Bring back the changes from /tmp/jjhursey-rmaps. Specific merge
command: svn merge -r 7567:7663 https://svn.open-mpi.org/svn/ompi/tmp/jjhursey-rmaps . (where "." is a trunk checkout) The logs from this branch are much more descriptive than I will put here (including a *really* long description from last night). Here's the short version: - fixed some broken implementations in ras and rmaps - "orterun --host ..." now works and has clearly defined semantics (this was the impetus for the branch and all these fixes -- LANL had a requirement for --host to work for 1.0) - there is still a little bit of cleanup left to do post-1.0 (we got correct functionality for 1.0 -- we did not fix bad implementations that still "work") - rds/hostfile and ras/hostfile handshaking - singleton node segment assignments in stage1 - remove the default hostfile (no need for it anymore with the localhost ras component) - clean up pls components to avoid duplicate ras mapping queries - [possible] -bynode/-byslot being specific to a single app context This commit was SVN r7664.
Этот коммит содержится в:
Jeff Squyres
родитель
fb19cc4177
Коммит
0629cdc2d7
@ -1 +1,30 @@
|
||||
localhost
|
||||
#
|
||||
# Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
# All rights reserved.
|
||||
# Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
# All rights reserved.
|
||||
# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
# University of Stuttgart. All rights reserved.
|
||||
# Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
# All rights reserved.
|
||||
# $COPYRIGHT$
|
||||
#
|
||||
# Additional copyrights may follow
|
||||
#
|
||||
# $HEADER$
|
||||
#
|
||||
# This is the default hostfile for Open MPI. Notice that it does not
|
||||
# contain any hosts (not even localhost). This file should only
|
||||
# contain hosts if a system administrator wants users to always have
|
||||
# the same set of default hosts, and is not using a batch scheduler
|
||||
# (such as SLURM, PBS, etc.).
|
||||
#
|
||||
# If you are primarily interested in running Open MPI on one node, you
|
||||
# should *not* simply list "localhost" in here (contrary to prior MPI
|
||||
# implementations, such as LAM/MPI). A localhost-only node list is
|
||||
# created by the RAS component named "localhost" if no other RAS
|
||||
# components were able to find any hosts to run on (this behavior can
|
||||
# be disabled by excluding the localhost RAS component by specifying
|
||||
# the value "^localhost" [without the quotes] to the "ras" MCA
|
||||
# parameter).
|
||||
|
||||
|
||||
@ -97,7 +97,7 @@ static inline int __poe_argv_append_int(int *argc, char ***argv, int varname, in
|
||||
*/
|
||||
int pls_poe_launch_interactive_orted(orte_jobid_t jobid)
|
||||
{
|
||||
opal_list_t nodes;
|
||||
opal_list_t nodes, mapping_list;
|
||||
opal_list_item_t* item;
|
||||
size_t num_nodes;
|
||||
orte_vpid_t vpid;
|
||||
@ -114,10 +114,12 @@ int pls_poe_launch_interactive_orted(orte_jobid_t jobid)
|
||||
int i;
|
||||
int status;
|
||||
FILE *hfp, *cfp;
|
||||
|
||||
/* query the list of nodes allocated to the job - don't need the entire
|
||||
* mapping - as the daemon/proxy is responsibe for determining the apps
|
||||
* to launch on each node.
|
||||
|
||||
/* Query the list of nodes allocated and mapped to this job.
|
||||
* We need the entire mapping for a couple of reasons:
|
||||
* - need the prefix to start with.
|
||||
* - need to know if we are launching on a subset of the allocated nodes
|
||||
* All other mapping responsibilities fall to orted in the fork PLS
|
||||
*/
|
||||
if (mca_pls_poe_component.verbose > 10) opal_output(0, "%s:--- BEGIN ---\n", __FUNCTION__);
|
||||
|
||||
@ -127,7 +129,8 @@ int pls_poe_launch_interactive_orted(orte_jobid_t jobid)
|
||||
if((cfp=fopen(mca_pls_poe_component.cmdfile,"w"))==NULL) return ORTE_ERR_OUT_OF_RESOURCE;
|
||||
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
rc = orte_ras_base_node_query_alloc(&nodes, jobid);
|
||||
OBJ_CONSTRUCT(&mapping_list, opal_list_t);
|
||||
rc = orte_rmaps_base_mapped_node_query(&mapping_list, &nodes, jobid);
|
||||
if(ORTE_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
@ -290,6 +293,12 @@ cleanup:
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
|
||||
while(NULL != (item = opal_list_remove_first(&mapping_list))) {
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&mapping_list);
|
||||
|
||||
if (mca_pls_poe_component.verbose > 10) opal_output(0, "%s: --- END rc(%d) ---\n", __FUNCTION__, rc);
|
||||
return rc;
|
||||
}
|
||||
@ -446,7 +455,7 @@ __poe_launch_interactive - launch an interactive job
|
||||
*/
|
||||
static inline int __poe_launch_interactive(orte_jobid_t jobid)
|
||||
{
|
||||
opal_list_t map, nodes;
|
||||
opal_list_t map, nodes, mapping_list;
|
||||
opal_list_item_t* item;
|
||||
orte_vpid_t vpid_start, vpid_range;
|
||||
size_t num_nodes, num_procs;
|
||||
@ -468,7 +477,8 @@ static inline int __poe_launch_interactive(orte_jobid_t jobid)
|
||||
mca_pls_poe_component.jobid = jobid;
|
||||
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
rc = orte_ras_base_node_query_alloc(&nodes, jobid);
|
||||
OBJ_CONSTRUCT(&mapping_list, opal_list_t);
|
||||
rc = orte_rmaps_base_mapped_node_query(&mapping_list, &nodes, jobid);
|
||||
if(ORTE_SUCCESS != rc) { goto cleanup; }
|
||||
|
||||
num_nodes = opal_list_get_size(&nodes);
|
||||
@ -582,6 +592,11 @@ cleanup:
|
||||
}
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
|
||||
while(NULL != (item = opal_list_remove_first(&mapping_list))) {
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&mapping_list);
|
||||
|
||||
if(mca_pls_poe_component.verbose>10) {
|
||||
opal_output(0, "%s: --- END rc(%d) ---\n", __FUNCTION__, rc);
|
||||
}
|
||||
|
||||
@ -375,7 +375,7 @@ static void orte_pls_rsh_wait_daemon(pid_t pid, int status, void* cbdata)
|
||||
|
||||
int orte_pls_rsh_launch(orte_jobid_t jobid)
|
||||
{
|
||||
opal_list_t nodes;
|
||||
opal_list_t nodes, mapping_list;
|
||||
opal_list_item_t* item;
|
||||
size_t num_nodes;
|
||||
orte_vpid_t vpid;
|
||||
@ -394,13 +394,15 @@ int orte_pls_rsh_launch(orte_jobid_t jobid)
|
||||
bool remote_bash = false, remote_csh = false;
|
||||
bool local_bash = false, local_csh = false;
|
||||
|
||||
/* query the list of nodes allocated to the job - don't need the entire
|
||||
* mapping - as the daemon/proxy is responsibe for determining the apps
|
||||
* to launch on each node.
|
||||
/* Query the list of nodes allocated and mapped to this job.
|
||||
* We need the entire mapping for a couple of reasons:
|
||||
* - need the prefix to start with.
|
||||
* - need to know if we are launching on a subset of the allocated nodes
|
||||
* All other mapping responsibilities fall to orted in the fork PLS
|
||||
*/
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
|
||||
rc = orte_ras_base_node_query_alloc(&nodes, jobid);
|
||||
OBJ_CONSTRUCT(&mapping_list, opal_list_t);
|
||||
rc = orte_rmaps_base_mapped_node_query(&mapping_list, &nodes, jobid);
|
||||
if (ORTE_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
@ -944,8 +946,15 @@ cleanup:
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
|
||||
while (NULL != (item = opal_list_remove_first(&mapping_list))) {
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&mapping_list);
|
||||
|
||||
free(jobid_string); /* done with this variable */
|
||||
opal_argv_free(argv);
|
||||
|
||||
return rc;
|
||||
}
|
||||
|
||||
|
||||
@ -81,7 +81,7 @@ extern char **environ;
|
||||
|
||||
static int pls_slurm_launch(orte_jobid_t jobid)
|
||||
{
|
||||
opal_list_t nodes;
|
||||
opal_list_t nodes, mapping_list;
|
||||
opal_list_item_t *item, *item2;
|
||||
size_t num_nodes;
|
||||
orte_vpid_t vpid;
|
||||
@ -102,13 +102,15 @@ static int pls_slurm_launch(orte_jobid_t jobid)
|
||||
int num_args, i;
|
||||
char *cur_prefix;
|
||||
|
||||
/* query the list of nodes allocated to the job - don't need the entire
|
||||
* mapping - as the daemon/proxy is responsibe for determining the apps
|
||||
* to launch on each node.
|
||||
/* Query the list of nodes allocated and mapped to this job.
|
||||
* We need the entire mapping for a couple of reasons:
|
||||
* - need the prefix to start with.
|
||||
* - need to know if we are launching on a subset of the allocated nodes
|
||||
* All other mapping responsibilities fall to orted in the fork PLS
|
||||
*/
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
|
||||
rc = orte_ras_base_node_query_alloc(&nodes, jobid);
|
||||
OBJ_CONSTRUCT(&mapping_list, opal_list_t);
|
||||
rc = orte_rmaps_base_mapped_node_query(&mapping_list, &nodes, jobid);
|
||||
if (ORTE_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
@ -379,6 +381,11 @@ cleanup:
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
|
||||
while (NULL != (item = opal_list_remove_first(&mapping_list))) {
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&mapping_list);
|
||||
return rc;
|
||||
}
|
||||
|
||||
|
||||
@ -89,7 +89,7 @@ extern char **environ;
|
||||
static int
|
||||
pls_tm_launch(orte_jobid_t jobid)
|
||||
{
|
||||
opal_list_t nodes;
|
||||
opal_list_t nodes, mapping_list;
|
||||
opal_list_item_t *item, *item2;
|
||||
size_t num_nodes;
|
||||
orte_vpid_t vpid;
|
||||
@ -104,13 +104,15 @@ pls_tm_launch(orte_jobid_t jobid)
|
||||
opal_list_t map;
|
||||
char *cur_prefix;
|
||||
|
||||
/* query the list of nodes allocated to the job - don't need the entire
|
||||
* mapping - as the daemon/proxy is responsibe for determining the apps
|
||||
* to launch on each node.
|
||||
/* Query the list of nodes allocated and mapped to this job.
|
||||
* We need the entire mapping for a couple of reasons:
|
||||
* - need the prefix to start with.
|
||||
* - need to know if we are launching on a subset of the allocated nodes
|
||||
* All other mapping responsibilities fall to orted in the fork PLS
|
||||
*/
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
|
||||
rc = orte_ras_base_node_query_alloc(&nodes, jobid);
|
||||
OBJ_CONSTRUCT(&mapping_list, opal_list_t);
|
||||
rc = orte_rmaps_base_mapped_node_query(&mapping_list, &nodes, jobid);
|
||||
if (ORTE_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
@ -391,6 +393,12 @@ cleanup:
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
|
||||
while (NULL != (item = opal_list_remove_first(&mapping_list))) {
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&mapping_list);
|
||||
|
||||
return rc;
|
||||
}
|
||||
|
||||
|
||||
@ -237,7 +237,7 @@ mca_pls_xgrid_set_node_name(orte_ras_node_t* node,
|
||||
|
||||
-(int) launchJob:(orte_jobid_t) jobid
|
||||
{
|
||||
opal_list_t nodes;
|
||||
opal_list_t nodes, mapping_list;
|
||||
opal_list_item_t *item;
|
||||
int ret;
|
||||
size_t num_nodes;
|
||||
@ -250,7 +250,8 @@ mca_pls_xgrid_set_node_name(orte_ras_node_t* node,
|
||||
|
||||
/* query the list of nodes allocated to the job */
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
ret = orte_ras_base_node_query_alloc(&nodes, jobid);
|
||||
OBJ_CONSTRUCT(&mapping_list, opal_list_t);
|
||||
rc = orte_rmaps_base_mapped_node_query(&mapping_list, &nodes, jobid);
|
||||
if (ORTE_SUCCESS != ret) goto cleanup;
|
||||
|
||||
/* allocate vpids for the daemons */
|
||||
@ -356,6 +357,12 @@ cleanup:
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
|
||||
while(NULL != (item = opal_list_remove_first(&mapping_list))) {
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&mapping_list);
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
|
||||
@ -14,8 +14,6 @@
|
||||
# $HEADER$
|
||||
#
|
||||
|
||||
|
||||
|
||||
noinst_LTLIBRARIES = libmca_ras_base.la
|
||||
|
||||
# Source code files
|
||||
@ -28,12 +26,12 @@ headers = \
|
||||
libmca_ras_base_la_SOURCES = \
|
||||
$(headers) \
|
||||
ras_base_alloc.c \
|
||||
ras_base_allocate.c \
|
||||
ras_base_close.c \
|
||||
ras_base_find_available.c \
|
||||
ras_base_node.h \
|
||||
ras_base_node.c \
|
||||
ras_base_open.c \
|
||||
ras_base_select.c
|
||||
|
||||
ras_base_open.c
|
||||
|
||||
# Conditionally install the header files
|
||||
|
||||
|
||||
@ -56,15 +56,14 @@ typedef struct orte_ras_base_cmp_t orte_ras_base_cmp_t;
|
||||
* function definitions
|
||||
*/
|
||||
ORTE_DECLSPEC int orte_ras_base_open(void);
|
||||
ORTE_DECLSPEC int orte_ras_base_find_available(void);
|
||||
ORTE_DECLSPEC int orte_ras_base_allocate(orte_jobid_t job,
|
||||
orte_ras_base_module_t **m);
|
||||
ORTE_DECLSPEC int orte_ras_base_finalize(void);
|
||||
ORTE_DECLSPEC int orte_ras_base_close(void);
|
||||
ORTE_DECLSPEC orte_ras_base_module_t* orte_ras_base_select(const char*);
|
||||
ORTE_DECLSPEC int orte_ras_base_allocate(orte_jobid_t job);
|
||||
ORTE_DECLSPEC int orte_ras_base_deallocate(orte_jobid_t job);
|
||||
ORTE_DECLSPEC int orte_ras_base_allocate_nodes_by_node(orte_jobid_t jobid,
|
||||
opal_list_t* nodes);
|
||||
ORTE_DECLSPEC int orte_ras_base_allocate_nodes_by_slot(orte_jobid_t jobid,
|
||||
opal_list_t* nodes);
|
||||
|
||||
ORTE_DECLSPEC int orte_ras_base_allocate_nodes(orte_jobid_t jobid,
|
||||
opal_list_t* nodes);
|
||||
|
||||
/*
|
||||
* globals that might be needed
|
||||
@ -74,7 +73,9 @@ ORTE_DECLSPEC int orte_ras_base_allocate_nodes_by_slot(orte_jobid_t jobid,
|
||||
typedef struct orte_ras_base_t {
|
||||
int ras_output;
|
||||
opal_list_t ras_opened;
|
||||
bool ras_opened_valid;
|
||||
opal_list_t ras_available;
|
||||
bool ras_available_valid;
|
||||
size_t ras_num_nodes;
|
||||
} orte_ras_base_t;
|
||||
|
||||
|
||||
@ -25,209 +25,28 @@
|
||||
#include "mca/rmgr/base/base.h"
|
||||
#include "mca/errmgr/errmgr.h"
|
||||
|
||||
|
||||
/*
|
||||
* Allocate one process per node on a round-robin basis, looping back
|
||||
* around to the beginning as necessary
|
||||
* Mark nodes as allocated on the registry
|
||||
*/
|
||||
int orte_ras_base_allocate_nodes_by_node(orte_jobid_t jobid,
|
||||
opal_list_t* nodes)
|
||||
int orte_ras_base_allocate_nodes(orte_jobid_t jobid,
|
||||
opal_list_t* nodes)
|
||||
{
|
||||
opal_list_t allocated;
|
||||
opal_list_item_t* item;
|
||||
size_t num_requested = 0;
|
||||
size_t num_allocated = 0;
|
||||
size_t num_constrained = 0;
|
||||
size_t slots;
|
||||
bool oversubscribe = false;
|
||||
int rc;
|
||||
|
||||
/* query for the number of process slots required */
|
||||
if (ORTE_SUCCESS !=
|
||||
(rc = orte_rmgr_base_get_job_slots(jobid, &num_requested))) {
|
||||
return rc;
|
||||
}
|
||||
|
||||
OBJ_CONSTRUCT(&allocated, opal_list_t);
|
||||
num_allocated = 0;
|
||||
|
||||
/* This loop continues until all procs have been allocated or we run
|
||||
out of resources. There are two definitions of "run out of
|
||||
resources":
|
||||
|
||||
1. All nodes have node_slots processes allocated to them
|
||||
2. All nodes have node_slots_max processes allocated to them
|
||||
|
||||
We first map until condition #1 is met. If there are still
|
||||
processes that haven't been allocated yet, then we continue
|
||||
until condition #2 is met. If we still have processes that
|
||||
haven't been allocated yet, then it's an "out of resources"
|
||||
error. */
|
||||
while (num_allocated < num_requested) {
|
||||
num_constrained = 0;
|
||||
|
||||
/* loop over all nodes until either all processes are
|
||||
allocated or they all become constrained */
|
||||
for (item = opal_list_get_first(nodes);
|
||||
item != opal_list_get_end(nodes) && num_allocated < num_requested;
|
||||
item = opal_list_get_next(item)) {
|
||||
orte_ras_node_t* node = (orte_ras_node_t*)item;
|
||||
|
||||
/* are any slots available? */
|
||||
slots = (oversubscribe ? node->node_slots_max : node->node_slots);
|
||||
if (node->node_slots_inuse < slots ||
|
||||
(oversubscribe && 0 == slots)) {
|
||||
++num_allocated;
|
||||
++node->node_slots_inuse; /* running total */
|
||||
++node->node_slots_alloc; /* this job */
|
||||
} else {
|
||||
++num_constrained;
|
||||
}
|
||||
}
|
||||
|
||||
/* if all nodes are constrained:
|
||||
- if this is the first time through the loop, then set
|
||||
"oversubscribe" to true, and we'll now start obeying
|
||||
node_slots_max instead of node_slots
|
||||
- if this is the second time through the loop, then all
|
||||
nodes are full to the max, and therefore we can't do
|
||||
anything more -- we're out of resources */
|
||||
if (opal_list_get_size(nodes) == num_constrained) {
|
||||
if (oversubscribe) {
|
||||
rc = ORTE_ERR_OUT_OF_RESOURCE;
|
||||
goto cleanup;
|
||||
} else {
|
||||
oversubscribe = true;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* move all nodes w/ allocations to the allocated list */
|
||||
item = opal_list_get_first(nodes);
|
||||
while(item != opal_list_get_end(nodes)) {
|
||||
/* Increment the allocation field on each node so we know that
|
||||
* it has been allocated to us. No further logic needed, that is left to rmaps */
|
||||
for (item = opal_list_get_first(nodes);
|
||||
item != opal_list_get_end(nodes);
|
||||
item = opal_list_get_next(item)) {
|
||||
orte_ras_node_t* node = (orte_ras_node_t*)item;
|
||||
opal_list_item_t* next = opal_list_get_next(item);
|
||||
if(node->node_slots_alloc) {
|
||||
opal_list_remove_item(nodes, item);
|
||||
opal_list_append(&allocated, item);
|
||||
}
|
||||
item = next;
|
||||
node->node_slots_alloc++;
|
||||
}
|
||||
|
||||
rc = orte_ras_base_node_assign(&allocated, jobid);
|
||||
if(ORTE_SUCCESS != rc) {
|
||||
ORTE_ERROR_LOG(rc);
|
||||
}
|
||||
cleanup:
|
||||
|
||||
while(NULL != (item = opal_list_remove_first(&allocated)))
|
||||
opal_list_append(nodes, item);
|
||||
OBJ_DESTRUCT(&allocated);
|
||||
return rc;
|
||||
}
|
||||
|
||||
|
||||
/*
|
||||
* Allocate processes to nodes, using all available slots on a node.
