnp_module.c \
notifier_hnp_recv.c \
notifier_hnp_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 MCA_BUILD_orte_notifier_hnp_DSO
component_noinst =
component_install = mca_notifier_hnp.la
else
component_noinst = libmca_notifier_hnp.la
component_install =
endif
mcacomponentdir = $(pkglibdir)
mcacomponent_LTLIBRARIES = $(component_install)
mca_notifier_hnp_la_SOURCES = $(sources)
mca_notifier_hnp_la_LDFLAGS = -module -avoid-version
noinst_LTLIBRARIES = $(component_noinst)
libmca_notifier_hnp_la_SOURCES =$(sources)
libmca_notifier_hnp_la_LDFLAGS = -module -avoid-version

19
orte/mca/notifier/hnp/configure.m4
Просмотреть файл

@ -1,19 +0,0 @@
# -*- shell-script -*-
#
# Copyright (c) 2011 Los Alamos National Security, LLC.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# MCA_notifier_hnp_CONFIG([action-if-found], [action-if-not-found])
# -----------------------------------------------------------
AC_DEFUN([MCA_orte_notifier_hnp_CONFIG], [
AC_CONFIG_FILES([orte/mca/notifier/hnp/Makefile])
AS_IF([test "$orte_without_full_support" = 0],
[$1],
[$2])
])

51
orte/mca/notifier/hnp/notifier_hnp.h
Просмотреть файл

@ -1,51 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
*/
#ifndef NOTIFIER_HNP_H
#define NOTIFIER_HNP_H
#include "orte_config.h"
#include "orte/types.h"
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/rml/rml.h"
#include "opal/class/opal_pointer_array.h"
BEGIN_C_DECLS
void orte_notifier_hnp_recv_cb(int status, orte_process_name_t* sender,
opal_buffer_t* buffer, orte_rml_tag_t tag,
void* cbdata);
/*
extern opal_pointer_array_t orte_notifier_hnp_tables;
extern opal_mutex_t orte_notifier_hnp_tables_lock;
*/
/*
* Notifier interfaces
*/
ORTE_MODULE_DECLSPEC extern orte_notifier_base_component_t mca_notifier_hnp_component;
extern orte_notifier_base_module_t orte_notifier_hnp_module;
END_C_DECLS
#endif

71
orte/mca/notifier/hnp/notifier_hnp_component.c
Просмотреть файл

@ -1,71 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* @file
*
* This component proxies notification messages up to the HNP. This
* component runs in both the HNP and non-HNP processes for ease of
* selection (e.g., so you can "--mca notifier hnp" (vs. "--mca
* notifier hnp,non_hnp"). It auto-detects where it is running and
* does the Right Thing -- if it's in the HNP process, it sets up to
* receive incoming proxied messages. If it's not in the HNP, then it
* proxies all messages to the HNP.
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "notifier_hnp.h"
static int orte_notifier_hnp_component_query(mca_base_module_t **module,
int *priority);
/*
* Struct of function pointers that need to be initialized
*/
orte_notifier_base_component_t mca_notifier_hnp_component = {
{
ORTE_NOTIFIER_BASE_VERSION_1_0_0,
"hnp", /* MCA module name */
ORTE_MAJOR_VERSION, /* MCA module major version */
ORTE_MINOR_VERSION, /* MCA module minor version */
ORTE_RELEASE_VERSION, /* MCA module release version */
NULL,
NULL,
orte_notifier_hnp_component_query /* module query */
},
{
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
}
};
static int orte_notifier_hnp_component_query(mca_base_module_t **module,
int *priority)
{
*priority = 10;
*module = (mca_base_module_t *)&orte_notifier_hnp_module;
return ORTE_SUCCESS;
}

205
orte/mca/notifier/hnp/notifier_hnp_module.c
Просмотреть файл

@ -1,205 +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 (c) 2007 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2009 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011-2012 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include <string.h>
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif /* HAVE_SYS_TIME_H */
#ifdef HAVE_HNP_H
#include <hnp.h>
#endif
#ifdef HAVE_STDARG_H
#include <stdarg.h>
#endif
#include "opal/util/show_help.h"
#include "opal/dss/dss.h"
#include "opal/dss/dss_types.h"
#include "orte/util/show_help.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/notifier/base/base.h"
#include "notifier_hnp.h"
/* Static API's */
static int init(void);
static void finalize(void);
static void mylog(orte_notifier_base_severity_t severity, int errcode,
const char *msg, va_list ap);
static void myhelplog(orte_notifier_base_severity_t severity, int errcode,
const char *filename, const char *topic, va_list ap);
static void mypeerlog(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc, const char *msg,
va_list ap);
static void myeventlog(const char *msg);
/* Module definition */
orte_notifier_base_module_t orte_notifier_hnp_module = {
init,
finalize,
mylog,
myhelplog,
mypeerlog,
myeventlog
};
static int send_command(orte_notifier_base_severity_t severity, int errcode,
char *msg)
{
opal_buffer_t *buf;
int rc;
uint8_t u8 = (uint8_t) severity;
uint32_t u32 = (uint32_t) errcode;
buf = OBJ_NEW(opal_buffer_t);
if (NULL == buf) {
return ORTE_ERR_OUT_OF_RESOURCE;
}
/* Pack the severity (need to use a fixed-size type) */
if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, &u8, 1, OPAL_UINT8))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buf);
return rc;
}
/* Pack the errcode (need to use a fixed-size type) */
if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, &u32, 1, OPAL_UINT32))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buf);
return rc;
}
/* Pack the message */
if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, &msg, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buf);
return rc;
}
/* Now send the buffer (rc = number of bytes sent) */
rc = orte_rml.send_buffer(ORTE_PROC_MY_HNP, buf,
ORTE_RML_TAG_NOTIFIER_HNP, 0);
if (rc <= 0) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buf);
return rc;
}
return ORTE_SUCCESS;
}
static int init(void)
{
int rc;
/* If I'm the HNP, post a non-blocking RML receive */
if (ORTE_PROC_IS_HNP) {
if (ORTE_SUCCESS !=
(rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_NOTIFIER_HNP,
ORTE_RML_PERSISTENT,
orte_notifier_hnp_recv_cb,
NULL))) {
ORTE_ERROR_LOG(rc);
return rc;
}
}
return ORTE_SUCCESS;
}
static void finalize(void)
{
/* If I'm the HNP, then cancel the non-blocking RML receive */
if (ORTE_PROC_IS_HNP) {
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_NOTIFIER_HNP);
}
}
static void mylog(orte_notifier_base_severity_t severity, int errcode,
const char *msg, va_list ap)
{
char *output;
/* If there was a message, output it */
vasprintf(&output, msg, ap);
if (NULL != output) {
if (ORTE_PROC_IS_HNP) {
/* output it locally */
orte_show_help("orte_notifier_hnp.txt", "notifier message", false, output);
} else {
send_command(severity, errcode, output);
}
free(output);
}
}
static void myhelplog(orte_notifier_base_severity_t severity, int errcode,
const char *filename, const char *topic, va_list ap)
{
char *output;
output = opal_show_help_vstring(filename, topic, false, ap);
if (NULL != output) {
if (ORTE_PROC_IS_HNP) {
/* output it locally */
orte_show_help("orte_notifier_hnp.txt", "notifier message", false, output);
} else {
send_command(severity, errcode, output);
}
free(output);
}
}
static void mypeerlog(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc, const char *msg,
va_list ap)
{
char *buf = orte_notifier_base_peer_log(errcode, peer_proc, msg, ap);
if (NULL != buf) {
if (ORTE_PROC_IS_HNP) {
/* output it locally */
orte_show_help("orte_notifier_hnp.txt", "notifier message", false, buf);
} else {
send_command(severity, errcode, buf);
}
free(buf);
}
}
static void myeventlog(const char *msg)
{
if (ORTE_PROC_IS_HNP) {
/* output it locally */
orte_show_help("orte_notifier_hnp.txt", "notifier message", false, (char*)msg);
} else {
send_command(ORTE_NOTIFIER_NOTICE, ORTE_SUCCESS, (char *)msg);
}
}