|
||||
*/
|
||||
int orte_ras_base_allocate_nodes_by_slot(orte_jobid_t jobid,
|
||||
opal_list_t* nodes)
|
||||
{
|
||||
opal_list_t allocated;
|
||||
opal_list_item_t* item;
|
||||
size_t num_requested = 0;
|
||||
size_t num_allocated = 0;
|
||||
size_t num_constrained = 0;
|
||||
size_t available;
|
||||
int rc;
|
||||
|
||||
/* query for the number of process slots required */
|
||||
if (ORTE_SUCCESS !=
|
||||
(rc = orte_rmgr_base_get_job_slots(jobid, &num_requested))) {
|
||||
return rc;
|
||||
}
|
||||
|
||||
OBJ_CONSTRUCT(&allocated, opal_list_t);
|
||||
num_allocated = 0;
|
||||
|
||||
/* In the first pass, just grab all available slots (i.e., stay <=
|
||||
node_slots) greedily off each node */
|
||||
for (item = opal_list_get_first(nodes);
|
||||
item != opal_list_get_end(nodes) && num_allocated < num_requested;
|
||||
item = opal_list_get_next(item)) {
|
||||
orte_ras_node_t* node = (orte_ras_node_t*)item;
|
||||
|
||||
/* are any slots available? */
|
||||
if (node->node_slots_inuse < node->node_slots) {
|
||||
available = node->node_slots - node->node_slots_inuse;
|
||||
if (num_requested - num_allocated < available) {
|
||||
node->node_slots_inuse +=
|
||||
(num_requested - num_allocated); /* running total */
|
||||
node->node_slots_alloc +=
|
||||
(num_requested - num_allocated); /* this job */
|
||||
num_allocated = num_requested;
|
||||
} else {
|
||||
num_allocated += available;
|
||||
node->node_slots_inuse += available; /* running total */
|
||||
node->node_slots_alloc += available; /* this job */
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* If we're not done, then we're in an oversubscribing situation.
|
||||
Switch to a round-robin-by-node policy -- take one slot from
|
||||
each node until we hit node_slots_max or we have no more
|
||||
resources; whichever occurs first. */
|
||||
while (num_allocated < num_requested) {
|
||||
num_constrained = 0;
|
||||
|
||||
/* loop over all nodes until either all processes are
|
||||
allocated or they all become constrained */
|
||||
for (item = opal_list_get_first(nodes);
|
||||
item != opal_list_get_end(nodes) && num_allocated < num_requested;
|
||||
item = opal_list_get_next(item)) {
|
||||
orte_ras_node_t* node = (orte_ras_node_t*)item;
|
||||
|
||||
/* are any slots available? */
|
||||
if (node->node_slots_inuse < node->node_slots_max ||
|
||||
0 == node->node_slots_max) {
|
||||
++num_allocated;
|
||||
++node->node_slots_inuse; /* running total */
|
||||
++node->node_slots_alloc; /* this job */
|
||||
} else {
|
||||
++num_constrained;
|
||||
}
|
||||
}
|
||||
|
||||
/* if all nodes are constrained, then we're out of resources
|
||||
-- thanks for playing */
|
||||
if (opal_list_get_size(nodes) == num_constrained) {
|
||||
rc = ORTE_ERR_OUT_OF_RESOURCE;
|
||||
goto cleanup;
|
||||
}
|
||||
}
|
||||
|
||||
/* move all nodes w/ allocations to the allocated list */
|
||||
item = opal_list_get_first(nodes);
|
||||
while(item != opal_list_get_end(nodes)) {
|
||||
orte_ras_node_t* node = (orte_ras_node_t*)item;
|
||||
opal_list_item_t* next = opal_list_get_next(item);
|
||||
if(node->node_slots_alloc) {
|
||||
opal_list_remove_item(nodes, item);
|
||||
opal_list_append(&allocated, item);
|
||||
}
|
||||
item = next;
|
||||
}
|
||||
|
||||
rc = orte_ras_base_node_assign(&allocated, jobid);
|
||||
rc = orte_ras_base_node_assign(nodes, jobid);
|
||||
if(ORTE_SUCCESS != rc) {
|
||||
ORTE_ERROR_LOG(rc);
|
||||
}
|
||||
|
||||
cleanup:
|
||||
|
||||
while(NULL != (item = opal_list_remove_first(&allocated)))
|
||||
opal_list_append(nodes, item);
|
||||
OBJ_DESTRUCT(&allocated);
|
||||
return rc;
|
||||
}
|
||||
|
||||
93
orte/mca/ras/base/ras_base_allocate.c
Обычный файл
93
orte/mca/ras/base/ras_base_allocate.c
Обычный файл
@ -0,0 +1,93 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
|
||||
#include "orte_config.h"
|
||||
|
||||
#include "opal/mca/mca.h"
|
||||
#include "opal/mca/base/base.h"
|
||||
#include "opal/util/output.h"
|
||||
#include "orte/include/orte_constants.h"
|
||||
#include "orte/mca/ras/base/base.h"
|
||||
#include "orte/mca/ras/base/ras_base_node.h"
|
||||
#include "orte/mca/errmgr/errmgr.h"
|
||||
|
||||
|
||||
/*
|
||||
* Function for selecting one component from all those that are
|
||||
* available.
|
||||
*/
|
||||
int orte_ras_base_allocate(orte_jobid_t jobid,
|
||||
orte_ras_base_module_t **module)
|
||||
{
|
||||
int ret;
|
||||
opal_list_item_t *item;
|
||||
orte_ras_base_cmp_t *cmp;
|
||||
|
||||
/* If the list is empty, return NULL */
|
||||
|
||||
if (opal_list_is_empty(&orte_ras_base.ras_available)) {
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:ras:base:select: no components available!");
|
||||
ret = ORTE_ERR_NOT_FOUND;
|
||||
ORTE_ERROR_LOG(ret);
|
||||
return ret;
|
||||
}
|
||||
|
||||
/* Otherwise, go through the [already sorted in priority order]
|
||||
list and initialize them until one of them puts something on
|
||||
the node segment */
|
||||
|
||||
for (item = opal_list_get_first(&orte_ras_base.ras_available);
|
||||
item != opal_list_get_end(&orte_ras_base.ras_available);
|
||||
item = opal_list_get_next(item)) {
|
||||
cmp = (orte_ras_base_cmp_t *) item;
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:ras:base:allocate: attemping to allocate using module: %s",
|
||||
cmp->component->ras_version.mca_component_name);
|
||||
|
||||
if (NULL != cmp->module->allocate) {
|
||||
ret = cmp->module->allocate(jobid);
|
||||
if (ORTE_SUCCESS == ret) {
|
||||
bool empty;
|
||||
|
||||
if (ORTE_SUCCESS !=
|
||||
(ret = orte_ras_base_node_segment_empty(&empty))) {
|
||||
ORTE_ERROR_LOG(ret);
|
||||
return ret;
|
||||
}
|
||||
|
||||
/* If this module put something on the node segment,
|
||||
we're done */
|
||||
|
||||
if (!empty) {
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:ras:base:allocate: found good module: %s",
|
||||
cmp->component->ras_version.mca_component_name);
|
||||
*module = cmp->module;
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* We didn't find anyone who put anything on the node segment */
|
||||
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:ras:base:allocate: no module put anything in the node segment");
|
||||
ret = ORTE_ERR_NOT_FOUND;
|
||||
ORTE_ERROR_LOG(ret);
|
||||
return ret;
|
||||
}
|
||||
@ -20,10 +20,9 @@
|
||||
|
||||
#include "include/orte_constants.h"
|
||||
|
||||
#include "mca/mca.h"
|
||||
#include "mca/base/base.h"
|
||||
|
||||
#include "mca/ras/base/base.h"
|
||||
#include "opal/mca/mca.h"
|
||||
#include "opal/mca/base/base.h"
|
||||
#include "orte/mca/ras/base/base.h"
|
||||
|
||||
|
||||
int orte_ras_base_finalize(void)
|
||||
@ -31,14 +30,20 @@ int orte_ras_base_finalize(void)
|
||||
opal_list_item_t* item;
|
||||
|
||||
/* Finalize all available modules */
|
||||
while((item = opal_list_remove_first(&orte_ras_base.ras_available)) != NULL) {
|
||||
orte_ras_base_cmp_t* cmp = (orte_ras_base_cmp_t*)item;
|
||||
cmp->module->finalize();
|
||||
OBJ_RELEASE(cmp);
|
||||
if (orte_ras_base.ras_available_valid) {
|
||||
while (NULL !=
|
||||
(item = opal_list_remove_first(&orte_ras_base.ras_available))) {
|
||||
orte_ras_base_cmp_t* cmp = (orte_ras_base_cmp_t*)item;
|
||||
cmp->module->finalize();
|
||||
OBJ_RELEASE(cmp);
|
||||
}
|
||||
OBJ_DESTRUCT(&orte_ras_base.ras_available);
|
||||
}
|
||||
return OMPI_SUCCESS;
|
||||
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
int orte_ras_base_close(void)
|
||||
{
|
||||
/* Close all remaining available components (may be one if this is a
|
||||
@ -47,7 +52,6 @@ int orte_ras_base_close(void)
|
||||
mca_base_components_close(orte_ras_base.ras_output,
|
||||
&orte_ras_base.ras_opened, NULL);
|
||||
|
||||
OBJ_DESTRUCT(&orte_ras_base.ras_available);
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
127
orte/mca/ras/base/ras_base_find_available.c
Обычный файл
127
orte/mca/ras/base/ras_base_find_available.c
Обычный файл
@ -0,0 +1,127 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
|
||||
#include "orte_config.h"
|
||||
|
||||
#include "include/orte_constants.h"
|
||||
|
||||
#include "mca/mca.h"
|
||||
#include "mca/base/base.h"
|
||||
#include "opal/util/output.h"
|
||||
#include "mca/ras/base/base.h"
|
||||
|
||||
|
||||
/*
|
||||
* Local functions
|
||||
*/
|
||||
static void orte_ras_base_cmp_constructor(orte_ras_base_cmp_t *cmp);
|
||||
static int compare(opal_list_item_t **a, opal_list_item_t **b);
|
||||
|
||||
|
||||
/*
|
||||
* Global variables
|
||||
*/
|
||||
OBJ_CLASS_INSTANCE(orte_ras_base_cmp_t,
|
||||
opal_list_item_t,
|
||||
orte_ras_base_cmp_constructor,
|
||||
NULL);
|
||||
|
||||
|
||||
/*
|
||||
* Find all available RAS components and sort them according to
|
||||
* priority
|
||||
*/
|
||||
int orte_ras_base_find_available(void)
|
||||
{
|
||||
opal_list_item_t *item;
|
||||
mca_base_component_list_item_t *cli;
|
||||
orte_ras_base_component_t *component;
|
||||
orte_ras_base_module_t *module;
|
||||
int priority;
|
||||
orte_ras_base_cmp_t *cmp;
|
||||
|
||||
OBJ_CONSTRUCT(&orte_ras_base.ras_available, opal_list_t);
|
||||
orte_ras_base.ras_available_valid = true;
|
||||
|
||||
for (item = opal_list_get_first(&orte_ras_base.ras_opened);
|
||||
opal_list_get_end(&orte_ras_base.ras_opened) != item;
|
||||
item = opal_list_get_next(item)) {
|
||||
cli = (mca_base_component_list_item_t *) item;
|
||||
component = (orte_ras_base_component_t *) cli->cli_component;
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:ras:base:open: querying component %s",
|
||||
component->ras_version.mca_component_name);
|
||||
|
||||
/* Call the component's init function and see if it wants to be
|
||||
selected */
|
||||
|
||||
module = component->ras_init(&priority);
|
||||
|
||||
/* If we got a non-NULL module back, then the component wants
|
||||
to be considered for selection */
|
||||
|
||||
if (NULL != module) {
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:ras:base:open: component %s returns priority %d",
|
||||
component->ras_version.mca_component_name,
|
||||
priority);
|
||||
|
||||
cmp = OBJ_NEW(orte_ras_base_cmp_t);
|
||||
cmp->component = component;
|
||||
cmp->module = module;
|
||||
cmp->priority = priority;
|
||||
|
||||
opal_list_append(&orte_ras_base.ras_available, &cmp->super);
|
||||
} else {
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:ras:base:open: component %s does NOT want to be considered for selection",
|
||||
component->ras_version.mca_component_name);
|
||||
}
|
||||
}
|
||||
|
||||
/* Sort the resulting available list in priority order */
|
||||
opal_list_sort(&orte_ras_base.ras_available, compare);
|
||||
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
|
||||
static void orte_ras_base_cmp_constructor(orte_ras_base_cmp_t *cmp)
|
||||
{
|
||||
cmp->component = NULL;
|
||||
cmp->module = NULL;
|
||||
cmp->priority = -1;
|
||||
}
|
||||
|
||||
/*
|
||||
* Need to make this an *opposite* compare (this is invoked by qsort)
|
||||
* so that we get the highest priority first (i.e., so the sort is
|
||||
* highest->lowest, not lowest->highest)
|
||||
*/
|
||||
static int compare(opal_list_item_t **a, opal_list_item_t **b)
|
||||
{
|
||||
orte_ras_base_cmp_t *aa = *((orte_ras_base_cmp_t **) a);
|
||||
orte_ras_base_cmp_t *bb = *((orte_ras_base_cmp_t **) b);
|
||||
|
||||
if (bb->priority > aa->priority) {
|
||||
return 1;
|
||||
} else if (bb->priority == aa->priority) {
|
||||
return 0;
|
||||
} else {
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
@ -134,137 +134,6 @@ int orte_ras_base_node_query(opal_list_t* nodes)
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
/*
|
||||
* Query the registry for all available nodes
|
||||
*/
|
||||
|
||||
int orte_ras_base_node_query_context(
|
||||
opal_list_t* nodes,
|
||||
orte_app_context_t** context,
|
||||
size_t num_context,
|
||||
bool *constrained)
|
||||
{
|
||||
size_t i, cnt;
|
||||
orte_gpr_value_t** values;
|
||||
int rc;
|
||||
opal_list_t required;
|
||||
opal_list_item_t* item;
|
||||
|
||||
/* expand the list of node/host specifications on the context structure into
|
||||
* a list of node names
|
||||
*/
|
||||
*constrained = false;
|
||||
OBJ_CONSTRUCT(&required, opal_list_t);
|
||||
for(i=0; i<num_context; i++) {
|
||||
orte_app_context_map_t** map = context[i]->map_data;
|
||||
size_t m, num_map = context[i]->num_map;
|
||||
for(m=0; m<num_map; m++) {
|
||||
if(map[m]->map_type == ORTE_APP_CONTEXT_MAP_HOSTNAME) {
|
||||
char** hosts = opal_argv_split(map[m]->map_data, ',');
|
||||
char** ptr = hosts;
|
||||
while(*ptr) {
|
||||
orte_ras_node_t* node = OBJ_NEW(orte_ras_node_t);
|
||||
node->node_name = strdup(*ptr);
|
||||
opal_list_append(&required, (opal_list_item_t*)node);
|
||||
ptr++;
|
||||
}
|
||||
opal_argv_free(hosts);
|
||||
*constrained = true;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* query all node entries */
|
||||
rc = orte_gpr.get(
|
||||
ORTE_GPR_KEYS_OR|ORTE_GPR_TOKENS_OR,
|
||||
ORTE_NODE_SEGMENT,
|
||||
NULL,
|
||||
NULL,
|
||||
&cnt,
|
||||
&values);
|
||||
if(ORTE_SUCCESS != rc) {
|
||||
ORTE_ERROR_LOG(rc);
|
||||
return rc;
|
||||
}
|
||||
|
||||
/* parse the response */
|
||||
for(i=0; i<cnt; i++) {
|
||||
orte_gpr_value_t* value = values[i];
|
||||
orte_ras_node_t* node = OBJ_NEW(orte_ras_node_t);
|
||||
size_t k;
|
||||
bool found = false;
|
||||
|
||||
for(k=0; k<value->cnt; k++) {
|
||||
orte_gpr_keyval_t* keyval = value->keyvals[k];
|
||||
if(strcmp(keyval->key, ORTE_NODE_NAME_KEY) == 0) {
|
||||
node->node_name = strdup(keyval->value.strptr);
|
||||
continue;
|
||||
}
|
||||
if(strcmp(keyval->key, ORTE_NODE_ARCH_KEY) == 0) {
|
||||
node->node_arch = strdup(keyval->value.strptr);
|
||||
continue;
|
||||
}
|
||||
if(strcmp(keyval->key, ORTE_NODE_STATE_KEY) == 0) {
|
||||
node->node_state = keyval->value.node_state;
|
||||
continue;
|
||||
}
|
||||
if(strcmp(keyval->key, ORTE_NODE_SLOTS_KEY) == 0) {
|
||||
node->node_slots = keyval->value.size;
|
||||
continue;
|
||||
}
|
||||
if(strncmp(keyval->key, ORTE_NODE_SLOTS_ALLOC_KEY, strlen(ORTE_NODE_SLOTS_ALLOC_KEY)) == 0) {
|
||||
node->node_slots_inuse += keyval->value.size;
|
||||
continue;
|
||||
}
|
||||
if(strcmp(keyval->key, ORTE_NODE_SLOTS_MAX_KEY) == 0) {
|
||||
node->node_slots_max = keyval->value.size;
|
||||
continue;
|
||||
}
|
||||
if(strcmp(keyval->key, ORTE_NODE_USERNAME_KEY) == 0) {
|
||||
node->node_username = strdup(keyval->value.strptr);
|
||||
continue;
|
||||
}
|
||||
if(strcmp(keyval->key, ORTE_CELLID_KEY) == 0) {
|
||||
node->node_cellid = keyval->value.cellid;
|
||||
continue;
|
||||
}
|
||||
}
|
||||
|
||||
/* contained in app_context? */
|
||||
for(item = opal_list_get_first(&required);
|
||||
item != opal_list_get_end(&required);
|
||||
item = opal_list_get_next(item)) {
|
||||
if(0 == strcmp(((orte_ras_node_t*)item)->node_name,node->node_name)) {
|
||||
opal_list_remove_item(&required, item);
|
||||
OBJ_RELEASE(item);
|
||||
found = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if(*constrained == false || found) {
|
||||
opal_list_append(nodes, &node->super);
|
||||
} else {
|
||||
OBJ_RELEASE(node);
|
||||
}
|
||||
OBJ_RELEASE(value);
|
||||
}
|
||||
|
||||
/* append any remaining specified nodes to the list with
|
||||
* with default settings for slots,etc.
|
||||
*/
|
||||
if(opal_list_get_size(&required)) {
|
||||
if(ORTE_SUCCESS != (rc = orte_ras_base_node_insert(&required))) {
|
||||
ORTE_ERROR_LOG(rc);
|
||||
return rc;
|
||||
}
|
||||
while(NULL != (item = opal_list_remove_first(&required))) {
|
||||
opal_list_append(nodes, item);
|
||||
}
|
||||
}
|
||||
OBJ_DESTRUCT(&required);
|
||||
if (NULL != values) free(values);
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
/*
|
||||
* Query the registry for all nodes allocated to a specified job
|
||||
@ -665,3 +534,33 @@ int orte_ras_base_node_assign(opal_list_t* nodes, orte_jobid_t jobid)
|
||||
return rc;
|
||||
}
|
||||
|
||||
|
||||
int orte_ras_base_node_segment_empty(bool *empty)
|
||||
{
|
||||
int ret;
|
||||
opal_list_t nodes;
|
||||
opal_list_item_t *item;
|
||||
|
||||
/* See what's already on the node segment */
|
||||
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
if (ORTE_SUCCESS != (ret = orte_ras_base_node_query(&nodes))) {
|
||||
ORTE_ERROR_LOG(ret);
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
return ret;
|
||||
}
|
||||
|
||||
*empty = opal_list_is_empty(&nodes) ? true : false;
|
||||
|
||||
/* Free the list */
|
||||
|
||||
while (NULL != (item = opal_list_remove_first(&nodes))) {
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
|
||||
/* All done */
|
||||
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
@ -13,10 +13,8 @@
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
/**
|
||||
* @file
|
||||
*
|
||||
*/
|
||||
/** @file */
|
||||
|
||||
#ifndef ORTE_RAS_BASE_NODE_H
|
||||
#define ORTE_RAS_BASE_NODE_H
|
||||
|
||||
@ -24,47 +22,42 @@
|
||||
#include "mca/soh/soh_types.h"
|
||||
#include "mca/rmgr/rmgr_types.h"
|
||||
#include "mca/ras/ras.h"
|
||||
|
||||
#if defined(c_plusplus) || defined(__cplusplus)
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
|
||||
/*
|
||||
* Convience routines to query/set node state in the registry
|
||||
*/
|
||||
|
||||
/*
|
||||
* Query the registry for all available nodes
|
||||
*/
|
||||
int orte_ras_base_node_query(opal_list_t*);
|
||||
|
||||
/*
|
||||
* Query the registry for all available nodes that satisfy any
|
||||
* constraints specified on the app_context(s)
|
||||
*/
|
||||
int orte_ras_base_node_query_context(opal_list_t*, orte_app_context_t**, size_t num_context, bool* constrained);
|
||||
|
||||
/*
|
||||
/**
|
||||
* Query the registry for all nodes allocated to a specific job
|
||||
*/
|
||||
int orte_ras_base_node_query_alloc(opal_list_t*, orte_jobid_t);
|
||||
|
||||
/*
|
||||
/**
|
||||
* Add the specified node definitions to the registry
|
||||
*/
|
||||
int orte_ras_base_node_insert(opal_list_t*);
|
||||
|
||||
/*
|
||||
/**
|
||||
* Delete the specified nodes from the registry
|
||||
*/
|
||||
int orte_ras_base_node_delete(opal_list_t*);
|
||||
|
||||
/*
|
||||
/**
|
||||
* Assign the allocated slots on the specified nodes to the
|
||||
* indicated jobid.
|
||||
*/
|
||||
int orte_ras_base_node_assign(opal_list_t*, orte_jobid_t);
|
||||
|
||||
/**
|
||||
* Check to see if the node segment is empty
|
||||
*/
|
||||
int orte_ras_base_node_segment_empty(bool *empty);
|
||||
|
||||
|
||||
#if defined(c_plusplus) || defined(__cplusplus)
|
||||
}
|
||||
|
||||
@ -33,51 +33,12 @@
|
||||
|
||||
#include "orte/mca/ras/base/static-components.h"
|
||||
|
||||
/**
|
||||
* Local functions.
|
||||
*/
|
||||
|
||||
static void orte_ras_base_cmp_constructor(orte_ras_base_cmp_t *cmp)
|
||||
{
|
||||
cmp->component = NULL;
|
||||
cmp->module = NULL;
|
||||
cmp->priority = -1;
|
||||
}
|
||||
|
||||
static void orte_ras_base_cmp_destructor(orte_ras_base_cmp_t *cmp)
|
||||
{
|
||||
}
|
||||
|
||||
/*
|
||||
* Need to make this an *opposite* compare (this is invoked by qsort)
|
||||
* so that we get the highest priority first (i.e., so the sort is
|
||||
* highest->lowest, not lowest->highest)
|
||||
*/
|
||||
static int compare(opal_list_item_t **a, opal_list_item_t **b)
|
||||
{
|
||||
orte_ras_base_cmp_t *aa = *((orte_ras_base_cmp_t **) a);
|
||||
orte_ras_base_cmp_t *bb = *((orte_ras_base_cmp_t **) b);
|
||||
|
||||
if (bb->priority > aa->priority) {
|
||||
return 1;
|
||||
} else if (bb->priority == aa->priority) {
|
||||
return 0;
|
||||
} else {
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
/*
|
||||
* Global variables
|
||||
*/
|
||||
orte_ras_base_module_t orte_ras;
|
||||
orte_ras_base_t orte_ras_base;
|
||||
OBJ_CLASS_INSTANCE(
|
||||
orte_ras_base_cmp_t,
|
||||
opal_list_item_t,
|
||||
orte_ras_base_cmp_constructor,
|
||||
orte_ras_base_cmp_destructor);
|
||||
|
||||
|
||||
/**
|
||||
@ -86,32 +47,24 @@ OBJ_CLASS_INSTANCE(
|
||||
*/
|
||||
int orte_ras_base_open(void)
|
||||
{
|
||||
opal_list_item_t *item;
|
||||
mca_base_component_list_item_t *cli;
|
||||
orte_ras_base_component_t *component;
|
||||
orte_ras_base_module_t *module;
|
||||
int param, priority, value;
|
||||
orte_ras_base_cmp_t *cmp;
|
||||
char * policy;
|
||||
int value;
|
||||
|
||||
/* Debugging / verbose output */
|
||||
|
||||
orte_ras_base.ras_output = opal_output_open(NULL);
|
||||
param = mca_base_param_reg_int_name("ras_base", "verbose",
|
||||
"Verbosity level for the ras framework",
|
||||
false, false, 0, &value);
|
||||
mca_base_param_reg_int_name("ras_base", "verbose",
|
||||
"Enable debugging for the RAS framework (nonzero = enabled)",
|
||||
false, false, 0, &value);
|
||||
if (value != 0) {
|
||||
orte_ras_base.ras_output = opal_output_open(NULL);
|
||||
} else {
|
||||
orte_ras_base.ras_output = -1;
|
||||
}
|
||||
|
||||
param = mca_base_param_reg_string_name("ras_base", "schedule_policy",
|
||||
"Scheduling Policy for RAS. [slot | node]",
|
||||
false, false, "slot", &policy);
|
||||
if (0 == strcmp(policy, "node")) {
|
||||
mca_base_param_set_string(param, "node");
|
||||
}
|
||||
/* Defaults */
|
||||
|
||||
orte_ras_base.ras_opened_valid = false;
|
||||
orte_ras_base.ras_available_valid = false;
|
||||
|
||||
/* Open up all available components */
|
||||
|
||||
@ -121,48 +74,10 @@ int orte_ras_base_open(void)
|
||||
&orte_ras_base.ras_opened, true)) {
|
||||
return ORTE_ERROR;
|
||||
}
|
||||
OBJ_CONSTRUCT(&orte_ras_base.ras_available, opal_list_t);
|
||||
|
||||
for (item = opal_list_get_first(&orte_ras_base.ras_opened);
|
||||
opal_list_get_end(&orte_ras_base.ras_opened) != item;
|
||||
item = opal_list_get_next(item)) {
|
||||
cli = (mca_base_component_list_item_t *) item;
|
||||
component = (orte_ras_base_component_t *) cli->cli_component;
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:base:open: querying component %s",
|
||||
component->ras_version.mca_component_name);
|
||||
|
||||
/* Call the component's init function and see if it wants to be
|
||||
selected */
|
||||
|
||||
module = component->ras_init(&priority);
|
||||
|
||||
/* If we got a non-NULL module back, then the component wants
|
||||
to be considered for selection */
|
||||
|
||||
if (NULL != module) {
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:base:open: component %s returns priority %d",
|
||||
component->ras_version.mca_component_name,
|
||||
priority);
|
||||
|
||||
cmp = OBJ_NEW(orte_ras_base_cmp_t);
|
||||
cmp->component = component;
|
||||
cmp->module = module;
|
||||
cmp->priority = priority;
|
||||
|
||||
opal_list_append(&orte_ras_base.ras_available, &cmp->super);
|
||||
} else {
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:base:open: component %s does NOT want to be considered for selection",
|
||||
component->ras_version.mca_component_name);
|
||||
}
|
||||
}
|
||||
|
||||
/* Sort the resulting available list in priority order */
|
||||
opal_list_sort(&orte_ras_base.ras_available, compare);
|
||||
|
||||
/* All done */
|
||||
|
||||
orte_ras_base.ras_opened_valid = true;
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
@ -1,102 +0,0 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
|
||||
#include "orte_config.h"
|
||||
|
||||
#include "include/orte_constants.h"
|
||||
|
||||
#include "mca/mca.h"
|
||||
#include "mca/base/base.h"
|
||||
#include "opal/util/output.h"
|
||||
#include "mca/ras/base/base.h"
|
||||
|
||||
|
||||
/*
|
||||
* Local functions
|
||||
*/
|
||||
static orte_ras_base_module_t *select_preferred(const char *name);
|
||||
static orte_ras_base_module_t *select_any(void);
|
||||
|
||||
/*
|
||||
* Function for selecting one component from all those that are
|
||||
* available.
|
||||
*/
|
||||
orte_ras_base_module_t* orte_ras_base_select(const char *preferred)
|
||||
{
|
||||
orte_ras_base_module_t *module;
|
||||
if (NULL != preferred) {
|
||||
module = select_preferred(preferred);
|
||||
} else {
|
||||
module = select_any();
|
||||
}
|
||||
orte_ras = *module;
|
||||
return module;
|
||||
}
|
||||
|
||||
static orte_ras_base_module_t *select_preferred(const char *name)
|
||||
{
|
||||
opal_list_item_t *item;
|
||||
orte_ras_base_cmp_t *cmp;
|
||||
|
||||
/* Look for a matching selected name */
|
||||
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:base:select: looking for component %s", name);
|
||||
for (item = opal_list_get_first(&orte_ras_base.ras_available);
|
||||
item != opal_list_get_end(&orte_ras_base.ras_available);
|
||||
item = opal_list_get_next(item)) {
|
||||
cmp = (orte_ras_base_cmp_t *) item;
|
||||
|
||||
if (0 == strcmp(name,
|
||||
cmp->component->ras_version.mca_component_name)) {
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:base:select: found module for compoent %s", name);
|
||||
return cmp->module;
|
||||
}
|
||||
}
|
||||
|
||||
/* Didn't find a matching name */
|
||||
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:base:select: did not find module for compoent %s", name);
|
||||
return NULL;
|
||||
}
|
||||
|
||||
|
||||
static orte_ras_base_module_t *select_any(void)
|
||||
{
|
||||
opal_list_item_t *item;
|
||||
orte_ras_base_cmp_t *cmp;
|
||||
|
||||
/* If the list is empty, return NULL */
|
||||
|
||||
if (opal_list_is_empty(&orte_ras_base.ras_available)) {
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:base:select: no components available!");
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/* Otherwise, return the first item (it's already sorted in
|
||||
priority order) */
|
||||
|
||||
item = opal_list_get_first(&orte_ras_base.ras_available);
|
||||
cmp = (orte_ras_base_cmp_t *) item;
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:base:select: highest priority component: %s",
|
||||
cmp->component->ras_version.mca_component_name);
|
||||
return cmp->module;
|
||||
}
|
||||
|
||||
@ -113,11 +113,10 @@ static int orte_ras_bjs_discover(
|
||||
opal_list_item_t* item;
|
||||
opal_list_t new_nodes;
|
||||
int rc;
|
||||
bool constrained;
|
||||
|
||||
|
||||
/* query the nodelist from the registry */
|
||||
OBJ_CONSTRUCT(&new_nodes, opal_list_t);
|
||||
if(ORTE_SUCCESS != (rc = orte_ras_base_node_query_context(nodelist, context, num_context, &constrained))) {
|
||||
if(ORTE_SUCCESS != (rc = orte_ras_base_node_query(nodelist))) {
|
||||
ORTE_ERROR_LOG(rc);
|
||||
return rc;
|
||||
}
|
||||
@ -158,10 +157,7 @@ static int orte_ras_bjs_discover(
|
||||
}
|
||||
item = next;
|
||||
}
|
||||
if(constrained) {
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
|
||||
/* parse the node list and check node status/access */
|
||||
nodes = getenv("NODES");
|
||||
@ -254,11 +250,8 @@ static int orte_ras_bjs_allocate(orte_jobid_t jobid)
|
||||
ORTE_ERROR_LOG(rc);
|
||||
goto cleanup;
|
||||
}
|
||||
if (0 == strcmp(mca_ras_bjs_component.schedule_policy, "node")) {
|
||||
rc = orte_ras_base_allocate_nodes_by_node(jobid, &nodes);
|
||||
} else {
|
||||
rc = orte_ras_base_allocate_nodes_by_slot(jobid, &nodes);
|
||||
}
|
||||
|
||||
rc = orte_ras_base_allocate_nodes(jobid, &nodes);
|
||||
if(ORTE_SUCCESS != rc) {
|
||||
ORTE_ERROR_LOG(rc);