76
orte/mca/notifier/hnp/notifier_hnp_recv.c
Просмотреть файл

@ -1,76 +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 (c) 2007 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2009 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011-2012 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/notifier/base/base.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/util/show_help.h"
#include "opal/class/opal_hash_table.h"
#include "notifier_hnp.h"
void orte_notifier_hnp_recv_cb(int status, orte_process_name_t* sender,
opal_buffer_t* buffer, orte_rml_tag_t tag,
void* cbdata)
{
uint8_t u8;
uint32_t u32;
int rc, count;
orte_notifier_base_severity_t severity;
int errcode;
char *msg;
/* Unpack the severity */
count = 1;
if (ORTE_SUCCESS !=
(rc = opal_dss.unpack(buffer, &u8, &count, OPAL_UINT8))) {
ORTE_ERROR_LOG(rc);
goto CLEAN_RETURN;
}
severity = (orte_notifier_base_severity_t) u8;
/* Unpack the errcode */
count = 1;
if (ORTE_SUCCESS !=
(rc = opal_dss.unpack(buffer, &u32, &count, OPAL_UINT32))) {
ORTE_ERROR_LOG(rc);
goto CLEAN_RETURN;
}
errcode = (int) u32;
/* Unpack the string */
count = 1;
if (ORTE_SUCCESS !=
(rc = opal_dss.unpack(buffer, &msg, &count, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
goto CLEAN_RETURN;
}
orte_show_help("orte_notifier_hnp.txt", "notifier message", false, msg);
CLEAN_RETURN:
return;
}

36
orte/mca/notifier/hnp/orte_notifier_hnp.txt
Просмотреть файл

@ -1,36 +0,0 @@
# -*- text -*-
#
# 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 (c) 2008 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# This is the US/English help file for HNP notifier messages.
#
# FORMAT:
# <severity> filename:linenum:functionname
# error message string
# stacktrace (optional)
[general message]
| |--<%s> at %s:%d:%s():
| | %s
#
[msg header]
| %s
#
# We marshall all the parameters into a single message when we
# relay it to the notifier.
[notifier message]
%s

180
orte/mca/notifier/notifier.h
Просмотреть файл

@ -1,180 +0,0 @@
/*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All Rights Reserved.
* Copyright (c) 2012 Los Alamos National Security, LLC.
* All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/** @file:
*
* The OpenRTE Notifier Framework
*
* The OpenRTE Notifier framework provides a mechanism for notifying
* system administrators or other fault monitoring systems that a
* problem with the underlying cluster has been detected - e.g., a
* failed connection in a network fabric
*/
#ifndef MCA_NOTIFIER_H
#define MCA_NOTIFIER_H
/*
* includes
*/
#include "orte_config.h"
#ifdef HAVE_STDARG_H
#include <stdarg.h>
#endif
#ifdef HAVE_LIMITS_H
#include <limits.h>
#endif
#ifdef HAVE_SYSLOG_H
#include <syslog.h>
#endif
#include "opal/mca/mca.h"
#include "orte/constants.h"
#include "orte/types.h"
#include "notifier_event_types.h"
BEGIN_C_DECLS
/* The maximum size of any on-stack buffers used in the notifier
* so we can try to avoid calling malloc in OUT_OF_RESOURCES conditions.
* The code has NOT been auditied for use of malloc, so this still
* may fail to get the "OUT_OF_RESOURCE" message out. Oh Well.
*/
#define ORTE_NOTIFIER_MAX_BUF 512
/* Severities */
typedef enum {
ORTE_NOTIFIER_EMERG = LOG_EMERG,
ORTE_NOTIFIER_ALERT = LOG_ALERT,
ORTE_NOTIFIER_CRIT = LOG_CRIT,
ORTE_NOTIFIER_ERROR = LOG_ERR,
ORTE_NOTIFIER_WARN = LOG_WARNING,
ORTE_NOTIFIER_NOTICE = LOG_NOTICE,
ORTE_NOTIFIER_INFO = LOG_INFO,
ORTE_NOTIFIER_DEBUG = LOG_DEBUG
} orte_notifier_base_severity_t;
/*
* Component functions - all MUST be provided!
*/
/* initialize the selected module */
typedef int (*orte_notifier_base_module_init_fn_t)(void);
/* finalize the selected module */
typedef void (*orte_notifier_base_module_finalize_fn_t)(void);
/* Log a failure message */
typedef void (*orte_notifier_base_module_log_fn_t)(orte_notifier_base_severity_t severity, int errcode, const char *msg, va_list ap)
__opal_attribute_format_funcptr__(__printf__, 3, 0);
/* Log a failure that is based upon a show_help message */
typedef void (*orte_notifier_base_module_log_show_help_fn_t)(orte_notifier_base_severity_t severity, int errcode, const char *file, const char *topic, va_list ap);
/* Log a failure related to a peer */
typedef void (*orte_notifier_base_module_log_peer_fn_t)(orte_notifier_base_severity_t severity, int errcode, orte_process_name_t *peer_proc, const char *msg, va_list ap)
__opal_attribute_format_funcptr__(__printf__, 4, 0);
/* Log an unusual event message */
typedef void (*orte_notifier_base_module_log_event_fn_t)(const char *msg);
/*
* Ver 1.0
*/
struct orte_notifier_base_module_1_0_0_t {
orte_notifier_base_module_init_fn_t init;
orte_notifier_base_module_finalize_fn_t finalize;
orte_notifier_base_module_log_fn_t log;
orte_notifier_base_module_log_show_help_fn_t help;
orte_notifier_base_module_log_peer_fn_t peer;
orte_notifier_base_module_log_event_fn_t log_event;
};
typedef struct orte_notifier_base_module_1_0_0_t orte_notifier_base_module_1_0_0_t;
typedef orte_notifier_base_module_1_0_0_t orte_notifier_base_module_t;
/*
* API functions
*/
/* Log a failure message */
typedef void (*orte_notifier_base_API_log_fn_t)(orte_notifier_base_severity_t severity, int errcode, const char *msg, ...);
/* Log a failure that is based upon a show_help message */
typedef void (*orte_notifier_base_API_log_show_help_fn_t)(orte_notifier_base_severity_t severity, int errcode, const char *file, const char *topic, ...);
/* Log a failure related to a peer */
typedef void (*orte_notifier_base_API_log_peer_fn_t)(orte_notifier_base_severity_t severity, int errcode, orte_process_name_t *peer_proc, const char *msg, ...);
/*
* Define a struct to hold the API functions that users will call
*/
struct orte_notifier_API_module_1_0_0_t {
orte_notifier_base_API_log_fn_t log;
orte_notifier_base_API_log_show_help_fn_t show_help;
orte_notifier_base_API_log_peer_fn_t log_peer;
};
typedef struct orte_notifier_API_module_1_0_0_t orte_notifier_API_module_1_0_0_t;
typedef orte_notifier_API_module_1_0_0_t orte_notifier_API_module_t;
ORTE_DECLSPEC extern orte_notifier_API_module_t orte_notifier;
/*
* the standard component data structure
*/
struct orte_notifier_base_component_1_0_0_t {
mca_base_component_t base_version;
mca_base_component_data_t base_data;
};
typedef struct orte_notifier_base_component_1_0_0_t orte_notifier_base_component_1_0_0_t;
typedef orte_notifier_base_component_1_0_0_t orte_notifier_base_component_t;
/*
* Macro for use in components that are of type notifier v1.0.0
*/
#define ORTE_NOTIFIER_BASE_VERSION_1_0_0 \
/* notifier v1.0 is chained to MCA v2.0 */ \
MCA_BASE_VERSION_2_0_0, \
/* notifier v1.0 */ \
"notifier", 1, 0, 0
/*
* To manage unusual events notifications
* Set to noop if not wanted
*/
#if ORTE_WANT_NOTIFIER_LOG_EVENT
#include "notifier_event_calls.h"
#else /* ORTE_WANT_NOTIFIER_LOG_EVENT */
#define ORTE_NOTIFIER_DEFINE_EVENT(i, m)
#define ORTE_NOTIFIER_LOG_EVENT(i, c, t) do {} while (0)
#endif /* ORTE_WANT_NOTIFIER_LOG_EVENT */
END_C_DECLS
#endif /* MCA_NOTIFIER_H */