|
||||
}
|
||||
|
||||
45
orte/mca/ras/dash_host/Makefile.am
Обычный файл
45
orte/mca/ras/dash_host/Makefile.am
Обычный файл
@ -0,0 +1,45 @@
|
||||
#
|
||||
# Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
# All rights reserved.
|
||||
# Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
# All rights reserved.
|
||||
# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
# University of Stuttgart. All rights reserved.
|
||||
# Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
# All rights reserved.
|
||||
# $COPYRIGHT$
|
||||
#
|
||||
# Additional copyrights may follow
|
||||
#
|
||||
# $HEADER$
|
||||
#
|
||||
|
||||
# Make the output library in this directory, and name it either
|
||||
# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
|
||||
# (for static builds).
|
||||
|
||||
if OMPI_BUILD_ras_dash_host_DSO
|
||||
component_noinst =
|
||||
component_install = mca_ras_dash_host.la
|
||||
else
|
||||
component_noinst = libmca_ras_dash_host.la
|
||||
component_install =
|
||||
endif
|
||||
|
||||
dash_host_SOURCES = \
|
||||
ras_dash_host.h \
|
||||
ras_dash_host_module.c \
|
||||
ras_dash_host_component.c
|
||||
|
||||
mcacomponentdir = $(libdir)/openmpi
|
||||
mcacomponent_LTLIBRARIES = $(component_install)
|
||||
mca_ras_dash_host_la_SOURCES = $(dash_host_SOURCES)
|
||||
mca_ras_dash_host_la_LIBADD = \
|
||||
$(top_ompi_builddir)/orte/liborte.la \
|
||||
$(top_ompi_builddir)/opal/libopal.la
|
||||
mca_ras_dash_host_la_LDFLAGS = -module -avoid-version
|
||||
|
||||
noinst_LTLIBRARIES = $(component_noinst)
|
||||
libmca_ras_dash_host_la_SOURCES = $(dash_host_SOURCES)
|
||||
libmca_ras_dash_host_la_LIBADD =
|
||||
libmca_ras_dash_host_la_LDFLAGS = -module -avoid-version
|
||||
21
orte/mca/ras/dash_host/configure.params
Обычный файл
21
orte/mca/ras/dash_host/configure.params
Обычный файл
@ -0,0 +1,21 @@
|
||||
# -*- shell-script -*-
|
||||
#
|
||||
# Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
# All rights reserved.
|
||||
# Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
# All rights reserved.
|
||||
# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
# University of Stuttgart. All rights reserved.
|
||||
# Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
# All rights reserved.
|
||||
# $COPYRIGHT$
|
||||
#
|
||||
# Additional copyrights may follow
|
||||
#
|
||||
# $HEADER$
|
||||
#
|
||||
|
||||
# Specific to this module
|
||||
|
||||
PARAM_INIT_FILE=ras_dash_host_component.c
|
||||
PARAM_CONFIG_FILES="Makefile"
|
||||
57
orte/mca/ras/dash_host/ras_dash_host.h
Обычный файл
57
orte/mca/ras/dash_host/ras_dash_host.h
Обычный файл
@ -0,0 +1,57 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
/**
|
||||
* @file
|
||||
*
|
||||
* Resource Allocation (dash_host)
|
||||
*/
|
||||
#ifndef ORTE_RAS_DASH_HOST_H
|
||||
#define ORTE_RAS_DASH_HOST_H
|
||||
|
||||
#include "mca/ras/ras.h"
|
||||
#if defined(c_plusplus) || defined(__cplusplus)
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
/**
|
||||
* Dash_host-specific RAS component struct
|
||||
*/
|
||||
struct orte_ras_dash_host_component_t {
|
||||
/** Base RAS component */
|
||||
orte_ras_base_component_t super;
|
||||
/** What's the priority of this component */
|
||||
int priority;
|
||||
};
|
||||
/**
|
||||
* Convenience typedef
|
||||
*/
|
||||
typedef struct orte_ras_dash_host_component_t orte_ras_dash_host_component_t;
|
||||
|
||||
/**
|
||||
* Component export structure
|
||||
*/
|
||||
OMPI_COMP_EXPORT extern orte_ras_dash_host_component_t mca_ras_dash_host_component;
|
||||
|
||||
/**
|
||||
* Module init function
|
||||
*/
|
||||
orte_ras_base_module_t *orte_ras_dash_host_init(int* priority);
|
||||
|
||||
#if defined(c_plusplus) || defined(__cplusplus)
|
||||
}
|
||||
#endif
|
||||
|
||||
#endif
|
||||
75
orte/mca/ras/dash_host/ras_dash_host_component.c
Обычный файл
75
orte/mca/ras/dash_host/ras_dash_host_component.c
Обычный файл
@ -0,0 +1,75 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
|
||||
#include "orte_config.h"
|
||||
|
||||
#include "opal/util/output.h"
|
||||
#include "opal/mca/base/base.h"
|
||||
#include "opal/mca/base/mca_base_param.h"
|
||||
#include "orte/include/orte_constants.h"
|
||||
#include "orte/util/proc_info.h"
|
||||
#include "orte/mca/ras/dash_host/ras_dash_host.h"
|
||||
|
||||
|
||||
/*
|
||||
* Local functions
|
||||
*/
|
||||
static int orte_ras_dash_host_open(void);
|
||||
|
||||
|
||||
orte_ras_dash_host_component_t mca_ras_dash_host_component = {
|
||||
{
|
||||
/* First, the mca_base_component_t struct containing meta
|
||||
information about the component itself */
|
||||
|
||||
{
|
||||
/* Indicate that we are a iof v1.0.0 component (which also
|
||||
implies a specific MCA version) */
|
||||
|
||||
ORTE_RAS_BASE_VERSION_1_0_0,
|
||||
|
||||
"dash_host", /* MCA component name */
|
||||
ORTE_MAJOR_VERSION, /* MCA component major version */
|
||||
ORTE_MINOR_VERSION, /* MCA component minor version */
|
||||
ORTE_RELEASE_VERSION, /* MCA component release version */
|
||||
orte_ras_dash_host_open, /* component open */
|
||||
NULL
|
||||
},
|
||||
|
||||
/* Next the MCA v1.0.0 component meta data */
|
||||
{
|
||||
/* Whether the component is checkpointable or not */
|
||||
false
|
||||
},
|
||||
|
||||
orte_ras_dash_host_init
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
/**
|
||||
* component open function
|
||||
*/
|
||||
static int orte_ras_dash_host_open(void)
|
||||
{
|
||||
mca_base_param_reg_int(&mca_ras_dash_host_component.super.ras_version,
|
||||
"priority",
|
||||
"Selection priority for the dash_host RAS component",
|
||||
false, false, 5,
|
||||
&mca_ras_dash_host_component.priority);
|
||||
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
208
orte/mca/ras/dash_host/ras_dash_host_module.c
Обычный файл
208
orte/mca/ras/dash_host/ras_dash_host_module.c
Обычный файл
@ -0,0 +1,208 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
|
||||
#include "orte_config.h"
|
||||
|
||||
#include "opal/util/output.h"
|
||||
#include "opal/util/argv.h"
|
||||
#include "orte/include/orte_constants.h"
|
||||
#include "orte/include/orte_types.h"
|
||||
#include "orte/mca/ras/base/base.h"
|
||||
#include "orte/mca/ras/base/ras_base_node.h"
|
||||
#include "orte/mca/rmgr/base/base.h"
|
||||
#include "orte/mca/ras/base/ras_base_node.h"
|
||||
#include "orte/mca/errmgr/errmgr.h"
|
||||
|
||||
#include "orte/mca/ras/dash_host/ras_dash_host.h"
|
||||
|
||||
|
||||
/*
|
||||
* Local functions
|
||||
*/
|
||||
static int orte_ras_dash_host_allocate(orte_jobid_t jobid);
|
||||
static int orte_ras_dash_host_deallocate(orte_jobid_t jobid);
|
||||
static int orte_ras_dash_host_finalize(void);
|
||||
|
||||
|
||||
/*
|
||||
* Local variables
|
||||
*/
|
||||
orte_ras_base_module_t orte_ras_dash_host_module = {
|
||||
orte_ras_dash_host_allocate,
|
||||
orte_ras_base_node_insert,
|
||||
orte_ras_base_node_query,
|
||||
orte_ras_dash_host_deallocate,
|
||||
orte_ras_dash_host_finalize
|
||||
};
|
||||
|
||||
|
||||
orte_ras_base_module_t *orte_ras_dash_host_init(int* priority)
|
||||
{
|
||||
*priority = mca_ras_dash_host_component.priority;
|
||||
return &orte_ras_dash_host_module;
|
||||
}
|
||||
|
||||
|
||||
static int orte_ras_dash_host_allocate(orte_jobid_t jobid)
|
||||
{
|
||||
opal_list_t nodes;
|
||||
opal_list_item_t* item;
|
||||
orte_app_context_t **context;
|
||||
size_t i, j, k, num_context;
|
||||
int rc;
|
||||
char **mapped_nodes = NULL, **mini_map;
|
||||
orte_ras_node_t *node;
|
||||
bool empty;
|
||||
|
||||
/* If the node segment is not empty, do nothing */
|
||||
|
||||
if (ORTE_SUCCESS != (rc = orte_ras_base_node_segment_empty(&empty))) {
|
||||
ORTE_ERROR_LOG(rc);
|
||||
return rc;
|
||||
}
|
||||
if (!empty) {
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:ras:dash_host: node segment not empty; not doing anything");
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
/* Otherwise, get the context */
|
||||
|
||||
rc = orte_rmgr_base_get_app_context(jobid, &context, &num_context);
|
||||
if (ORTE_SUCCESS != rc) {
|
||||
ORTE_ERROR_LOG(rc);
|
||||
return rc;
|
||||
}
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
|
||||
/* If there's nothing to do, skip to the end */
|
||||
|
||||
if (0 == num_context) {
|
||||
rc = ORTE_SUCCESS;
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
/* Otherwise, go through the contexts */
|
||||
|
||||
for (i = 0; i < num_context; ++i) {
|
||||
if (context[i]->num_map > 0) {
|
||||
orte_app_context_map_t** map = context[i]->map_data;
|
||||
|
||||
/* Accumulate all of the host name mappings */
|
||||
for (j = 0; j < context[i]->num_map; ++j) {
|
||||
if (ORTE_APP_CONTEXT_MAP_HOSTNAME == map[j]->map_type) {
|
||||
mini_map = opal_argv_split(map[j]->map_data, ',');
|
||||
|
||||
if (mapped_nodes == NULL) {
|
||||
mapped_nodes = mini_map;
|
||||
} else {
|
||||
for (k = 0; NULL != mini_map[k]; ++k) {
|
||||
rc = opal_argv_append_nosize(&mapped_nodes,
|
||||
mini_map[k]);
|
||||
if (OPAL_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* Did we find anything? */
|
||||
|
||||
if (NULL != mapped_nodes) {
|
||||
|
||||
/* Go through the names found and add them to the host list.
|
||||
If they're not unique, then bump the slots count for each
|
||||
duplicate */
|
||||
|
||||
for (i = 0; NULL != mapped_nodes[i]; ++i) {
|
||||
for (item = opal_list_get_begin(&nodes);
|
||||
item != opal_list_get_end(&nodes);
|
||||
item = opal_list_get_next(item)) {
|
||||
node = (orte_ras_node_t*) item;
|
||||
if (0 == strcmp(node->node_name, mapped_nodes[i])) {
|
||||
++node->node_slots;
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
/* If we didn't find it, add it to the list */
|
||||
|
||||
if (item == opal_list_get_end(&nodes)) {
|
||||
node = OBJ_NEW(orte_ras_node_t);
|
||||
if (NULL == node) {
|
||||
return ORTE_ERR_OUT_OF_RESOURCE;
|
||||
}
|
||||
node->node_name = strdup(mapped_nodes[i]);
|
||||
node->node_arch = NULL;
|
||||
node->node_state = ORTE_NODE_STATE_UP;
|
||||
/* JMS: this should not be hard-wired to 0, but there's no
|
||||
other value to put it to [yet]... */
|
||||
node->node_cellid = 0;
|
||||
node->node_slots_inuse = 0;
|
||||
node->node_slots_max = 0;
|
||||
node->node_slots = 1;
|
||||
opal_list_append(&nodes, &node->super);
|
||||
}
|
||||
}
|
||||
|
||||
/* Put them on the segment and allocate them */
|
||||
|
||||
if (ORTE_SUCCESS !=
|
||||
(rc = orte_ras_base_node_insert(&nodes)) ||
|
||||
ORTE_SUCCESS !=
|
||||
(rc = orte_ras_base_allocate_nodes(jobid, &nodes))) {
|
||||
goto cleanup;
|
||||
}
|
||||
}
|
||||
|
||||
cleanup:
|
||||
if (NULL != mapped_nodes) {
|
||||
opal_argv_free(mapped_nodes);
|
||||
}
|
||||
|
||||
while (NULL != (item = opal_list_remove_first(&nodes))) {
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
for (i = 0; i < num_context; i++) {
|
||||
OBJ_RELEASE(context[i]);
|
||||
}
|
||||
free(context);
|
||||
return rc;
|
||||
}
|
||||
|
||||
|
||||
static int orte_ras_dash_host_deallocate(orte_jobid_t jobid)
|
||||
{
|
||||
/* Nothing to do */
|
||||
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"ras:dash_host:deallocate: success (nothing to do)");
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
static int orte_ras_dash_host_finalize(void)
|
||||
{
|
||||
/* Nothing to do */
|
||||
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"ras:dash_host:finalize: success (nothing to do)");
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
@ -1,111 +0,0 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
#include "orte_config.h"
|
||||
#include "include/orte_constants.h"
|
||||
#include "include/orte_types.h"
|
||||
|
||||
#include "mca/ras/base/base.h"
|
||||
#include "mca/ras/base/ras_base_node.h"
|
||||
#include "mca/ras/host/ras_host.h"
|
||||
#include "mca/rmgr/base/base.h"
|
||||
#include "mca/ras/base/ras_base_node.h"
|
||||
#include "mca/errmgr/errmgr.h"
|
||||
|
||||
#if HAVE_STRING_H
|
||||
#include <string.h>
|
||||
#endif /* HAVE_STRING_H */
|
||||
|
||||
/**
|
||||
* Discover available (pre-allocated) nodes. Allocate the
|
||||
* requested number of nodes/process slots to the job.
|
||||
*
|
||||
*/
|
||||
|
||||
static int orte_ras_host_allocate(orte_jobid_t jobid)
|
||||
{
|
||||
opal_list_t nodes;
|
||||
opal_list_item_t* item;
|
||||
orte_app_context_t **context;
|
||||
size_t i, num_context;
|
||||
bool constrained;
|
||||
int rc;
|
||||
|
||||
rc = orte_rmgr_base_get_app_context(jobid, &context, &num_context);
|
||||
if(ORTE_SUCCESS != rc) {
|
||||
ORTE_ERROR_LOG(rc);
|
||||
return rc;
|
||||
}
|
||||
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
if(ORTE_SUCCESS != (rc = orte_ras_base_node_query_context(
|
||||
&nodes, context, num_context, &constrained))) {
|
||||
ORTE_ERROR_LOG(rc);
|
||||
return rc;
|
||||
}
|
||||
|
||||
if (0 == strcmp(mca_ras_host_component.schedule_policy, "node")) {
|
||||
if (ORTE_SUCCESS !=
|
||||
(rc = orte_ras_base_allocate_nodes_by_node(jobid, &nodes))) {
|
||||
goto cleanup;
|
||||
}
|
||||
} else {
|
||||
if (ORTE_SUCCESS !=
|
||||
(rc = orte_ras_base_allocate_nodes_by_slot(jobid, &nodes))) {
|
||||
goto cleanup;
|
||||
}
|
||||
}
|
||||
|
||||
cleanup:
|
||||
while(NULL != (item = opal_list_remove_first(&nodes))) {
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
for(i=0; i<num_context; i++)
|
||||
OBJ_RELEASE(context[i]);
|
||||
free(context);
|
||||
return rc;
|
||||
}
|
||||
|
||||
static int orte_ras_host_node_insert(opal_list_t *nodes)
|
||||
{
|
||||
return orte_ras_base_node_insert(nodes);
|
||||
}
|
||||
|
||||
static int orte_ras_host_node_query(opal_list_t *nodes)
|
||||
{
|
||||
return orte_ras_base_node_query(nodes);
|
||||
}
|
||||
|
||||
static int orte_ras_host_deallocate(orte_jobid_t jobid)
|
||||
{
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
static int orte_ras_host_finalize(void)
|
||||
{
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
orte_ras_base_module_t orte_ras_host_module = {
|
||||
orte_ras_host_allocate,
|
||||
orte_ras_host_node_insert,
|
||||
orte_ras_host_node_query,
|
||||
orte_ras_host_deallocate,
|
||||
orte_ras_host_finalize
|
||||
};
|
||||
|
||||
@ -1,116 +0,0 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
|
||||
#include "orte_config.h"
|
||||
#include "include/orte_constants.h"
|
||||
#include "mca/base/base.h"
|
||||
#include "mca/base/mca_base_param.h"
|
||||
#include "util/proc_info.h"
|
||||
#include "opal/util/output.h"
|
||||
#include "mca/ras/host/ras_host.h"
|
||||
|
||||
/*
|
||||
* Local functions
|
||||
*/
|
||||
|
||||
static int orte_ras_host_open(void);
|
||||
static int orte_ras_host_close(void);
|
||||
static orte_ras_base_module_t* orte_ras_host_init(int*);
|
||||
|
||||
|
||||
orte_ras_host_component_t mca_ras_host_component = {
|
||||
{
|
||||
/* First, the mca_base_component_t struct containing meta
|
||||
information about the component itself */
|
||||
|
||||
{
|
||||
/* Indicate that we are a iof v1.0.0 component (which also
|
||||
implies a specific MCA version) */
|
||||
|
||||
ORTE_RAS_BASE_VERSION_1_0_0,
|
||||
|
||||
"host", /* MCA component name */
|
||||
ORTE_MAJOR_VERSION, /* MCA component major version */
|
||||
ORTE_MINOR_VERSION, /* MCA component minor version */
|
||||
ORTE_RELEASE_VERSION, /* MCA component release version */
|
||||
orte_ras_host_open, /* component open */
|
||||
orte_ras_host_close /* component close */
|
||||
},
|
||||
|
||||
/* Next the MCA v1.0.0 component meta data */
|
||||
{
|
||||
/* Whether the component is checkpointable or not */
|
||||
false
|
||||
},
|
||||
|
||||
orte_ras_host_init
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
/**
|
||||
* component open/close/init function
|
||||
*/
|
||||
static int orte_ras_host_open(void)
|
||||
{
|
||||
size_t id;
|
||||
|
||||
mca_base_param_reg_int(&mca_ras_host_component.super.ras_version, "debug",
|
||||
"Toggle debug output for Host RAS component",
|
||||
false, false, 1,
|
||||
&mca_ras_host_component.debug);
|
||||
|
||||
mca_base_param_reg_int(&mca_ras_host_component.super.ras_version, "debug",
|
||||
"Selection priority for the Host RAS component",
|
||||
false, false, 1,
|
||||
&mca_ras_host_component.priority);
|
||||
|
||||
/* JMS To be changed post-beta to LAM's C/N command line notation */
|
||||
id = mca_base_param_find("ras_base", NULL, "schedule_policy");
|
||||
if (0 > id) {
|
||||
id = mca_base_param_reg_string_name("ras_base", "schedule_policy",
|
||||
"Scheduling Policy for RAS. [slot | node]",
|
||||
false, false, "slot",
|
||||
&mca_ras_host_component.schedule_policy);
|
||||
}
|
||||
else {
|
||||
mca_base_param_lookup_string(id, &mca_ras_host_component.schedule_policy);
|
||||
}
|
||||
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
static orte_ras_base_module_t *orte_ras_host_init(int* priority)
|
||||
{
|
||||
*priority = mca_ras_host_component.priority;
|
||||
return &orte_ras_host_module;
|
||||
}
|
||||
|
||||
/**
|
||||
* Close all subsystems.
|
||||
*/
|
||||
|
||||
static int orte_ras_host_close(void)
|
||||
{
|
||||
if (NULL != mca_ras_host_component.schedule_policy) {
|
||||
free(mca_ras_host_component.schedule_policy);
|
||||
}
|
||||
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
@ -14,36 +14,32 @@
|
||||
# $HEADER$
|
||||
#
|
||||
|
||||
# Use the top-level Makefile.options
|
||||
|
||||
|
||||
|
||||
# Make the output library in this directory, and name it either
|
||||
# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
|
||||
# (for static builds).
|
||||
|
||||
if OMPI_BUILD_ras_host_DSO
|
||||
if OMPI_BUILD_ras_hostfile_DSO
|
||||
component_noinst =
|
||||
component_install = mca_ras_host.la
|
||||
component_install = mca_ras_hostfile.la
|
||||
else
|
||||
component_noinst = libmca_ras_host.la
|
||||
component_noinst = libmca_ras_hostfile.la
|
||||
component_install =
|
||||
endif
|
||||
|
||||
host_SOURCES = \
|
||||
ras_host.c \
|
||||
ras_host.h \
|
||||
ras_host_component.c
|
||||
hostfile_SOURCES = \
|
||||
ras_hostfile.h \
|
||||
ras_hostfile_module.c \
|
||||
ras_hostfile_component.c
|
||||
|
||||
mcacomponentdir = $(libdir)/openmpi
|
||||
mcacomponent_LTLIBRARIES = $(component_install)
|
||||
mca_ras_host_la_SOURCES = $(host_SOURCES)
|
||||
mca_ras_host_la_LIBADD = \
|
||||
mca_ras_hostfile_la_SOURCES = $(hostfile_SOURCES)
|
||||
mca_ras_hostfile_la_LIBADD = \
|
||||
$(top_ompi_builddir)/orte/liborte.la \
|
||||
$(top_ompi_builddir)/opal/libopal.la
|
||||
mca_ras_host_la_LDFLAGS = -module -avoid-version
|
||||
mca_ras_hostfile_la_LDFLAGS = -module -avoid-version
|
||||
|
||||
noinst_LTLIBRARIES = $(component_noinst)
|
||||
libmca_ras_host_la_SOURCES = $(host_SOURCES)
|
||||
libmca_ras_host_la_LIBADD =
|
||||
libmca_ras_host_la_LDFLAGS = -module -avoid-version
|
||||
libmca_ras_hostfile_la_SOURCES = $(hostfile_SOURCES)
|
||||
libmca_ras_hostfile_la_LIBADD =
|
||||
libmca_ras_hostfile_la_LDFLAGS = -module -avoid-version
|
||||
@ -17,5 +17,5 @@
|
||||
|
||||
# Specific to this module
|
||||
|
||||
PARAM_INIT_FILE=ras_host.c
|
||||
PARAM_INIT_FILE=ras_hostfile_component.c
|
||||
PARAM_CONFIG_FILES="Makefile"
|
||||
57
orte/mca/ras/hostfile/ras_hostfile.h
Обычный файл
57
orte/mca/ras/hostfile/ras_hostfile.h
Обычный файл
@ -0,0 +1,57 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
/**
|
||||
* @file
|
||||
*
|
||||
* Resource Allocation (hostfile)
|
||||
*/
|
||||
#ifndef ORTE_RAS_HOSTFILE_H
|
||||
#define ORTE_RAS_HOSTFILE_H
|
||||
|
||||
#include "mca/ras/ras.h"
|
||||
#if defined(c_plusplus) || defined(__cplusplus)
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
/**
|
||||
* hostfile-specific RAS component struct
|
||||
*/
|
||||
struct orte_ras_hostfile_component_t {
|
||||
/** Base RAS component */
|
||||
orte_ras_base_component_t super;
|
||||
/** What's the priority of this component */
|
||||
int priority;
|
||||
};
|
||||
/**
|
||||
* Convenience typedef
|
||||
*/
|
||||
typedef struct orte_ras_hostfile_component_t orte_ras_hostfile_component_t;
|
||||
|
||||
/**
|
||||
* Component export structure
|
||||
*/
|
||||
OMPI_COMP_EXPORT extern orte_ras_hostfile_component_t mca_ras_hostfile_component;
|
||||
|
||||
/**
|
||||
* Module init function
|
||||
*/
|
||||
orte_ras_base_module_t *orte_ras_hostfile_init(int* priority);
|
||||
|
||||
#if defined(c_plusplus) || defined(__cplusplus)
|
||||
}
|
||||
#endif
|
||||
|
||||
#endif
|
||||
75
orte/mca/ras/hostfile/ras_hostfile_component.c
Обычный файл
75
orte/mca/ras/hostfile/ras_hostfile_component.c
Обычный файл
@ -0,0 +1,75 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
|
||||
#include "orte_config.h"
|
||||
|
||||
#include "opal/util/output.h"
|
||||
#include "opal/mca/base/base.h"
|
||||
#include "opal/mca/base/mca_base_param.h"
|
||||
#include "orte/include/orte_constants.h"
|
||||
#include "orte/util/proc_info.h"
|
||||
#include "orte/mca/ras/hostfile/ras_hostfile.h"
|
||||
|
||||
|
||||
/*
|
||||
* Local functions
|
||||
*/
|
||||
static int orte_ras_hostfile_open(void);
|
||||
|
||||
|
||||
orte_ras_hostfile_component_t mca_ras_hostfile_component = {
|
||||
{
|
||||
/* First, the mca_base_component_t struct containing meta
|
||||
information about the component itself */
|
||||
|
||||
{
|
||||
/* Indicate that we are a iof v1.0.0 component (which also
|
||||
implies a specific MCA version) */
|
||||
|
||||
ORTE_RAS_BASE_VERSION_1_0_0,
|
||||
|
||||
"hostfile", /* MCA component name */
|
||||
ORTE_MAJOR_VERSION, /* MCA component major version */
|
||||
ORTE_MINOR_VERSION, /* MCA component minor version */
|
||||
ORTE_RELEASE_VERSION, /* MCA component release version */
|
||||
orte_ras_hostfile_open, /* component open */
|
||||
NULL
|
||||
},
|
||||
|
||||
/* Next the MCA v1.0.0 component meta data */
|
||||
{
|
||||
/* Whether the component is checkpointable or not */
|
||||
false
|
||||
},
|
||||
|
||||
orte_ras_hostfile_init
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
/**
|
||||
* component open function
|
||||
*/
|
||||
static int orte_ras_hostfile_open(void)
|
||||
{
|
||||
mca_base_param_reg_int(&mca_ras_hostfile_component.super.ras_version,
|
||||
"priority",
|
||||
"Selection priority for the hostfile RAS component",
|
||||
false, false, 10,
|
||||
&mca_ras_hostfile_component.priority);
|
||||
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
130
orte/mca/ras/hostfile/ras_hostfile_module.c
Обычный файл
130
orte/mca/ras/hostfile/ras_hostfile_module.c
Обычный файл
@ -0,0 +1,130 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
|
||||
#include "orte_config.h"
|
||||
|
||||
#include "opal/util/output.h"
|
||||
#include "opal/util/argv.h"
|
||||
#include "orte/include/orte_constants.h"
|
||||
#include "orte/include/orte_types.h"
|
||||
#include "orte/mca/ras/base/base.h"
|
||||
#include "orte/mca/ras/base/ras_base_node.h"
|
||||
#include "orte/mca/rmgr/base/base.h"
|
||||
#include "orte/mca/ras/base/ras_base_node.h"
|
||||
#include "orte/mca/errmgr/errmgr.h"
|
||||
|
||||
#include "orte/mca/ras/hostfile/ras_hostfile.h"
|
||||
|
||||
|
||||
/*
|
||||
* Local functions
|
||||
*/
|
||||
static int orte_ras_hostfile_allocate(orte_jobid_t jobid);
|
||||
static int orte_ras_hostfile_deallocate(orte_jobid_t jobid);
|
||||
static int orte_ras_hostfile_finalize(void);
|
||||
|
||||
|
||||
/*
|
||||
* Local variables
|
||||
*/
|
||||
orte_ras_base_module_t orte_ras_hostfile_module = {
|
||||
orte_ras_hostfile_allocate,
|
||||
orte_ras_base_node_insert,
|
||||
orte_ras_base_node_query,
|
||||
orte_ras_hostfile_deallocate,
|
||||
orte_ras_hostfile_finalize
|
||||
};
|
||||
|
||||
|
||||
orte_ras_base_module_t *orte_ras_hostfile_init(int* priority)
|
||||
{
|
||||
*priority = mca_ras_hostfile_component.priority;
|
||||
return &orte_ras_hostfile_module;
|
||||
}
|
||||
|
||||
|
||||
/*
|
||||
* THIS FUNCTION NEEDS TO CHANGE POST-1.0.
|
||||
*
|
||||
* After 1.0, this function, and the rds/hostfile need to change to
|
||||
* clean up properly. They're not "broken" right now, so we're not
|
||||
* fixing them. But they're implemented wrong, so they should be
|
||||
* adapted to the model that they're supposed to implement, not the
|
||||
* workarounds that they currently have. The end result will be much,
|
||||
* much cleaner.
|
||||
*
|
||||
* Specifically, the rds/hostfile currently puts all of its nodes on
|
||||
* the resource segment *and* the node segment. It should not. It
|
||||
* should only put its nodes on the resource segment, appropriately
|
||||
* tagged that they came from a hostfile. The ras/hostfile should
|
||||
* then examine the resources segment and pull out all nodes that came
|
||||
* from a hostfile and put them on the nodes segment.
|
||||
*/
|
||||
static int orte_ras_hostfile_allocate(orte_jobid_t jobid)
|
||||
{
|
||||
opal_list_t nodes;
|
||||
opal_list_item_t* item;
|
||||
int rc;
|
||||
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
|
||||
/* Query for all nodes in the node segment that have been
|
||||
allocated to this job */
|
||||
if (ORTE_SUCCESS != (rc = orte_ras_base_node_query_alloc(&nodes, jobid))) {
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
/* If there are nodes allocated, then query for *all* nodes */
|
||||
if (opal_list_is_empty(&nodes)) {
|
||||
if (ORTE_SUCCESS != (rc = orte_ras_base_node_query(&nodes))) {
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
/* If there are any nodes at all, allocate them all to this job */
|
||||
if (!opal_list_is_empty(&nodes)) {
|
||||
rc = orte_ras_base_allocate_nodes(jobid, &nodes);
|
||||
goto cleanup;
|
||||
}
|
||||
}
|
||||
|
||||
cleanup:
|
||||
while (NULL != (item = opal_list_remove_first(&nodes))) {
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
|
||||
return rc;
|
||||
}
|
||||
|
||||
|
||||
static int orte_ras_hostfile_deallocate(orte_jobid_t jobid)
|
||||
{
|
||||
/* Nothing to do */
|
||||
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"ras:hostfile:deallocate: success (nothing to do)");
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
static int orte_ras_hostfile_finalize(void)
|
||||
{
|
||||
/* Nothing to do */
|
||||
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"ras:hostfile:finalize: success (nothing to do)");
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
45
orte/mca/ras/localhost/Makefile.am
Обычный файл
45
orte/mca/ras/localhost/Makefile.am
Обычный файл
@ -0,0 +1,45 @@
|
||||
#
|
||||
# Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
# All rights reserved.
|
||||
# Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
# All rights reserved.
|
||||
# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
# University of Stuttgart. All rights reserved.
|
||||
# Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
# All rights reserved.
|
||||
# $COPYRIGHT$
|
||||
#
|
||||
# Additional copyrights may follow
|
||||
#
|
||||
# $HEADER$
|
||||
#
|
||||
|
||||
# Make the output library in this directory, and name it either
|
||||
# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
|
||||
# (for static builds).
|
||||
|
||||
if OMPI_BUILD_ras_localhost_DSO
|
||||
component_noinst =
|
||||
component_install = mca_ras_localhost.la
|
||||
else
|
||||
component_noinst = libmca_ras_localhost.la
|
||||
component_install =
|
||||
endif
|
||||
|
||||
localhost_SOURCES = \
|
||||
ras_localhost.h \
|
||||
ras_localhost_module.c \
|
||||
ras_localhost_component.c
|
||||
|
||||
mcacomponentdir = $(libdir)/openmpi
|
||||
mcacomponent_LTLIBRARIES = $(component_install)
|
||||
mca_ras_localhost_la_SOURCES = $(localhost_SOURCES)
|
||||
mca_ras_localhost_la_LIBADD = \
|
||||
$(top_ompi_builddir)/orte/liborte.la \
|
||||
$(top_ompi_builddir)/opal/libopal.la
|
||||
mca_ras_localhost_la_LDFLAGS = -module -avoid-version
|
||||
|
||||
noinst_LTLIBRARIES = $(component_noinst)
|
||||
libmca_ras_localhost_la_SOURCES = $(localhost_SOURCES)