170
orte/mca/notifier/notifier_event_calls.h
Просмотреть файл

@ -1,170 +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 (c) 2009 Bull SAS. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#ifndef ORTE_NOTIFIER_EVENTS_CALLS_H
#define ORTE_NOTIFIER_EVENTS_CALLS_H
#include "orte_config.h"
#ifdef HAVE_STDIO_H
#include <stdio.h>
#endif /* HAVE_STDIO_H */
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif /* HAVE_SYS_TIME_H */
#include "opal/class/opal_object.h"
BEGIN_C_DECLS
#define ORTE_NOTIFIER_LOG_0 0 /* Initial log format needed (no delay) */
#define ORTE_NOTIFIER_LOG_1 1 /* Intermediate log format needed (delay) */
#define ORTE_NOTIFIER_LOG_2 2 /* Final log format needed (at finalize) */
ORTE_DECLSPEC bool notifier_log_event_enabled(void);
ORTE_DECLSPEC void notifier_event_store(orte_notifier_event_t *);
ORTE_DECLSPEC void notifier_trace_event(int, int, int32_t, time_t, time_t,
const char *);
/*
* Do not use this function directly: use ORTE_NOTIFIER_DEFINE_EVENT() instead
*/
static inline orte_notifier_event_t *notifier_alloc_event(int ev_id,
const char *msg)
{
orte_notifier_event_t *ev;
ev = OBJ_NEW(orte_notifier_event_t);
if (NULL == ev) {
return NULL;
}
asprintf(&ev->ev_msg, msg);
if (NULL == ev->ev_msg) {
OBJ_RELEASE(ev);
return NULL;
}
ev->ev_id = ev_id;
/*
* Store the allocated event into a list to be able to manage the
* unconditional event tracing and freeing during finalize.
*/
notifier_event_store(ev);
return ev;
}
static inline void notifier_count_and_log_event(orte_notifier_event_t *ev,
int ev_id,
int cnt_thresh,
int time_thresh)
{
time_t now, delay;
int32_t count;
opal_atomic_add_32(&ev->ev_cnt, 1);
if (ev->ev_cnt <= cnt_thresh) {
return;
}
count = ev->ev_cnt;
now = time(NULL);
if (ev->ev_already_traced) {
if (now > ev->ev_time_trc + time_thresh) {
delay = now - ev->ev_time_trc;
ev->ev_cnt = 0;
ev->ev_time_trc = now;
notifier_trace_event(ORTE_NOTIFIER_LOG_1, ev_id, count, now, delay,
ev->ev_msg);
}
} else {
ev->ev_already_traced = 1;
ev->ev_cnt = 0;
ev->ev_time_trc = now;
/* We don't care about the delay for the very 1st trace */
notifier_trace_event(ORTE_NOTIFIER_LOG_0, ev_id, count, now, now,
ev->ev_msg);
}
}
#define notifier_event_fn_prefix(i) notifier_log_event_ ## i
/*
* This macro should be called each time a new event will be traced.
* It expands to a static inline function suffixed by the event number.
*/
#define ORTE_NOTIFIER_DEFINE_EVENT(i, m) \
static inline void notifier_event_fn_prefix(i) (int c_thr, int t_thr) \
{ \
static orte_notifier_event_t *prefix_ ## i = NULL; \
if (!notifier_log_event_enabled()) { \
return; \
} \
if (NULL == prefix_ ## i) { \
prefix_ ## i = notifier_alloc_event(i, m); \
if (NULL == prefix_ ## i) { \
return; \
} \
} \
notifier_count_and_log_event(prefix_ ## i, i, c_thr, t_thr); \
}
/*
* This is the log interface that should be called whenever an unsual event
* should be warned about.
* The event should have been defined before, using
* ORTE_NOTIFIER_DEFINE_EVENT():
*
* (1) Event definition:
*
* Typically in a header file call the following:
* ORTE_NOTIFIER_DEFINE_EVENT(0, "message 0")
* This macro expands to
* static inline void notifier_log_event_0(int c_thr, int t_thr)
* {
* static orte_notifier_event_t *prefix_0 = NULL;
* if (!notifier_log_event_enabled()) {
* return;
* }
* if (NULL == prefix_0) {
* prefix_0 = notifier_alloc_event(0, "message 0");
* if (NULL == prefix_0) {
* return;
* }
* }
* notifier_count_and_log_event(prefix_0, 0, c_thr, t_thr);
* }
*
* (2) Event accounting and tracing:
*
* Whenever you want to trace the unusual event whose id is 0, just call:
* ORTE_NOTIFIER_LOG_EVENT(0, 100, 1);
* 100 and 1 are respectively the counter and time thresholds.
* This actually expands to
* notifier_log_event_0(100, 1);
*/
#define ORTE_NOTIFIER_LOG_EVENT(i, c, t) notifier_event_fn_prefix(i) (c, t)
END_C_DECLS
#endif /* ORTE_NOTIFIER_EVENT_CALLS_H */