|
||||
libmca_ras_localhost_la_LIBADD =
|
||||
libmca_ras_localhost_la_LDFLAGS = -module -avoid-version
|
||||
21
orte/mca/ras/localhost/configure.params
Обычный файл
21
orte/mca/ras/localhost/configure.params
Обычный файл
@ -0,0 +1,21 @@
|
||||
# -*- shell-script -*-
|
||||
#
|
||||
# Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
# All rights reserved.
|
||||
# Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
# All rights reserved.
|
||||
# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
# University of Stuttgart. All rights reserved.
|
||||
# Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
# All rights reserved.
|
||||
# $COPYRIGHT$
|
||||
#
|
||||
# Additional copyrights may follow
|
||||
#
|
||||
# $HEADER$
|
||||
#
|
||||
|
||||
# Specific to this module
|
||||
|
||||
PARAM_INIT_FILE=ras_localhost_component.c
|
||||
PARAM_CONFIG_FILES="Makefile"
|
||||
@ -18,30 +18,38 @@
|
||||
*
|
||||
* Resource Allocation (Host)
|
||||
*/
|
||||
#ifndef ORTE_RAS_HOST_H
|
||||
#define ORTE_RAS_HOST_H
|
||||
#ifndef ORTE_RAS_LOCALHOST_H
|
||||
#define ORTE_RAS_LOCALHOST_H
|
||||
|
||||
#include "mca/ras/ras.h"
|
||||
#if defined(c_plusplus) || defined(__cplusplus)
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
/**
|
||||
* Localhost-specific RAS component struct
|
||||
*/
|
||||
struct orte_ras_localhost_component_t {
|
||||
/** Base RAS component */
|
||||
orte_ras_base_component_t super;
|
||||
/** What's the priority of this component */
|
||||
int priority;
|
||||
};
|
||||
/**
|
||||
* Convenience typedef
|
||||
*/
|
||||
typedef struct orte_ras_localhost_component_t orte_ras_localhost_component_t;
|
||||
|
||||
/**
|
||||
* RAS Component
|
||||
*/
|
||||
struct orte_ras_host_component_t {
|
||||
orte_ras_base_component_t super;
|
||||
int debug;
|
||||
int priority;
|
||||
char *schedule_policy;
|
||||
};
|
||||
typedef struct orte_ras_host_component_t orte_ras_host_component_t;
|
||||
|
||||
OMPI_COMP_EXPORT extern orte_ras_host_component_t mca_ras_host_component;
|
||||
OMPI_COMP_EXPORT extern orte_ras_base_module_t orte_ras_host_module;
|
||||
|
||||
|
||||
/**
|
||||
* Component export structure
|
||||
*/
|
||||
OMPI_COMP_EXPORT extern orte_ras_localhost_component_t mca_ras_localhost_component;
|
||||
|
||||
/**
|
||||
* Module init function
|
||||
*/
|
||||
orte_ras_base_module_t *orte_ras_localhost_init(int* priority);
|
||||
|
||||
#if defined(c_plusplus) || defined(__cplusplus)
|
||||
}
|
||||
#endif
|
||||
75
orte/mca/ras/localhost/ras_localhost_component.c
Обычный файл
75
orte/mca/ras/localhost/ras_localhost_component.c
Обычный файл
@ -0,0 +1,75 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
|
||||
#include "orte_config.h"
|
||||
|
||||
#include "opal/util/output.h"
|
||||
#include "opal/mca/base/base.h"
|
||||
#include "opal/mca/base/mca_base_param.h"
|
||||
#include "orte/include/orte_constants.h"
|
||||
#include "orte/util/proc_info.h"
|
||||
#include "orte/mca/ras/localhost/ras_localhost.h"
|
||||
|
||||
|
||||
/*
|
||||
* Local functions
|
||||
*/
|
||||
static int orte_ras_localhost_open(void);
|
||||
|
||||
|
||||
orte_ras_localhost_component_t mca_ras_localhost_component = {
|
||||
{
|
||||
/* First, the mca_base_component_t struct containing meta
|
||||
information about the component itself */
|
||||
|
||||
{
|
||||
/* Indicate that we are a iof v1.0.0 component (which also
|
||||
implies a specific MCA version) */
|
||||
|
||||
ORTE_RAS_BASE_VERSION_1_0_0,
|
||||
|
||||
"localhost", /* MCA component name */
|
||||
ORTE_MAJOR_VERSION, /* MCA component major version */
|
||||
ORTE_MINOR_VERSION, /* MCA component minor version */
|
||||
ORTE_RELEASE_VERSION, /* MCA component release version */
|
||||
orte_ras_localhost_open, /* component open */
|
||||
NULL
|
||||
},
|
||||
|
||||
/* Next the MCA v1.0.0 component meta data */
|
||||
{
|
||||
/* Whether the component is checkpointable or not */
|
||||
false
|
||||
},
|
||||
|
||||
orte_ras_localhost_init
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
/**
|
||||
* component open function
|
||||
*/
|
||||
static int orte_ras_localhost_open(void)
|
||||
{
|
||||
mca_base_param_reg_int(&mca_ras_localhost_component.super.ras_version,
|
||||
"priority",
|
||||
"Selection priority for the localhost RAS component",
|
||||
false, false, 0,
|
||||
&mca_ras_localhost_component.priority);
|
||||
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
136
orte/mca/ras/localhost/ras_localhost_module.c
Обычный файл
136
orte/mca/ras/localhost/ras_localhost_module.c
Обычный файл
@ -0,0 +1,136 @@
|
||||
/*
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
* Additional copyrights may follow
|
||||
*
|
||||
* $HEADER$
|
||||
*/
|
||||
|
||||
#include "orte_config.h"
|
||||
|
||||
#include "opal/class/opal_list.h"
|
||||
#include "opal/util/output.h"
|
||||
#include "orte/include/orte_constants.h"
|
||||
#include "orte/include/orte_types.h"
|
||||
#include "orte/mca/ras/base/base.h"
|
||||
#include "orte/mca/ras/base/ras_base_node.h"
|
||||
#include "orte/mca/rmgr/base/base.h"
|
||||
#include "orte/mca/ras/base/ras_base_node.h"
|
||||
#include "orte/mca/errmgr/errmgr.h"
|
||||
|
||||
#include "orte/mca/ras/localhost/ras_localhost.h"
|
||||
|
||||
|
||||
/*
|
||||
* Local functions
|
||||
*/
|
||||
static int orte_ras_localhost_allocate(orte_jobid_t jobid);
|
||||
static int orte_ras_localhost_deallocate(orte_jobid_t jobid);
|
||||
static int orte_ras_localhost_finalize(void);
|
||||
|
||||
|
||||
/*
|
||||
* Local variables
|
||||
*/
|
||||
orte_ras_base_module_t orte_ras_localhost_module = {
|
||||
orte_ras_localhost_allocate,
|
||||
orte_ras_base_node_insert,
|
||||
orte_ras_base_node_query,
|
||||
orte_ras_localhost_deallocate,
|
||||
orte_ras_localhost_finalize
|
||||
};
|
||||
|
||||
|
||||
orte_ras_base_module_t *orte_ras_localhost_init(int* priority)
|
||||
{
|
||||
*priority = mca_ras_localhost_component.priority;
|
||||
return &orte_ras_localhost_module;
|
||||
}
|
||||
|
||||
|
||||
static int orte_ras_localhost_allocate(orte_jobid_t jobid)
|
||||
{
|
||||
bool empty;
|
||||
int ret;
|
||||
opal_list_t nodes;
|
||||
orte_ras_node_t *node;
|
||||
opal_list_item_t *item;
|
||||
|
||||
/* If the node segment is not empty, do nothing */
|
||||
|
||||
if (ORTE_SUCCESS != (ret = orte_ras_base_node_segment_empty(&empty))) {
|
||||
ORTE_ERROR_LOG(ret);
|
||||
return ret;
|
||||
}
|
||||
if (!empty) {
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:ras:localhost: node segment not empty; not doing anything");
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"orte:ras:localhost: node segment empty; adding \"localhost\"");
|
||||
|
||||
/* Ok, the node segment is empty -- so add a localhost node */
|
||||
|
||||
node = OBJ_NEW(orte_ras_node_t);
|
||||
if (NULL == node) {
|
||||
return ORTE_ERR_OUT_OF_RESOURCE;
|
||||
}
|
||||
node->node_name = strdup("localhost");
|
||||
node->node_arch = NULL;
|
||||
node->node_state = ORTE_NODE_STATE_UP;
|
||||
/* JMS: this should not be hard-wired to 0, but there's no
|
||||
other value to put it to [yet]... */
|
||||
node->node_cellid = 0;
|
||||
node->node_slots_inuse = 0;
|
||||
node->node_slots_max = 0;
|
||||
node->node_slots = 1;
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
opal_list_append(&nodes, &node->super);
|
||||
|
||||
/* Put it on the segment and allocate it */
|
||||
|
||||
if (ORTE_SUCCESS !=
|
||||
(ret = orte_ras_base_node_insert(&nodes)) ||
|
||||
ORTE_SUCCESS !=
|
||||
(ret = orte_ras_base_allocate_nodes(jobid, &nodes))) {
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
cleanup:
|
||||
item = opal_list_remove_first(&nodes);
|
||||
OBJ_RELEASE(item);
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
|
||||
/* All done */
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
|
||||
static int orte_ras_localhost_deallocate(orte_jobid_t jobid)
|
||||
{
|
||||
/* Nothing to do */
|
||||
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"ras:localhost:deallocate: success (nothing to do)");
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
static int orte_ras_localhost_finalize(void)
|
||||
{
|
||||
/* Nothing to do */
|
||||
|
||||
opal_output(orte_ras_base.ras_output,
|
||||
"ras:localhost:finalize: success (nothing to do)");
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
@ -55,7 +55,6 @@ orte_ras_base_module_t orte_ras_slurm_module = {
|
||||
finalize
|
||||
};
|
||||
|
||||
|
||||
/**
|
||||
* Discover available (pre-allocated) nodes. Allocate the
|
||||
* requested number of nodes/process slots to the job.
|
||||
@ -81,7 +80,8 @@ static int allocate(orte_jobid_t jobid)
|
||||
"ras:slurm:allocate: discover failed!");
|
||||
return ret;
|
||||
}
|
||||
ret = orte_ras_base_allocate_nodes_by_slot(jobid, &nodes);
|
||||
|
||||
ret = orte_ras_base_allocate_nodes(jobid, &nodes);
|
||||
|
||||
while (NULL != (item = opal_list_remove_first(&nodes))) {
|
||||
OBJ_RELEASE(item);
|
||||
|
||||
@ -84,7 +84,7 @@ static int allocate(orte_jobid_t jobid)
|
||||
tm_finalize();
|
||||
return ret;
|
||||
}
|
||||
ret = orte_ras_base_allocate_nodes_by_slot(jobid, &nodes);
|
||||
ret = orte_ras_base_allocate_nodes(jobid, &nodes);
|
||||
|
||||
while (NULL != (item = opal_list_remove_first(&nodes))) {
|
||||
OBJ_RELEASE(item);
|
||||
|
||||
@ -71,7 +71,7 @@ static int allocate(orte_jobid_t jobid)
|
||||
"ras:xgrid:allocate: discover failed!");
|
||||
return ret;
|
||||
}
|
||||
ret = orte_ras_base_allocate_nodes_by_node(jobid, &nodes);
|
||||
ret = orte_ras_base_allocate_nodes(jobid, &nodes);
|
||||
|
||||
while (NULL != (item = opal_list_remove_first(&nodes))) {
|
||||
OBJ_RELEASE(item);
|
||||
|
||||
@ -20,20 +20,21 @@
|
||||
#include <errno.h>
|
||||
#include <string.h>
|
||||
|
||||
#include "include/orte_constants.h"
|
||||
#include "opal/class/opal_list.h"
|
||||
#include "opal/util/output.h"
|
||||
#include "opal/util/argv.h"
|
||||
#include "util/sys_info.h"
|
||||
#include "mca/mca.h"
|
||||
#include "mca/base/base.h"
|
||||
#include "mca/ns/ns.h"
|
||||
#include "mca/errmgr/errmgr.h"
|
||||
#include "mca/ras/ras.h"
|
||||
#include "mca/rds/rds.h"
|
||||
#include "mca/rds/base/base.h"
|
||||
#include "mca/rds/hostfile/rds_hostfile.h"
|
||||
#include "mca/rds/hostfile/rds_hostfile_lex.h"
|
||||
#include "opal/mca/mca.h"
|
||||
#include "opal/mca/base/base.h"
|
||||
#include "orte/include/orte_constants.h"
|
||||
#include "orte/util/sys_info.h"
|
||||
#include "orte/mca/ns/ns.h"
|
||||
#include "orte/mca/errmgr/errmgr.h"
|
||||
#include "orte/mca/ras/ras.h"
|
||||
#include "orte/mca/ras/base/ras_base_node.h"
|
||||
#include "orte/mca/rds/rds.h"
|
||||
#include "orte/mca/rds/base/base.h"
|
||||
#include "orte/mca/rds/hostfile/rds_hostfile.h"
|
||||
#include "orte/mca/rds/hostfile/rds_hostfile_lex.h"
|
||||
|
||||
#include "runtime/runtime_types.h"
|
||||
|
||||
@ -334,7 +335,7 @@ static int orte_rds_hostfile_query(void)
|
||||
OBJ_CONSTRUCT(&existing, opal_list_t);
|
||||
OBJ_CONSTRUCT(&updates, opal_list_t);
|
||||
OBJ_CONSTRUCT(&rds_updates, opal_list_t);
|
||||
rc = orte_ras.node_query(&existing);
|
||||
rc = orte_ras_base_node_query(&existing);
|
||||
if(ORTE_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
@ -418,7 +419,7 @@ static int orte_rds_hostfile_query(void)
|
||||
* resources listed in the hostfile have been
|
||||
* already allocated for our use.
|
||||
*/
|
||||
rc = orte_ras.node_insert(&updates);
|
||||
rc = orte_ras_base_node_insert(&updates);
|
||||
if (ORTE_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
@ -195,6 +195,94 @@ orte_rmaps_lookup_node(opal_list_t* nodes, char* node_name, orte_rmaps_base_proc
|
||||
return node;
|
||||
}
|
||||
|
||||
/**
|
||||
* Obtain the mapping for this job, and the list of nodes confined to that mapping.
|
||||
*
|
||||
* Use this instead of orte_ras_base_node_query when past the RMAPS framework
|
||||
* since components like the PLS are only conserned with those nodes that they
|
||||
* been mapped on, not all of the nodes allocated to their job. In the case
|
||||
* where we are allocated 10 nodes from the RAS, but only map to 2 of them
|
||||
* then we don't try to launch orteds on all 10 nodes, just the 2 mapped.
|
||||
*/
|
||||
int orte_rmaps_base_mapped_node_query(opal_list_t* mapping_list, opal_list_t* nodes_alloc, orte_jobid_t jobid) {
|
||||
opal_list_t nodes;
|
||||
opal_list_item_t *item_a, *item_m, *item_n;
|
||||
int num_mapping = 0;
|
||||
int rc = ORTE_SUCCESS;
|
||||
bool matched = false;
|
||||
|
||||
/* get all nodes allocated to this job */
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
rc = orte_ras_base_node_query_alloc(&nodes, jobid);
|
||||
if (ORTE_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
/* get the mapping for this job */
|
||||
rc = orte_rmaps_base_get_map(jobid, mapping_list);
|
||||
if (ORTE_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
num_mapping = opal_list_get_size(mapping_list);
|
||||
|
||||
/* Create a list of nodes that are in the mapping */
|
||||
for( item_m = opal_list_get_first(mapping_list);
|
||||
item_m != opal_list_get_end(mapping_list);
|
||||
item_m = opal_list_get_next(item_m)) {
|
||||
orte_rmaps_base_map_t* map = (orte_rmaps_base_map_t*)item_m;
|
||||
|
||||
/* Iterate over all the nodes mapped and check them against the
|
||||
* allocated node list */
|
||||
for( item_n = opal_list_get_first(&(map->nodes));
|
||||
item_n != opal_list_get_end(&(map->nodes));
|
||||
item_n = opal_list_get_next(item_n)) {
|
||||
matched = false;
|
||||
|
||||
/* If this node is in the list already, skip it */
|
||||
for( item_a = opal_list_get_first(nodes_alloc);
|
||||
item_a != opal_list_get_end(nodes_alloc);
|
||||
item_a = opal_list_get_next(item_a)) {
|
||||
if( 0 == strcmp( ((orte_ras_node_t*) item_a)->node_name,
|
||||
((orte_rmaps_base_node_t*) item_n)->node_name) ) {
|
||||
matched = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if(matched){
|
||||
continue;
|
||||
}
|
||||
|
||||
/* Otherwise
|
||||
* - Find it in the node list from the node segment,
|
||||
* - Add it to the allocated list of nodes
|
||||
*/
|
||||
matched = false;
|
||||
for( item_a = opal_list_get_first(&nodes);
|
||||
item_a != opal_list_get_end(&nodes);
|
||||
item_a = opal_list_get_next(item_a)) {
|
||||
if( 0 == strcmp( ((orte_ras_node_t*) item_a)->node_name,
|
||||
((orte_rmaps_base_node_t*) item_n)->node_name) ) {
|
||||
matched = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if(!matched) {
|
||||
printf("Unable to find the matched node in the allocation. This should never happen\n");
|
||||
return ORTE_ERROR;
|
||||
}
|
||||
opal_list_remove_item(&nodes, item_a);
|
||||
opal_list_append(nodes_alloc, item_a);
|
||||
}
|
||||
}
|
||||
|
||||
cleanup:
|
||||
while (NULL != (item_a = opal_list_remove_first(&nodes))) {
|
||||
OBJ_RELEASE(item_a);
|
||||
}
|
||||
|
||||
return rc;
|
||||
}
|
||||
|
||||
/**
|
||||
* Query the process mapping from the registry.
|
||||
|
||||
@ -97,7 +97,7 @@ typedef struct orte_rmaps_base_map_t orte_rmaps_base_map_t;
|
||||
|
||||
OBJ_CLASS_DECLARATION(orte_rmaps_base_map_t);
|
||||
|
||||
|
||||
int orte_rmaps_base_mapped_node_query(opal_list_t* mapping_list, opal_list_t* nodes_alloc, orte_jobid_t jobid);
|
||||
int orte_rmaps_base_get_map(orte_jobid_t, opal_list_t* mapping);
|
||||
int orte_rmaps_base_set_map(orte_jobid_t, opal_list_t* mapping);
|
||||
int orte_rmaps_base_get_node_map(orte_cellid_t, orte_jobid_t, const char*, opal_list_t* mapping);
|
||||
|
||||
@ -61,6 +61,7 @@ int orte_rmaps_base_open(void)
|
||||
orte_rmaps_base_module_t *module;
|
||||
int param, priority, value;
|
||||
orte_rmaps_base_cmp_t *cmp;
|
||||
char *policy;
|
||||
|
||||
/* Debugging / verbose output */
|
||||
|
||||
@ -73,6 +74,15 @@ int orte_rmaps_base_open(void)
|
||||
orte_rmaps_base.rmaps_output = -1;
|
||||
}
|
||||
|
||||
/* Are we scheduling by node or by slot? */
|
||||
|
||||
param = mca_base_param_reg_string_name("rmaps_base", "schedule_policy",
|
||||
"Scheduling Policy for RMAPS. [slot | node]",
|
||||
false, false, "slot", &policy);
|
||||
if (0 == strcmp(policy, "node")) {
|
||||
mca_base_param_set_string(param, "node");
|
||||
}
|
||||
|
||||
/* Open up all the components that we can find */
|
||||
|
||||
if (ORTE_SUCCESS !=
|
||||
|
||||
@ -21,4 +21,20 @@
|
||||
|
||||
Either request fewer slots for your application, or make more slots available
|
||||
for use.
|
||||
[orte-rmaps-rr:no-mapped-node]
|
||||
There are no allocated resources for the application
|
||||
%s
|
||||
that match the requested mapping:
|
||||
%s
|
||||
|
||||
Verify that you have mapped the allocated resources properly using the
|
||||
--host specification.
|
||||
[orte-rmaps-rr:not-all-mapped-alloc]
|
||||
Some of the requested hosts are not included in the current allocation for the
|
||||
application:
|
||||
%s
|
||||
The requested hosts were:
|
||||
%s
|
||||
|
||||
Verify that you have mapped the allocated resources properly using the
|
||||
--host specification.