49
orte/mca/notifier/notifier_event_types.h
Просмотреть файл

@ -1,49 +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 (c) 2009 Bull SAS. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#ifndef ORTE_NOTIFIER_BASE_EVENTS_H
#define ORTE_NOTIFIER_BASE_EVENTS_H
#include "orte_config.h"
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif /* HAVE_SYS_TIME_H */
#include "opal/sys/atomic.h"
#include "opal/class/opal_list.h"
#include "opal/class/opal_object.h"
BEGIN_C_DECLS
typedef struct {
opal_list_item_t super;
volatile int32_t ev_cnt;
int ev_id;
int ev_already_traced;
time_t ev_time_trc;
char *ev_msg;
} orte_notifier_event_t;
ORTE_DECLSPEC OBJ_CLASS_DECLARATION(orte_notifier_event_t);
END_C_DECLS
#endif /* ORTE_NOTIFIER_BASE_EVENTS_H */

51
orte/mca/notifier/smtp/Makefile.am
Просмотреть файл

@ -1,51 +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 (c) 2009-2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
AM_CPPFLAGS = $(notifier_smtp_CPPFLAGS)
dist_pkgdata_DATA = \
help-orte-notifier-smtp.txt
sources = \
notifier_smtp.h \
notifier_smtp_module.c \
notifier_smtp_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 MCA_BUILD_orte_notifier_smtp_DSO
component_noinst =
component_install = mca_notifier_smtp.la
else
component_noinst = libmca_notifier_smtp.la
component_install =
endif
mcacomponentdir = $(pkglibdir)
mcacomponent_LTLIBRARIES = $(component_install)
mca_notifier_smtp_la_SOURCES = $(sources)
mca_notifier_smtp_la_LDFLAGS = -module -avoid-version $(notifier_smtp_LDFLAGS)
mca_notifier_smtp_la_LIBADD = $(notifier_smtp_LIBS)
noinst_LTLIBRARIES = $(component_noinst)
libmca_notifier_smtp_la_SOURCES =$(sources)
libmca_notifier_smtp_la_LDFLAGS = -module -avoid-version $(notifier_smtp_LDFLAGS)
libmca_notifier_smtp_la_LIBADD = $(notifier_smtp_LIBS)

41
orte/mca/notifier/smtp/configure.m4
Просмотреть файл

@ -1,41 +0,0 @@
# -*- shell-script -*-
#
# 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 (c) 2009-2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# MCA_notifier_smtp_CONFIG([action-if-found], [action-if-not-found])
# -----------------------------------------------------------
AC_DEFUN([MCA_orte_notifier_smtp_CONFIG], [
AC_CONFIG_FILES([orte/mca/notifier/smtp/Makefile])
OPAL_SETUP_COMPONENT_PACKAGE([notifier],
[smtp],
[esmtp],
[include/libesmtp.h],
[libesmtp*],
[libesmtp.h],
[esmtp],
[smtp_create_session],
[],
[orte_notifier_want_smtp=1],
[orte_notifier_want_smtp=0])
AS_IF([test "$orte_notifier_want_smtp" = 1 -a "$orte_without_full_support" = 0],
[$1],
[$2])
])dnl

33
orte/mca/notifier/smtp/help-orte-notifier-smtp.txt
Просмотреть файл

@ -1,33 +0,0 @@
# -*- text -*-
#
# Copyright (c) 2009 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# This is the US/English help file for Open MPI's SMTP notifier support
#
[to/from not specified]
Error: the Open MPI SMTP notifier component had no "to" and/or "from"
email addresses specified.
#
[server not specified]
Error: the Open MPI SMTP notifier component had no SMTP server name or
IP address specified.
#
[unable to resolve server]
Sorry, Open MPI's SMTP notifier component was unable to resolve the IP
address of the server provided.
Server: %s
#
[send_email failed]
Oops! Open MPI's SMTP notifier failed to send an email.
Reason: %s
libESMTP function: %s
libESMTP message: %s
Message: %s
#

64
orte/mca/notifier/smtp/notifier_smtp.h
Просмотреть файл

@ -1,64 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
*/
#ifndef NOTIFIER_SMTP_H
#define NOTIFIER_SMTP_H
#include "orte_config.h"
#include <netdb.h>
#include "libesmtp.h"
#include "orte/mca/notifier/notifier.h"
BEGIN_C_DECLS
typedef struct {
orte_notifier_base_component_t super;
/* SMTP server name and port */
char *server;
int port;
/* To, From, Subject */
char *to, **to_argv, *from_name, *from_addr, *subject;
/* Mail body prefix and suffix */
char *body_prefix, *body_suffix;
/* struct hostent from resolved SMTP server name */
struct hostent *server_hostent;
/* Priority of this component */
int priority;
} orte_notifier_smtp_component_t;
/*
* Notifier interfaces
*/
ORTE_MODULE_DECLSPEC extern orte_notifier_smtp_component_t
mca_notifier_smtp_component;
extern orte_notifier_base_module_t orte_notifier_smtp_module;
END_C_DECLS
#endif

229
orte/mca/notifier/smtp/notifier_smtp_component.c
Просмотреть файл

@ -1,229 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/*
* Simple smtp notifier (using libesmtp)
*/
#include "orte_config.h"
#include "opal/mca/base/mca_base_param.h"
#include "orte/constants.h"
#include "orte/util/show_help.h"
#include "notifier_smtp.h"
static int smtp_component_query(mca_base_module_t **module, int *priority);
static int smtp_close(void);
static int smtp_register(void);
/*
* Struct of function pointers that need to be initialized
*/
orte_notifier_smtp_component_t mca_notifier_smtp_component = {
{
{
ORTE_NOTIFIER_BASE_VERSION_1_0_0,
"smtp",
ORTE_MAJOR_VERSION,
ORTE_MINOR_VERSION,
ORTE_RELEASE_VERSION,
NULL,
smtp_close,
smtp_component_query,
smtp_register,
},
{
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
}
},
/* SMTP server and port */
"localhost",
25,
/* To, from, subject */
NULL,
NULL,
"Open MPI Notifier",
NULL,
"Open MPI notifier",
/* Mail body prefix, suffix */
"The Open MPI SMTP notifier wishes to inform you of the following message:\n\n",
"\n\nSincerely,\nOscar the Open MPI Owl",
/* Struct hostent */
NULL,
/* Priority */
10,
};
static int smtp_register(void)
{
char version[256];
/* Server stuff */
mca_base_param_reg_string(&mca_notifier_smtp_component.super.base_version,
"server",
"SMTP server name or IP address",
false, false,
mca_notifier_smtp_component.server,
&mca_notifier_smtp_component.server);
mca_base_param_reg_int(&mca_notifier_smtp_component.super.base_version,
"port",
"SMTP server port",
false, false,
mca_notifier_smtp_component.port,
&mca_notifier_smtp_component.port);
/* Email stuff */
mca_base_param_reg_string(&mca_notifier_smtp_component.super.base_version,
"to",
"Comma-delimited list of email addresses to send to",
false, false,
mca_notifier_smtp_component.to,
&mca_notifier_smtp_component.to);
mca_base_param_reg_string(&mca_notifier_smtp_component.super.base_version,
"from_addr",
"Email address that messages will be from",
false, false,
mca_notifier_smtp_component.from_addr,
&mca_notifier_smtp_component.from_addr);
mca_base_param_reg_string(&mca_notifier_smtp_component.super.base_version,
"from_name",
"Email name that messages will be from",
false, false,
mca_notifier_smtp_component.from_name,
&mca_notifier_smtp_component.from_name);
mca_base_param_reg_string(&mca_notifier_smtp_component.super.base_version,
"subject",
"Email subject",
false, false,
mca_notifier_smtp_component.subject,
&mca_notifier_smtp_component.subject);
/* Mail body prefix and suffix */
mca_base_param_reg_string(&mca_notifier_smtp_component.super.base_version,
"body_prefix",
"Text to put at the beginning of the mail message",
false, false,
mca_notifier_smtp_component.body_prefix,
&mca_notifier_smtp_component.body_prefix);
mca_base_param_reg_string(&mca_notifier_smtp_component.super.base_version,
"body_suffix",
"Text to put at the beginning of the mail message",
false, false,
mca_notifier_smtp_component.body_suffix,
&mca_notifier_smtp_component.body_suffix);
/* Priority */
mca_base_param_reg_int(&mca_notifier_smtp_component.super.base_version,
"priority",
"Priority of this component",
false, false,
mca_notifier_smtp_component.priority,
&mca_notifier_smtp_component.priority);
/* Libesmtp version */
smtp_version(version, sizeof(version), 0);
version[sizeof(version) - 1] = '\0';
mca_base_param_reg_string(&mca_notifier_smtp_component.super.base_version,
"libesmtp_version",
"Version of libesmtp that this component is linked against",
false, true, version, NULL);
return ORTE_SUCCESS;
}
static int smtp_close(void)
{
if (NULL != mca_notifier_smtp_component.server) {
free(mca_notifier_smtp_component.server);
}
if (NULL != mca_notifier_smtp_component.to) {
free(mca_notifier_smtp_component.to);
}
if (NULL != mca_notifier_smtp_component.from_name) {
free(mca_notifier_smtp_component.from_name);
}
if (NULL != mca_notifier_smtp_component.from_addr) {
free(mca_notifier_smtp_component.from_addr);
}
if (NULL != mca_notifier_smtp_component.subject) {
free(mca_notifier_smtp_component.subject);
}
if (NULL != mca_notifier_smtp_component.body_prefix) {
free(mca_notifier_smtp_component.body_prefix);
}
if (NULL != mca_notifier_smtp_component.body_suffix) {
free(mca_notifier_smtp_component.body_suffix);
}
return ORTE_SUCCESS;
}
static int smtp_component_query(mca_base_module_t **module,
int *priority)
{
*priority = 0;
*module = NULL;
/* If there's no to or from, there's no love */
if (NULL == mca_notifier_smtp_component.to ||
'\0' == mca_notifier_smtp_component.to[0] ||
NULL == mca_notifier_smtp_component.from_addr ||
'\0' == mca_notifier_smtp_component.from_addr[0]) {
orte_show_help("help-orte-notifier-smtp.txt",
"to/from not specified", true);
return ORTE_ERR_NOT_FOUND;
}
/* Sanity checks */
if (NULL == mca_notifier_smtp_component.server ||
'\0' == mca_notifier_smtp_component.server[0]) {
orte_show_help("help-orte-notifier-smtp.txt",
"server not specified", true);
return ORTE_ERR_NOT_FOUND;
}
/* Since we have to open a socket later, try to resolve the IP
address of the server now. Save the result, or abort if we
can't resolve it. */
mca_notifier_smtp_component.server_hostent =
gethostbyname(mca_notifier_smtp_component.server);
if (NULL == mca_notifier_smtp_component.server_hostent) {
orte_show_help("help-orte-notifier-smtp.txt",
"unable to resolve server",
true, mca_notifier_smtp_component.server);
return ORTE_ERR_NOT_FOUND;
}
*priority = 10;
*module = (mca_base_module_t *)&orte_notifier_smtp_module;
return ORTE_SUCCESS;
}