|
||||
|
||||
@ -22,6 +22,7 @@
|
||||
#include "include/orte_types.h"
|
||||
#include "opal/util/output.h"
|
||||
#include "opal/util/show_help.h"
|
||||
#include "opal/util/argv.h"
|
||||
#include "mca/ns/ns.h"
|
||||
#include "mca/gpr/gpr.h"
|
||||
#include "mca/rmaps/base/base.h"
|
||||
@ -36,9 +37,108 @@
|
||||
*/
|
||||
static opal_list_item_t *cur_node_item = NULL;
|
||||
|
||||
/*
|
||||
* A sanity check to ensure that all of the requested nodes are actually
|
||||
* allocated to this application.
|
||||
*/
|
||||
static bool are_all_mapped_valid(char **mapping,
|
||||
int num_mapped,
|
||||
opal_list_t* nodes)
|
||||
{
|
||||
opal_list_item_t *item;
|
||||
int i;
|
||||
bool matched;
|
||||
|
||||
for (i = 0; i < num_mapped; ++i) {
|
||||
matched = false;
|
||||
|
||||
for(item = opal_list_get_first(nodes);
|
||||
item != opal_list_get_end(nodes);
|
||||
item = opal_list_get_next(item) ) {
|
||||
if( 0 == strcmp( ((orte_ras_node_t*) item)->node_name, mapping[i]) ) {
|
||||
matched = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
/* If we find one requested resource that is not allocated,
|
||||
* then return an error */
|
||||
if(!matched) {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
/*
|
||||
* If the node in question is in the current mapping.
|
||||
*/
|
||||
static bool is_mapped(opal_list_item_t *item,
|
||||
char **mapping,
|
||||
int num_mapped,
|
||||
opal_list_t* nodes)
|
||||
{
|
||||
int i;
|
||||
|
||||
for ( i = 0; i < num_mapped; ++i) {
|
||||
if ( 0 == strcmp( ((orte_ras_node_t*) item)->node_name, mapping[i])){
|
||||
return true;
|
||||
}
|
||||
}
|
||||
|
||||
return false;
|
||||
}
|
||||
|
||||
/*
|
||||
* Return a point to the next node allocated, included in the mapping.
|
||||
*/
|
||||
static opal_list_item_t* get_next_mapped(opal_list_item_t *node_item,
|
||||
char **mapping,
|
||||
int num_mapped,
|
||||
opal_list_t* nodes)
|
||||
{
|
||||
opal_list_item_t *item;
|
||||
|
||||
/* Wrap around to beginning if we are at the end of the list */
|
||||
if (opal_list_get_end(nodes) == opal_list_get_next(node_item)) {
|
||||
item = opal_list_get_first(nodes);
|
||||
}
|
||||
else {
|
||||
item = opal_list_get_next(node_item);
|
||||
}
|
||||
|
||||
do {
|
||||
/* See if current node is in the mapping and contains slots */
|
||||
if( is_mapped(item, mapping, num_mapped, nodes) ) {
|
||||
return item;
|
||||
}
|
||||
|
||||
/*
|
||||
* We just rechecked the current item and concluded that
|
||||
* it wasn't in the list, thus the list contains no matches
|
||||
* in this mapping. Return an error.
|
||||
*/
|
||||
if(node_item == item){
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/* Access next item in Round Robin Manner */
|
||||
if (opal_list_get_end(nodes) == opal_list_get_next(item)) {
|
||||
item = opal_list_get_first(nodes);
|
||||
}
|
||||
else {
|
||||
item = opal_list_get_next(item);
|
||||
}
|
||||
|
||||
} while( true );
|
||||
}
|
||||
|
||||
static int claim_slot(orte_rmaps_base_map_t *map,
|
||||
orte_ras_node_t *current_node,
|
||||
orte_jobid_t jobid, orte_vpid_t vpid, int proc_index)
|
||||
orte_jobid_t jobid,
|
||||
orte_vpid_t vpid,
|
||||
int proc_index)
|
||||
{
|
||||
orte_rmaps_base_proc_t *proc;
|
||||
orte_process_name_t *proc_name;
|
||||
@ -50,8 +150,9 @@ static int claim_slot(orte_rmaps_base_map_t *map,
|
||||
if (NULL == rmaps_node) {
|
||||
return ORTE_ERR_OUT_OF_RESOURCE;
|
||||
}
|
||||
|
||||
rmaps_node->node_name = strdup(current_node->node_name);
|
||||
|
||||
|
||||
proc = OBJ_NEW(orte_rmaps_base_proc_t);
|
||||
if (NULL == proc) {
|
||||
OBJ_RELEASE(rmaps_node);
|
||||
@ -77,9 +178,9 @@ static int claim_slot(orte_rmaps_base_map_t *map,
|
||||
/* Save this node on the map */
|
||||
opal_list_append(&map->nodes, &rmaps_node->super);
|
||||
|
||||
/* Decrease the number of slots available for allocation
|
||||
on this node */
|
||||
--current_node->node_slots_alloc;
|
||||
/* Be sure to demarcate this slot claim for the node */
|
||||
current_node->node_slots_inuse++;
|
||||
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
@ -87,11 +188,6 @@ static int claim_slot(orte_rmaps_base_map_t *map,
|
||||
/*
|
||||
* Create a default mapping for the application, scheduling round
|
||||
* robin by node.
|
||||
*
|
||||
* NOTE: This function assumes that the allocator has already setup
|
||||
* the list of nodes such that the sum of the node_slots_alloc fields
|
||||
* from all entries will be the total number of processes in all the
|
||||
* apps.
|
||||
*/
|
||||
static int map_app_by_node(
|
||||
orte_app_context_t* app,
|
||||
@ -99,124 +195,9 @@ static int map_app_by_node(
|
||||
orte_jobid_t jobid,
|
||||
orte_vpid_t vpid_start,
|
||||
int rank,
|
||||
opal_list_t* nodes)
|
||||
{
|
||||
int rc;
|
||||
size_t num_alloc = 0;
|
||||
size_t proc_index = 0;
|
||||
opal_list_item_t *start, *next;
|
||||
orte_ras_node_t *node;
|
||||
bool did_alloc;
|
||||
|
||||
/* Note that cur_node_item already points to the Right place in
|
||||
the node list to start looking (i.e., if this is the first time
|
||||
through, it'll point to the first item. If this is not the
|
||||
first time through -- i.e., we have multiple app contexts --
|
||||
it'll point to where we left off last time.). If we're at the
|
||||
end, bounce back to the front (as would happen in the loop
|
||||
below)
|
||||
|
||||
But do a bozo check to ensure that we don't have a empty node
|
||||
list. */
|
||||
if (0 == opal_list_get_size(nodes)) {
|
||||
return ORTE_ERR_TEMP_OUT_OF_RESOURCE;
|
||||
} else if (opal_list_get_end(nodes) == cur_node_item) {
|
||||
cur_node_item = opal_list_get_first(nodes);
|
||||
}
|
||||
start = cur_node_item;
|
||||
|
||||
/* This loop continues until all procs have been mapped or we run
|
||||
out of resources. There are two definitions of "run out of
|
||||
resources":
|
||||
|
||||
1. All nodes have node_slots processes mapped to them
|
||||
2. All nodes have node_slots_max processes mapped to them
|
||||
|
||||
We first map until condition #1 is met. If there are still
|
||||
processes that haven't been mapped yet, then we continue until
|
||||
condition #2 is met. If we still have processes that haven't
|
||||
been mapped yet, then it's an "out of resources" error. */
|
||||
did_alloc = false;
|
||||
while (num_alloc < app->num_procs) {
|
||||
node = (orte_ras_node_t*) cur_node_item;
|
||||
next = opal_list_get_next(cur_node_item);
|
||||
|
||||
/* If we have an available slot on this node, claim it */
|
||||
if (node->node_slots_alloc > 0) {
|
||||
fflush(stdout);
|
||||
rc = claim_slot(map, node, jobid, vpid_start + rank, proc_index);
|
||||
if (ORTE_SUCCESS != rc) {
|
||||
return rc;
|
||||
}
|
||||
|
||||
if (node->node_slots_alloc == 0) {
|
||||
opal_list_remove_item(nodes, (opal_list_item_t*)node);
|
||||
OBJ_RELEASE(node);
|
||||
}
|
||||
|
||||
++rank;
|
||||
++proc_index;
|
||||
|
||||
/* Save the fact that we successfully allocated a process
|
||||
to a node in this round */
|
||||
did_alloc = true;
|
||||
|
||||
/* Increase the number of procs allocated and see if we're
|
||||
done */
|
||||
++num_alloc;
|
||||
}
|
||||
|
||||
/* Move on to the next node */
|
||||
|
||||
cur_node_item = next;
|
||||
if (opal_list_get_end(nodes) == cur_node_item) {
|
||||
cur_node_item = opal_list_get_first(nodes);
|
||||
}
|
||||
|
||||
/* Are we done? */
|
||||
if (num_alloc == app->num_procs) {
|
||||
break;
|
||||
}
|
||||
|
||||
/* Double check that the list is not empty */
|
||||
if (opal_list_get_end(nodes) == cur_node_item) {
|
||||
opal_show_help("help-orte-rmaps-rr.txt", "orte-rmaps-rr:alloc-error",
|
||||
true, num_alloc, app->num_procs);
|
||||
return ORTE_ERR_TEMP_OUT_OF_RESOURCE;
|
||||
}
|
||||
|
||||
/* If we looped around without allocating any new processes,
|
||||
then we're full */
|
||||
if (start == cur_node_item) {
|
||||
if (!did_alloc) {
|
||||
opal_show_help("help-orte-rmaps-rr.txt", "orte-rmaps-rr:alloc-error",
|
||||
true, num_alloc, app->num_procs);
|
||||
return ORTE_ERR_TEMP_OUT_OF_RESOURCE;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
map->num_procs = num_alloc;
|
||||
return ORTE_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
/*
|
||||
* Create a default mapping for the application, scheduling one round
|
||||
* robin by slot.
|
||||
*
|
||||
* NOTE: This function assumes that the allocator has already setup
|
||||
* the list of nodes such that the sum of the node_slots_alloc fields
|
||||
* from all entries will be the total number of processes in all the
|
||||
* apps.
|
||||
*/
|
||||
static int map_app_by_slot(
|
||||
orte_app_context_t* app,
|
||||
orte_rmaps_base_map_t* map,
|
||||
orte_jobid_t jobid,
|
||||
orte_vpid_t vpid_start,
|
||||
int rank,
|
||||
opal_list_t* nodes)
|
||||
opal_list_t* nodes,
|
||||
char **mapped_nodes,
|
||||
int num_mapped_nodes)
|
||||
{
|
||||
int rc;
|
||||
size_t num_alloc = 0;
|
||||
@ -224,67 +205,179 @@ static int map_app_by_slot(
|
||||
opal_list_item_t *next;
|
||||
orte_ras_node_t *node;
|
||||
|
||||
/* Note that cur_node_item already points to the Right place in
|
||||
the node list to start looking (i.e., if this is the first time
|
||||
through, it'll point to the first item. If this is not the
|
||||
first time through -- i.e., we have multiple app contexts --
|
||||
it'll point to where we left off last time.). If we're at the
|
||||
end, bounce back to the front (as would happen in the loop
|
||||
below)
|
||||
|
||||
But do a bozo check to ensure that we don't have a empty node
|
||||
list. */
|
||||
if (0 == opal_list_get_size(nodes)) {
|
||||
return ORTE_ERR_TEMP_OUT_OF_RESOURCE;
|
||||
} else if (opal_list_get_end(nodes) == cur_node_item) {
|
||||
cur_node_item = opal_list_get_first(nodes);
|
||||
}
|
||||
|
||||
/* Go through all nodes and take up to node_slots_alloc slots and
|
||||
map it to this job */
|
||||
|
||||
while (opal_list_get_end(nodes) != cur_node_item &&
|
||||
num_alloc < app->num_procs) {
|
||||
/* This loop continues until all procs have been mapped or we run
|
||||
out of resources. We determine that we have "run out of
|
||||
resources" when all nodes have node_slots_max processes mapped to them,
|
||||
thus there are no free slots for a process to be mapped.
|
||||
If we still have processes that haven't been mapped yet, then it's an
|
||||
"out of resources" error. */
|
||||
while (num_alloc < app->num_procs) {
|
||||
node = (orte_ras_node_t*) cur_node_item;
|
||||
next = opal_list_get_next(cur_node_item);
|
||||
|
||||
/* Find the next node we can use before claiming slots, since
|
||||
* we may need to prune the nodes list removing overused nodes */
|
||||
if ( 0 < app->num_map ) {
|
||||
next = get_next_mapped(cur_node_item, mapped_nodes, num_mapped_nodes, nodes);
|
||||
if (NULL == next ) {
|
||||
/* Not allocated anything */
|
||||
opal_show_help("help-orte-rmaps-rr.txt", "orte-rmaps-rr:no-mapped-node",
|
||||
true, app->app, opal_argv_join(mapped_nodes, ','));
|
||||
rc = ORTE_ERR_OUT_OF_RESOURCE;
|
||||
goto cleanup;
|
||||
}
|
||||
}
|
||||
else {
|
||||
if (opal_list_get_end(nodes) == opal_list_get_next(cur_node_item)) {
|
||||
next = opal_list_get_first(nodes);
|
||||
}
|
||||
else {
|
||||
next = opal_list_get_next(cur_node_item);
|
||||
}
|
||||
}
|
||||
|
||||
/* If we have available slots on this node, claim it */
|
||||
while (node->node_slots_alloc > 0 &&
|
||||
num_alloc < app->num_procs) {
|
||||
fflush(stdout);
|
||||
/* Remove this node if it has reached its max number of allocatable slots */
|
||||
if( 0 != node->node_slots_max &&
|
||||
node->node_slots_inuse >= node->node_slots_max) {
|
||||
opal_list_remove_item(nodes, (opal_list_item_t*)node);
|
||||
if(0 >= opal_list_get_size(nodes) ) {
|
||||
/* No more nodes to allocate :( */
|
||||
opal_show_help("help-orte-rmaps-rr.txt", "orte-rmaps-rr:alloc-error",
|
||||
true, num_alloc, app->num_procs);
|
||||
rc = ORTE_ERR_OUT_OF_RESOURCE;
|
||||
goto cleanup;
|
||||
}
|
||||
}
|
||||
/* Allocate this node */
|
||||
else {
|
||||
rc = claim_slot(map, node, jobid, vpid_start + rank, proc_index);
|
||||
if (ORTE_SUCCESS != rc) {
|
||||
return rc;
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
++rank;
|
||||
++proc_index;
|
||||
|
||||
/* Increase the number of procs allocated and see if we're
|
||||
done */
|
||||
++num_alloc;
|
||||
}
|
||||
if (node->node_slots_alloc == 0) {
|
||||
opal_list_remove_item(nodes, (opal_list_item_t*)node);
|
||||
OBJ_RELEASE(node);
|
||||
|
||||
/* Move on to the next node since we have allocated all of
|
||||
this node's slots */
|
||||
|
||||
cur_node_item = next;
|
||||
}
|
||||
}
|
||||
|
||||
/* Did we allocate everything? */
|
||||
|
||||
if (num_alloc < app->num_procs) {
|
||||
opal_show_help("help-rmaps-rr.txt", "rmaps-rr:alloc-error",
|
||||
true, num_alloc, app->num_procs);
|
||||
return ORTE_ERR_OUT_OF_RESOURCE;
|
||||
cur_node_item = next;
|
||||
}
|
||||
|
||||
map->num_procs = num_alloc;
|
||||
return ORTE_SUCCESS;
|
||||
|
||||
cleanup:
|
||||
|
||||
return rc;
|
||||
}
|
||||
|
||||
|
||||
/*
|
||||
* Create a default mapping for the application, scheduling one round
|
||||
* robin by slot.
|
||||
*/
|
||||
static int map_app_by_slot(
|
||||
orte_app_context_t* app,
|
||||
orte_rmaps_base_map_t* map,
|
||||
orte_jobid_t jobid,
|
||||
orte_vpid_t vpid_start,
|
||||
int rank,
|
||||
opal_list_t* nodes,
|
||||
char **mapped_nodes,
|
||||
int num_mapped_nodes)
|
||||
{
|
||||
int rc = ORTE_SUCCESS;
|
||||
size_t i;
|
||||
size_t num_alloc = 0;
|
||||
size_t proc_index = 0;
|
||||
orte_ras_node_t *node;
|
||||
opal_list_item_t *start, *next;
|
||||
bool oversubscribe;
|
||||
|
||||
|
||||
/* This loop continues until all procs have been mapped or we run
|
||||
out of resources. We determine that we have "run out of
|
||||
resources" when all nodes have node_slots_max processes mapped to them,
|
||||
thus there are no free slots for a process to be mapped.
|
||||
If we still have processes that haven't been mapped yet, then it's an
|
||||
"out of resources" error. */
|
||||
num_alloc = 0;
|
||||
start = cur_node_item;
|
||||
oversubscribe = false;
|
||||
while ( num_alloc < app->num_procs) {
|
||||
node = (orte_ras_node_t*) cur_node_item;
|
||||
|
||||
/* Find the next node we can use before claiming slots, since
|
||||
* we may need to prune the nodes list removing over used nodes */
|
||||
if ( 0 < app->num_map ) {
|
||||
next = get_next_mapped(cur_node_item, mapped_nodes, num_mapped_nodes, nodes);
|
||||
if (NULL == next ) {
|
||||
/* Not allocated anything */
|
||||
opal_show_help("help-orte-rmaps-rr.txt", "orte-rmaps-rr:no-mapped-node",
|
||||
true, app->app, opal_argv_join(mapped_nodes, ','));
|
||||
rc = ORTE_ERR_OUT_OF_RESOURCE;
|
||||
goto cleanup;
|
||||
}
|
||||
}
|
||||
else {
|
||||
if (opal_list_get_end(nodes) == opal_list_get_next(cur_node_item)) {
|
||||
next = opal_list_get_first(nodes);
|
||||
}
|
||||
else {
|
||||
next = opal_list_get_next(cur_node_item);
|
||||
}
|
||||
}
|
||||
|
||||
/* If we have available slots on this node, claim all of them
|
||||
* If node_slots == 0, assume 1 slot for that node.