354
orte/mca/notifier/smtp/notifier_smtp_module.c
Просмотреть файл

@ -1,354 +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 (c) 2007 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/*
* Send an email upon notifier events.
*/
#include "orte_config.h"
#include <stdio.h>
#include <string.h>
#ifdef HAVE_STDARG_H
#include <stdarg.h>
#endif
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#ifdef HAVE_SIGNAL_H
#include <signal.h>
#endif
#include "opal/util/show_help.h"
#include "opal/util/argv.h"
#include "orte/constants.h"
#include "orte/mca/ess/ess.h"
#include "orte/util/error_strings.h"
#include "orte/util/name_fns.h"
#include "orte/util/show_help.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/notifier/base/base.h"
#include "notifier_smtp.h"
/* Static API's */
static void mylog(orte_notifier_base_severity_t severity, int errcode,
const char *msg, va_list ap);
static void myhelplog(orte_notifier_base_severity_t severity, int errcode,
const char *filename,
const char *topic, va_list ap);
static void mypeerlog(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc,
const char *msg, va_list ap);
/* Module */
orte_notifier_base_module_t orte_notifier_smtp_module = {
NULL,
NULL,
mylog,
myhelplog,
mypeerlog,
NULL
};
typedef enum {
SENT_NONE,
SENT_HEADER,
SENT_BODY_PREFIX,
SENT_BODY,
SENT_BODY_SUFFIX,
SENT_ALL
} sent_flag_t;
typedef struct {
sent_flag_t sent_flag;
char *msg;
char *prev_string;
} message_status_t;
/*
* Convert lone \n's to \r\n
*/
static char *crnl(char *orig)
{
int i, j, max, count;
char *str;
return strdup(orig);
/* Count how much space we need */
count = max = strlen(orig);
for (i = 0; i < max; ++i) {
if (orig[i] == '\n' && i > 0 && orig[i - 1] != '\r') {
++count;
}
}
/* Copy, changing \n to \r\n */
str = malloc(count + 1);
for (j = i = 0; i < max; ++i) {
if (orig[i] == '\n' && i > 0 && orig[i - 1] != '\r') {
str[j++] = '\n';
}
str[j++] = orig[i];
}
str[j] = '\0';
return str;
}
/*
* Callback function invoked via smtp_start_session()
*/
static const char *message_cb(void **buf, int *len, void *arg)
{
message_status_t *ms = (message_status_t*) arg;
if (NULL == *buf) {
*buf = malloc(8192);
}
if (NULL == len) {
ms->sent_flag = SENT_NONE;
return NULL;
}
/* Free the previous string */
if (NULL != ms->prev_string) {
free(ms->prev_string);
ms->prev_string = NULL;
}
switch (ms->sent_flag) {
case SENT_NONE:
/* Send a blank line to signify the end of the header */
ms->sent_flag = SENT_HEADER;
ms->prev_string = NULL;
*len = 2;
return "\r\n";
case SENT_HEADER:
if (NULL != mca_notifier_smtp_component.body_prefix) {
ms->sent_flag = SENT_BODY_PREFIX;
ms->prev_string = crnl(mca_notifier_smtp_component.body_prefix);
*len = strlen(ms->prev_string);
return ms->prev_string;
}
case SENT_BODY_PREFIX:
ms->sent_flag = SENT_BODY;
ms->prev_string = crnl(ms->msg);
*len = strlen(ms->prev_string);
return ms->prev_string;
case SENT_BODY:
if (NULL != mca_notifier_smtp_component.body_suffix) {
ms->sent_flag = SENT_BODY_SUFFIX;
ms->prev_string = crnl(mca_notifier_smtp_component.body_suffix);
*len = strlen(ms->prev_string);
return ms->prev_string;
}
case SENT_BODY_SUFFIX:
case SENT_ALL:
default:
ms->sent_flag = SENT_ALL;
*len = 0;
return NULL;
}
}
/*
* Back-end function to actually send the email
*/
static int send_email(char *msg)
{
int i, err = ORTE_SUCCESS;
char *str = NULL;
char *errmsg = NULL;
struct sigaction sig, oldsig;
bool set_oldsig = false;
smtp_session_t session = NULL;
smtp_message_t message = NULL;
message_status_t ms;
orte_notifier_smtp_component_t *c = &mca_notifier_smtp_component;
if (NULL == c->to_argv) {
c->to_argv = opal_argv_split(c->to, ',');
if (NULL == c->to_argv ||
NULL == c->to_argv[0]) {
return ORTE_ERR_OUT_OF_RESOURCE;
}
}
ms.sent_flag = SENT_NONE;
ms.prev_string = NULL;
ms.msg = msg;
/* Temporarily disable SIGPIPE so that if remote servers timeout
or hang up on us, it doesn't kill this application. We'll
restore the original SIGPIPE handler when we're done. */
sig.sa_handler = SIG_IGN;
sigemptyset(&sig.sa_mask);
sig.sa_flags = 0;
sigaction(SIGPIPE, &sig, &oldsig);
set_oldsig = true;
/* Try to get a libesmtp session. If so, assume that libesmtp is
happy and proceeed */
session = smtp_create_session();
if (NULL == session) {
err = ORTE_ERR_NOT_SUPPORTED;
errmsg = "stmp_create_session";
goto error;
}
/* Create the message */
message = smtp_add_message(session);
if (NULL == message) {
err = ORTE_ERROR;
errmsg = "stmp_add_message";
goto error;
}
/* Set the SMTP server (yes, it's a weird return status!) */
asprintf(&str, "%s:%d", c->server, c->port);
if (0 == smtp_set_server(session, str)) {
err = ORTE_ERROR;
errmsg = "stmp_set_server";
goto error;
}
free(str);
str = NULL;
/* Add the sender */
if (0 == smtp_set_reverse_path(message, c->from_addr)) {
err = ORTE_ERROR;
errmsg = "stmp_set_reverse_path";
goto error;
}
/* Set the subject and some headers */
asprintf(&str, "Open MPI SMTP Notifier v%d.%d.%d",
c->super.base_version.mca_component_major_version,
c->super.base_version.mca_component_minor_version,
c->super.base_version.mca_component_release_version);
if (0 == smtp_set_header(message, "Subject", c->subject) ||
0 == smtp_set_header_option(message, "Subject", Hdr_OVERRIDE, 1) ||
0 == smtp_set_header(message, "To", NULL, NULL) ||
0 == smtp_set_header(message, "From",
(NULL != c->from_name ?
c->from_name : c->from_addr),
c->from_addr) ||
0 == smtp_set_header(message, "X-Mailer", str) ||
0 == smtp_set_header_option(message, "Subject", Hdr_OVERRIDE, 1)) {
err = ORTE_ERROR;
errmsg = "smtp_set_header";
goto error;
}
free(str);
str = NULL;
/* Add the recipients */
for (i = 0; NULL != c->to_argv[i]; ++i) {
if (NULL == smtp_add_recipient(message, c->to_argv[i])) {
err = ORTE_ERR_OUT_OF_RESOURCE;
errmsg = "stmp_add_recipient";
goto error;
}
}
/* Set the callback to get the message */
if (0 == smtp_set_messagecb(message, message_cb, &ms)) {
err = ORTE_ERROR;
errmsg = "smtp_set_messagecb";
goto error;
}
/* Send it! */
if (0 == smtp_start_session(session)) {
err = ORTE_ERROR;
errmsg = "smtp_start_session";
goto error;
}
/* Fall through */
error:
if (NULL != str) {
free(str);
}
if (NULL != session) {
smtp_destroy_session(session);
}
/* Restore the SIGPIPE handler */
if (set_oldsig) {
sigaction(SIGPIPE, &oldsig, NULL);
}
if (ORTE_SUCCESS != err) {
int e;
char em[256];
e = smtp_errno();
smtp_strerror(e, em, sizeof(em));
orte_show_help("help-orte-notifier-smtp.txt",
"send_email failed",
true, "libesmtp library call failed",
errmsg, em, e, msg);
}
return err;
}
static void mylog(orte_notifier_base_severity_t severity, int errcode,
const char *msg, va_list ap)
{
char *output;
/* If there was a message, output it */
vasprintf(&output, msg, ap);
if (NULL != output) {
send_email(output);
free(output);
}
}
static void myhelplog(orte_notifier_base_severity_t severity, int errcode,
const char *filename,
const char *topic, va_list ap)
{
char *output = opal_show_help_vstring(filename, topic, false, ap);
if (NULL != output) {
send_email(output);
free(output);
}
}
static void mypeerlog(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc, const char *msg,
va_list ap)
{
char *buf = orte_notifier_base_peer_log(errcode, peer_proc, msg, ap);
if (NULL != buf) {
send_email(buf);
free(buf);
}
}

44
orte/mca/notifier/syslog/Makefile.am
Просмотреть файл

@ -1,44 +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 (c) 2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
sources = \
notifier_syslog.h \
notifier_syslog_module.c \
notifier_syslog_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 MCA_BUILD_orte_notifier_syslog_DSO
component_noinst =
component_install = mca_notifier_syslog.la
else
component_noinst = libmca_notifier_syslog.la
component_install =
endif
mcacomponentdir = $(pkglibdir)
mcacomponent_LTLIBRARIES = $(component_install)
mca_notifier_syslog_la_SOURCES = $(sources)
mca_notifier_syslog_la_LDFLAGS = -module -avoid-version
noinst_LTLIBRARIES = $(component_noinst)
libmca_notifier_syslog_la_SOURCES =$(sources)
libmca_notifier_syslog_la_LDFLAGS = -module -avoid-version

19
orte/mca/notifier/syslog/configure.m4
Просмотреть файл

@ -1,19 +0,0 @@
# -*- shell-script -*-
#
# Copyright (c) 2011 Los Alamos National Security, LLC.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# MCA_notifier_syslog_CONFIG([action-if-found], [action-if-not-found])
# -----------------------------------------------------------
AC_DEFUN([MCA_orte_notifier_syslog_CONFIG], [
AC_CONFIG_FILES([orte/mca/notifier/syslog/Makefile])
AS_IF([test "$orte_without_full_support" = 0],
[$1],
[$2])
])

39
orte/mca/notifier/syslog/notifier_syslog.h
Просмотреть файл

@ -1,39 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
*/
#ifndef NOTIFIER_SYSLOG_H
#define NOTIFIER_SYSLOG_H
#include "orte_config.h"
#include "orte/mca/notifier/notifier.h"
BEGIN_C_DECLS
/*
* Notifier interfaces
*/
ORTE_MODULE_DECLSPEC extern orte_notifier_base_component_t mca_notifier_syslog_component;
extern orte_notifier_base_module_t orte_notifier_syslog_module;
END_C_DECLS
#endif

63
orte/mca/notifier/syslog/notifier_syslog_component.c
Просмотреть файл

@ -1,63 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2008 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 (c) 2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/** @file:
*
*/
/*
* includes
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "notifier_syslog.h"
static int orte_notifier_syslog_component_query(mca_base_module_t **module,
int *priority);
/*
* Struct of function pointers that need to be initialized
*/
orte_notifier_base_component_t mca_notifier_syslog_component = {
{
ORTE_NOTIFIER_BASE_VERSION_1_0_0,
"syslog", /* MCA module name */
ORTE_MAJOR_VERSION, /* MCA module major version */
ORTE_MINOR_VERSION, /* MCA module minor version */
ORTE_RELEASE_VERSION, /* MCA module release version */
NULL,
NULL,
orte_notifier_syslog_component_query /* module query */
},
{
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
}
};
static int orte_notifier_syslog_component_query(mca_base_module_t **module,
int *priority)
{
*priority = 1;
*module = (mca_base_module_t *)&orte_notifier_syslog_module;
return ORTE_SUCCESS;
}

127
orte/mca/notifier/syslog/notifier_syslog_module.c
Просмотреть файл

@ -1,127 +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 (c) 2007 Sun Microsystems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include <string.h>
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif /* HAVE_SYS_TIME_H */
#ifdef HAVE_SYSLOG_H
#include <syslog.h>
#endif
#ifdef HAVE_STDARG_H
#include <stdarg.h>
#endif
#include "opal/util/show_help.h"
#include "orte/util/error_strings.h"
#include "orte/util/name_fns.h"
#include "orte/mca/notifier/base/base.h"
#include "notifier_syslog.h"
/* Static API's */
static int init(void);
static void finalize(void);
static void mylog(orte_notifier_base_severity_t severity, int errcode,
const char *msg, va_list ap);
static void myhelplog(orte_notifier_base_severity_t severity, int errcode,
const char *filename, const char *topic, va_list ap);
static void mypeerlog(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc, const char *msg,
va_list ap);
static void myeventlog(const char *msg);
/* Module def */
orte_notifier_base_module_t orte_notifier_syslog_module = {
init,
finalize,
mylog,
myhelplog,
mypeerlog,
myeventlog
};
static int init(void)
{
int opts;
opts = LOG_CONS | LOG_PID;
openlog("Open MPI Error Report:", opts, LOG_USER);
return ORTE_SUCCESS;
}
static void finalize(void)
{
closelog();
}
static void mylog(orte_notifier_base_severity_t severity, int errcode,
const char *msg, va_list ap)
{
/* If there was a message, output it */
#if defined(HAVE_VSYSLOG)
vsyslog(severity, msg, ap);
#else
char *output;
vasprintf(&output, msg, ap);
syslog(severity, output, NULL);
free(output);
#endif
}
static void myhelplog(orte_notifier_base_severity_t severity, int errcode,
const char *filename, const char *topic, va_list ap)
{
char *output = opal_show_help_vstring(filename, topic, false, ap);
/* if nothing came back, then nothing to do */
if (NULL == output) {
return;
}
/* go ahead and output it */
syslog(severity, output, NULL);
free(output);
}
static void mypeerlog(orte_notifier_base_severity_t severity, int errcode,
orte_process_name_t *peer_proc, const char *msg,
va_list ap)
{
char *buf = orte_notifier_base_peer_log(errcode, peer_proc, msg, ap);
if (NULL != buf) {
syslog(severity, buf, NULL);
free(buf);
}
}
static void myeventlog(const char *msg)
{
/* If there was a message, output it */
syslog(LOG_LOCAL0 | LOG_NOTICE, msg, NULL);
}

5
orte/mca/oob/tcp/oob_tcp_peer.c
Просмотреть файл

@ -64,7 +64,6 @@
#include "orte/runtime/orte_globals.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/ess/ess.h"
#include "orte/mca/notifier/notifier.h"
#include "orte/runtime/orte_wait.h"
#include "oob_tcp.h"
@ -641,10 +640,6 @@ void mca_oob_tcp_peer_shutdown(mca_oob_tcp_peer_t* peer)
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&(peer->peer_name)),
(NULL == host) ? "NULL" : host);
/* provide a notifier message */
orte_notifier.log_peer(ORTE_NOTIFIER_CRIT, ORTE_ERR_COMM_FAILURE, &(peer->peer_name),
"OOB connection retries exceeded. Cannot communicate with peer %s.",
ORTE_JOBID_PRINT(peer->peer_name.jobid));
/* There are cases during the initial connection setup where
the peer_send_msg is NULL but there are things in the queue

1
orte/mca/oob/ud/oob_ud_peer.c
Просмотреть файл

@ -18,7 +18,6 @@
#include "orte/util/name_fns.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/ess/ess.h"
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/state/state.h"
#include "orte/mca/routed/routed.h"