|
||||
* JJH - is this assumption fully justified? */
|
||||
for( i = 0; i < ((node->node_slots == 0) ? 1 : node->node_slots); ++i) {
|
||||
/* If we are not oversubscribing, and this node is full, skip it. */
|
||||
if( !oversubscribe &&
|
||||
0 != node->node_slots &&
|
||||
node->node_slots_inuse > node->node_slots) {
|
||||
break;
|
||||
}
|
||||
/* If this node has reached its max number of slots,
|
||||
* take it out of the list, and skip it */
|
||||
else if( 0 != node->node_slots_max &&
|
||||
node->node_slots_inuse >= node->node_slots_max){
|
||||
opal_list_remove_item(nodes, (opal_list_item_t*)node);
|
||||
if( 0 >= opal_list_get_size(nodes) ) {
|
||||
/* No more nodes to allocate */
|
||||
opal_show_help("help-orte-rmaps-rr.txt", "orte-rmaps-rr:alloc-error",
|
||||
true, num_alloc, app->num_procs);
|
||||
rc = ORTE_ERR_OUT_OF_RESOURCE;
|
||||
goto cleanup;
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
rc = claim_slot(map, node, jobid, vpid_start + rank, proc_index);
|
||||
if (ORTE_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
/* Increase the number of procs allocated */
|
||||
++num_alloc;
|
||||
++rank;
|
||||
++proc_index;
|
||||
if(num_alloc >= app->num_procs) {
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
cur_node_item = next;
|
||||
|
||||
/* Since we have now looped back around, go ahead and oversubscribe nodes */
|
||||
if(start == cur_node_item) {
|
||||
oversubscribe = true;
|
||||
}
|
||||
}
|
||||
|
||||
map->num_procs = num_alloc;
|
||||
|
||||
cleanup:
|
||||
|
||||
return rc;
|
||||
}
|
||||
|
||||
|
||||
@ -294,8 +387,9 @@ static int map_app_by_slot(
|
||||
|
||||
static int orte_rmaps_rr_map(orte_jobid_t jobid)
|
||||
{
|
||||
orte_app_context_t** context;
|
||||
size_t i, num_context;
|
||||
orte_app_context_t** context, *app;
|
||||
orte_rmaps_base_map_t* map;
|
||||
size_t i, j, k, num_context;
|
||||
opal_list_t nodes;
|
||||
opal_list_t mapping;
|
||||
opal_list_item_t* item;
|
||||
@ -304,6 +398,8 @@ static int orte_rmaps_rr_map(orte_jobid_t jobid)
|
||||
int rank = 0;
|
||||
int rc = ORTE_SUCCESS;
|
||||
bool bynode = true;
|
||||
char **mapped_nodes = NULL;
|
||||
int num_mapped_nodes = 0;
|
||||
|
||||
/* query for the application context and allocated nodes */
|
||||
if(ORTE_SUCCESS != (rc = orte_rmgr_base_get_app_context(jobid, &context, &num_context))) {
|
||||
@ -317,9 +413,26 @@ static int orte_rmaps_rr_map(orte_jobid_t jobid)
|
||||
bynode = false;
|
||||
}
|
||||
|
||||
/* total number of procs required */
|
||||
/* query for all nodes allocated to this job */
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
if(ORTE_SUCCESS != (rc = orte_ras_base_node_query_alloc(&nodes, jobid))) {
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
return rc;
|
||||
}
|
||||
|
||||
/* Sanity check to make sure we have been allocated nodes */
|
||||
if (0 == opal_list_get_size(&nodes)) {
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
return ORTE_ERR_TEMP_OUT_OF_RESOURCE;
|
||||
}
|
||||
|
||||
/* Total number of procs required
|
||||
* DEVEL NOTE: Need to extend this when implementing C/N notation
|
||||
* Will need to set the app->num_procs approprately before this,
|
||||
* Since we will have the allocated node information at this point.
|
||||
*/
|
||||
for(i=0; i<num_context; i++) {
|
||||
orte_app_context_t* app = context[i];
|
||||
app = context[i];
|
||||
num_procs += app->num_procs;
|
||||
}
|
||||
|
||||
@ -328,19 +441,14 @@ static int orte_rmaps_rr_map(orte_jobid_t jobid)
|
||||
return rc;
|
||||
}
|
||||
|
||||
/* query for all nodes allocated to this job */
|
||||
OBJ_CONSTRUCT(&nodes, opal_list_t);
|
||||
if(ORTE_SUCCESS != (rc = orte_ras_base_node_query_alloc(&nodes, jobid))) {
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
return rc;
|
||||
}
|
||||
|
||||
/* construct a default mapping */
|
||||
/* construct a default mapping by application */
|
||||
OBJ_CONSTRUCT(&mapping, opal_list_t);
|
||||
cur_node_item = opal_list_get_first(&nodes);
|
||||
for(i=0; i<num_context; i++) {
|
||||
orte_app_context_t* app = context[i];
|
||||
orte_rmaps_base_map_t* map = OBJ_NEW(orte_rmaps_base_map_t);
|
||||
app = context[i];
|
||||
|
||||
map = OBJ_NEW(orte_rmaps_base_map_t);
|
||||
if(NULL == map) {
|
||||
rc = ORTE_ERR_OUT_OF_RESOURCE;
|
||||
goto cleanup;
|
||||
@ -353,15 +461,86 @@ static int orte_rmaps_rr_map(orte_jobid_t jobid)
|
||||
rc = ORTE_ERR_OUT_OF_RESOURCE;
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
/* Extract the requested mapping for this application */
|
||||
|
||||
/* Note that cur_node_item already points to the Right place in
|
||||
the node list to start looking (i.e., if this is the first time
|
||||
through, it'll point to the first item. If this is not the
|
||||
first time through -- i.e., we have multiple app contexts --
|
||||
it'll point to where we left off last time.). If we're at the
|
||||
end, bounce back to the front (as would happen in the loop
|
||||
below)
|
||||
|
||||
But do a bozo check to ensure that we don't have a empty node list.*/
|
||||
if (0 == opal_list_get_size(&nodes)) {
|
||||
rc = ORTE_ERR_TEMP_OUT_OF_RESOURCE;
|
||||
goto cleanup;
|
||||
} else if (opal_list_get_end(&nodes) == cur_node_item) {
|
||||
cur_node_item = opal_list_get_first(&nodes);
|
||||
}
|
||||
|
||||
/* If this application has a mapping then
|
||||
* - if the current node is in the mapping, use it
|
||||
* - ow get the next node in that mapping.
|
||||
*/
|
||||
if ( 0 < app->num_map ) {
|
||||
orte_app_context_map_t** loc_map = app->map_data;
|
||||
|
||||
/* Accumulate all of the host name mappings */
|
||||
for(k = 0; k < app->num_map; ++k) {
|
||||
if ( ORTE_APP_CONTEXT_MAP_HOSTNAME == loc_map[k]->map_type ) {
|
||||
if(mapped_nodes == NULL) {
|
||||
mapped_nodes = opal_argv_split(loc_map[k]->map_data, ',');
|
||||
num_mapped_nodes = opal_argv_count(mapped_nodes);
|
||||
}
|
||||
else { /* Append to the existing mapping */
|
||||
char ** mini_map = opal_argv_split(loc_map[k]->map_data, ',');
|
||||
size_t mini_num_map = opal_argv_count(mini_map);
|
||||
for (j = 0; j < mini_num_map; ++j) {
|
||||
rc = opal_argv_append(&num_mapped_nodes, &mapped_nodes, mini_map[j]);
|
||||
if (OPAL_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
}
|
||||
opal_argv_free(mini_map);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if( !are_all_mapped_valid(mapped_nodes, num_mapped_nodes, &nodes) ) {
|
||||
opal_show_help("help-orte-rmaps-rr.txt", "orte-rmaps-rr:not-all-mapped-alloc",
|
||||
true, app->app, opal_argv_join(mapped_nodes, ','));
|
||||
rc = ORTE_ERR_OUT_OF_RESOURCE;
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
/* If the current node is not in the current mapping
|
||||
* Then get the next node that is in the mapping */
|
||||
if( !is_mapped(cur_node_item, mapped_nodes, num_mapped_nodes, &nodes) ) {
|
||||
cur_node_item = get_next_mapped(cur_node_item, mapped_nodes, num_mapped_nodes, &nodes);
|
||||
if( NULL == cur_node_item) {
|
||||
opal_show_help("help-orte-rmaps-rr.txt", "orte-rmaps-rr:no-mapped-node",
|
||||
true, app->app, opal_argv_join(mapped_nodes, ','));
|
||||
rc = ORTE_ERR_OUT_OF_RESOURCE;
|
||||
goto cleanup;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* Make assignments */
|
||||
if (bynode) {
|
||||
rc = map_app_by_node(app, map, jobid, vpid_start, rank, &nodes);
|
||||
rc = map_app_by_node(app, map, jobid, vpid_start, rank, &nodes, mapped_nodes, num_mapped_nodes);
|
||||
} else {
|
||||
rc = map_app_by_slot(app, map, jobid, vpid_start, rank, &nodes);
|
||||
rc = map_app_by_slot(app, map, jobid, vpid_start, rank, &nodes, mapped_nodes, num_mapped_nodes);
|
||||
}
|
||||
if (ORTE_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
|
||||
rank += app->num_procs;
|
||||
opal_argv_free(mapped_nodes);
|
||||
mapped_nodes = NULL;
|
||||
}
|
||||
|
||||
/* save mapping to the registry */
|
||||
@ -377,10 +556,17 @@ cleanup:
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
OBJ_DESTRUCT(&nodes);
|
||||
|
||||
while(NULL != (item = opal_list_remove_first(&mapping))) {
|
||||
OBJ_RELEASE(item);
|
||||
}
|
||||
|
||||
OBJ_DESTRUCT(&mapping);
|
||||
|
||||
if( NULL != mapped_nodes ) {
|
||||
opal_argv_free(mapped_nodes);
|
||||
}
|
||||
|
||||
return rc;
|
||||
}
|
||||
|
||||
|
||||
@ -80,10 +80,10 @@ static int orte_rmaps_round_robin_open(void)
|
||||
|
||||
/* JMS To be changed post-beta to LAM's C/N command line notation */
|
||||
|
||||
id = mca_base_param_find("ras_base", NULL, "schedule_policy");
|
||||
id = mca_base_param_find("rmaps_base", NULL, "schedule_policy");
|
||||
if (0 > id) {
|
||||
id = mca_base_param_reg_string_name("ras_base", "schedule_policy",
|
||||
"Scheduling Policy for RAS. [slot | node]",
|
||||
id = mca_base_param_reg_string_name("rmaps_base", "schedule_policy",
|
||||
"Scheduling Policy for RMAPS. [slot | node]",
|
||||
false, false, "slot",
|
||||
&mca_rmaps_round_robin_component.schedule_policy);
|
||||
}
|
||||
|
||||
@ -103,7 +103,7 @@ int orte_rmgr_base_put_app_context(
|
||||
if(ORTE_SUCCESS != rc) {
|
||||
goto cleanup;
|
||||
}
|
||||
rc = orte_rmgr_base_set_job_slots(jobid, job_slots);
|
||||
rc = orte_rmgr_base_set_job_slots(jobid, job_slots); /* JJH C/N napping breaks here */
|
||||
|
||||
cleanup:
|
||||
OBJ_RELEASE(value);
|
||||
|
||||
@ -123,7 +123,7 @@ static int orte_rmgr_urm_create(
|
||||
return rc;
|
||||
}
|
||||
|
||||
/* create and initialize job segment */
|
||||
/* create and initialize job segment */ /* JJH C/N mapping before this */
|
||||
if (ORTE_SUCCESS !=
|
||||
(rc = orte_rmgr_base_put_app_context(*jobid, app_context,
|
||||
num_context))) {
|
||||
@ -144,9 +144,9 @@ static int orte_rmgr_urm_create(
|
||||
|
||||
static int orte_rmgr_urm_allocate(orte_jobid_t jobid)
|
||||
{
|
||||
OPAL_TRACE(1);
|
||||
OPAL_TRACE(1);
|
||||
|
||||
return mca_rmgr_urm_component.urm_ras->allocate(jobid);
|
||||
return orte_ras_base_allocate(jobid, &mca_rmgr_urm_component.urm_ras);
|
||||
}
|
||||
|
||||
static int orte_rmgr_urm_deallocate(orte_jobid_t jobid)
|
||||
@ -308,7 +308,7 @@ static int orte_rmgr_urm_spawn(
|
||||
|
||||
/*
|
||||
* Initialize job segment and allocate resources
|
||||
*/
|
||||
*/ /* JJH Insert C/N mapping stuff here */
|
||||
if (ORTE_SUCCESS !=
|
||||
(rc = orte_rmgr_urm_create(app_context,num_context,jobid))) {
|
||||
ORTE_ERROR_LOG(rc);
|
||||
|
||||
@ -176,9 +176,9 @@ static orte_rmgr_base_module_t *orte_rmgr_urm_init(int* priority)
|
||||
mca_rmgr_urm_component.urm_rds = false;
|
||||
|
||||
/**
|
||||
* Select RAS component
|
||||
* Find available RAS components
|
||||
*/
|
||||
if (NULL == (mca_rmgr_urm_component.urm_ras = orte_ras_base_select(NULL))) {
|
||||
if (ORTE_SUCCESS != (rc = orte_ras_base_find_available())) {
|
||||
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
|
||||
return NULL;
|
||||
}
|
||||
|
||||
@ -40,6 +40,7 @@
|
||||
#include "mca/sds/base/base.h"
|
||||
#include "mca/gpr/base/base.h"
|
||||
#include "mca/ras/base/base.h"
|
||||
#include "mca/ras/base/ras_base_node.h"
|
||||
#include "mca/rds/base/base.h"
|
||||
#include "mca/rmgr/base/base.h"
|
||||
#include "mca/rmaps/base/base.h"
|
||||
@ -427,6 +428,7 @@ int orte_init_stage1(bool infrastructure)
|
||||
orte_rds_cell_desc_t *rds_item;
|
||||
orte_rds_cell_attr_t *new_attr;
|
||||
orte_ras_node_t *ras_item;
|
||||
orte_ras_base_module_t *module;
|
||||
|
||||
OBJ_CONSTRUCT(&single_host, opal_list_t);
|
||||
OBJ_CONSTRUCT(&rds_single_host, opal_list_t);
|
||||
@ -487,13 +489,28 @@ int orte_init_stage1(bool infrastructure)
|
||||
goto error;
|
||||
}
|
||||
|
||||
ret = orte_ras.node_insert(&single_host);
|
||||
/* JMS: This isn't quite right and should be fixed after
|
||||
1.0 -- we shouldn't be doing this manually here. We
|
||||
should somehow be invoking a real RAS component to do
|
||||
this for us. */
|
||||
ret = orte_ras_base_node_insert(&single_host);
|
||||
if (ORTE_SUCCESS != ret ) {
|
||||
ORTE_ERROR_LOG(ret);
|
||||
error = "orte_ras.node_insert";
|
||||
goto error;;
|
||||
}
|
||||
|
||||
/* JMS: Same as above -- fix this after 1.0: force a
|
||||
selection so that orte_ras has initialized pointers in
|
||||
case anywhere else tries to use it. This may end up
|
||||
putting a bunch more nodes on the node segment (e.g.,
|
||||
if you're in a SLURM allocation and you "./a.out",
|
||||
you'll end up with the localhost *and* all the other
|
||||
nodes in your allocation on the node segment -- which
|
||||
is probably fine) */
|
||||
orte_ras_base_allocate(my_jobid, &module);
|
||||
orte_ras = *module;
|
||||
|
||||
OBJ_DESTRUCT(&single_host);
|
||||
OBJ_DESTRUCT(&rds_single_host);
|
||||
}
|
||||
|
||||
@ -715,11 +715,11 @@ static int parse_globals(int argc, char* argv[])
|
||||
|
||||
/* JMS To be changed post-beta to LAM's C/N command line notation */
|
||||
/* Don't initialize the MCA parameter here unless we have to,
|
||||
* since it really should be initialized in ras_base_open */
|
||||
* since it really should be initialized in rmaps_base_open */
|
||||
if (orterun_globals.by_node || orterun_globals.by_slot) {
|
||||
char *policy = NULL;
|
||||
id = mca_base_param_reg_string_name("ras_base", "schedule_policy",
|
||||
"Scheduling Policy for RAS. [slot | node]",
|
||||
id = mca_base_param_reg_string_name("rmaps_base", "schedule_policy",
|
||||
"Scheduling policy for RMAPS. [slot | node]",
|
||||
false, false, "slot", &policy);
|
||||
|
||||
if (orterun_globals.by_node) {
|
||||
@ -1144,6 +1144,10 @@ static int create_app(int argc, char* argv[], orte_app_context_t **app_ptr,
|
||||
}
|
||||
app->map_data[i]->map_type = value[0] - '0';
|
||||
app->map_data[i]->map_data = strdup(value2);
|
||||
/* map_data = true;
|
||||
* JJH - This activates the C/N mapping stuff,
|
||||
* or at least allows us to pass the 'num_procs' check below.
|
||||
* since it is not implemented yet, leave commented. */
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
Загрузка…
x
Ссылка в новой задаче
Block a user