1
orte/mca/plm/base/plm_base_launch_support.c
Просмотреть файл

@ -57,7 +57,6 @@
#include "orte/mca/filem/filem.h"
#include "orte/mca/filem/base/base.h"
#include "orte/mca/grpcomm/base/base.h"
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/rml/base/rml_contact.h"
#include "orte/mca/sensor/sensor.h"
#include "orte/runtime/orte_globals.h"

3
orte/mca/rml/rml_types.h
Просмотреть файл

@ -121,9 +121,6 @@ BEGIN_C_DECLS
/* tag for receiving heartbeats */
#define ORTE_RML_TAG_HEARTBEAT 39
/* notifier data */
#define ORTE_RML_TAG_NOTIFIER_HNP 40
/* Process Migration Tool Tag */
#define ORTE_RML_TAG_MIGRATE 43

30
orte/mca/snapc/base/snapc_base_fns.c
Просмотреть файл

@ -55,7 +55,6 @@
#include "orte/mca/errmgr/errmgr.h"
#include "orte/runtime/orte_globals.h"
#include "orte/util/name_fns.h"
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/sstore/sstore.h"
#include "orte/mca/sstore/base/base.h"
@ -418,34 +417,29 @@ static void snapc_none_global_cmdline_request(int status,
* Utility functions
********************/
/* Report the checkpoint status over the notifier interface */
/* Report the checkpoint status */
void orte_snapc_ckpt_state_notify(int state)
{
switch(state) {
case ORTE_SNAPC_CKPT_STATE_ESTABLISHED:
orte_notifier.log(ORTE_NOTIFIER_INFO, ORTE_SNAPC_CKPT_NOTIFY(state),
"%d: Checkpoint established for process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
opal_output(0, "%d: Checkpoint established for process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
break;
case ORTE_SNAPC_CKPT_STATE_NO_CKPT:
orte_notifier.log(ORTE_NOTIFIER_WARN, ORTE_SNAPC_CKPT_NOTIFY(state),
"%d: Process %s is not checkpointable.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
opal_output(0, "%d: Process %s is not checkpointable.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
break;
case ORTE_SNAPC_CKPT_STATE_ERROR:
orte_notifier.log(ORTE_NOTIFIER_WARN, ORTE_SNAPC_CKPT_NOTIFY(state),
"%d: Failed to checkpoint process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
opal_output(0, "%d: Failed to checkpoint process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
break;
case ORTE_SNAPC_CKPT_STATE_RECOVERED:
orte_notifier.log(ORTE_NOTIFIER_INFO, ORTE_SNAPC_CKPT_NOTIFY(state),
"%d: Successfully restarted process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
opal_output(0, "%d: Successfully restarted process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
break;
case ORTE_SNAPC_CKPT_STATE_NO_RESTART:
orte_notifier.log(ORTE_NOTIFIER_WARN, ORTE_SNAPC_CKPT_NOTIFY(state),
"%d: Failed to restart process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
opal_output(0, "%d: Failed to restart process %s.",
orte_process_info.pid, ORTE_JOBID_PRINT(ORTE_PROC_MY_NAME->jobid));
break;
/* ADK: We currently do not notify for these states, but good to
* have them around anyways. */
@ -677,7 +671,7 @@ int orte_snapc_base_global_coord_ckpt_update_cmd(orte_process_name_t* peer,
}
/*
* Pass on the checkpoint state over the notifier interface.
* Pass on the checkpoint state.
*/
orte_snapc_ckpt_state_notify(ckpt_status);

14
orte/mca/state/hnp/state_hnp.c
Просмотреть файл

@ -26,7 +26,6 @@
#include "orte/mca/ras/base/base.h"
#include "orte/mca/rmaps/base/base.h"
#include "orte/mca/routed/routed.h"
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/sensor/sensor.h"
#include "orte/util/session_dir.h"
#include "orte/runtime/orte_quit.h"
@ -356,7 +355,6 @@ static void check_all_complete(int fd, short args, void *cbdata)
orte_std_cntr_t index;
bool one_still_alive;
orte_vpid_t lowest=0;
char *msg;
OPAL_OUTPUT_VERBOSE((2, orte_state_base_output,
"%s state:hnp:check_job_complete on job %s",
@ -559,18 +557,6 @@ static void check_all_complete(int fd, short args, void *cbdata)
* wasn't already set by an error condition
*/
ORTE_UPDATE_EXIT_STATUS(0);
/* provide a notifier message if that framework is active - ignored otherwise */
if (NULL != (job = (orte_job_t*)opal_pointer_array_get_item(orte_job_data, 1))) {
if (0 == orte_exit_status) {
asprintf(&msg, "Job %s complete", ORTE_JOBID_PRINT(job->jobid));
orte_notifier.log(ORTE_NOTIFIER_INFO, 0, msg);
} else {
asprintf(&msg, "Job %s terminated abnormally", ORTE_JOBID_PRINT(job->jobid));
orte_notifier.log(ORTE_NOTIFIER_ALERT, orte_exit_status, msg);
}
free(msg);
/* this job object will be release during finalize */
}
/* order daemon termination - this tells us to cleanup
* our local procs as well as telling remote daemons

11
orte/tools/orte-info/components.c
Просмотреть файл

@ -65,8 +65,6 @@
#include "orte/mca/state/state.h"
#include "orte/mca/state/base/base.h"
#if !ORTE_DISABLE_FULL_SUPPORT
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/notifier/base/base.h"
#include "orte/mca/iof/iof.h"
#include "orte/mca/iof/base/base.h"
#include "orte/mca/oob/oob.h"
@ -351,14 +349,6 @@ void orte_info_open_components(void)
opal_pointer_array_add(&component_map, map);
#if !ORTE_DISABLE_FULL_SUPPORT
if (ORTE_SUCCESS != orte_notifier_base_open()) {
goto error;
}
map = OBJ_NEW(orte_info_component_map_t);
map->type = strdup("notifier");
map->components = &orte_notifier_base_components_available;
opal_pointer_array_add(&component_map, map);
if (ORTE_SUCCESS != mca_oob_base_open()) {
goto error;
}
@ -504,7 +494,6 @@ void orte_info_close_components()
#if OPAL_ENABLE_FT_CR == 1
(void) orte_snapc_base_close();
#endif
(void) orte_notifier_base_close();
(void) orte_filem_base_close();
(void) orte_iof_base_close();
(void) orte_plm_base_close();

1
orte/tools/orte-info/orte-info.c
Просмотреть файл

@ -240,7 +240,6 @@ int main(int argc, char *argv[])
opal_pointer_array_add(&mca_types, "ess");
opal_pointer_array_add(&mca_types, "grpcomm");
opal_pointer_array_add(&mca_types, "db");
opal_pointer_array_add(&mca_types, "notifier");
/* Execute the desired action(s) */