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

Restore the paffinity capability, along with all the required logic to ensure we "do the right thing" when the user gives us inaccurate information about the number of slots on a remote node.

Просмотр исходного кода 
This commit was SVN r12780.
Этот коммит содержится в:
Ralph Castain 2006-12-06 15:59:34 +00:00
родитель b1e16fffac
Коммит d4bd60c9fe
20 изменённых файлов: 275 добавлений и 94 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

109
orte/mca/odls/default/odls_default_module.c
Просмотреть файл

@ -119,15 +119,17 @@ int orte_odls_default_subscribe_launch_data(orte_jobid_t job, orte_gpr_notify_cb
char *glob_keys[] = {
ORTE_JOB_APP_CONTEXT_KEY,
ORTE_JOB_VPID_START_KEY,
ORTE_JOB_VPID_RANGE_KEY
ORTE_JOB_VPID_RANGE_KEY,
ORTE_JOB_OVERSUBSCRIBE_OVERRIDE_KEY
};
int num_glob_keys = 3;
int num_glob_keys = 4;
char* keys[] = {
ORTE_PROC_NAME_KEY,
ORTE_PROC_APP_CONTEXT_KEY,
ORTE_NODE_NAME_KEY,
ORTE_NODE_OVERSUBSCRIBED_KEY
};
int num_keys = 3;
int num_keys = 4;
int i, rc;
/* get the job segment name */
@ -749,7 +751,22 @@ static int odls_default_fork_local_proc(
opal_unsetenv(param, &environ_copy);
free(param);
/* Handle processor affinity */
/* setup yield schedule and processor affinity
* We default here to always setting the affinity processor if we want
* it. The processor affinity system then determines
* if processor affinity is enabled/requested - if so, it then uses
* this value to select the process to which the proc is "assigned".
* Otherwise, the paffinity subsystem just ignores this value anyway
*/
if (oversubscribed) {
param = mca_base_param_environ_variable("mpi", NULL, "yield_when_idle");
opal_setenv(param, "1", true, &environ_copy);
} else {
param = mca_base_param_environ_variable("mpi", NULL, "yield_when_idle");
opal_setenv(param, "0", true, &environ_copy);
}
free(param);
if (want_processor) {
param = mca_base_param_environ_variable("mpi", NULL,
"paffinity_processor");
@ -757,15 +774,11 @@ static int odls_default_fork_local_proc(
opal_setenv(param, param2, true, &environ_copy);
free(param);
free(param2);
}
/* handle oversubscription - setup yield schedule */
if (oversubscribed) {
param = mca_base_param_environ_variable("mpi", NULL, "yield_when_idle");
opal_setenv(param, "1", true, &environ_copy);
} else {
param = mca_base_param_environ_variable("mpi", NULL, "yield_when_idle");
opal_setenv(param, "0", true, &environ_copy);
param = mca_base_param_environ_variable("mpi", NULL,
"paffinity_processor");
opal_unsetenv(param, &environ_copy);
free(param);
}
/* setup universe info */
@ -910,8 +923,8 @@ int orte_odls_default_launch_local_procs(orte_gpr_notify_data_t *data, char **ba
opal_list_t app_context_list;
orte_odls_child_t *child;
odls_default_app_context_t *app_item;
size_t num_processors;
bool want_processor=false, oversubscribed;
int num_processors;
bool oversubscribed=false, want_processor, *bptr, override_oversubscribed=false;
opal_list_item_t *item, *item2;
/* parse the returned data to create the required structures
@ -992,6 +1005,15 @@ int orte_odls_default_launch_local_procs(orte_gpr_notify_data_t *data, char **ba
opal_list_append(&app_context_list, &app_item->super);
kval->value->data = NULL; /* protect the data storage from later release */
}
if (strcmp(kval->key, ORTE_JOB_OVERSUBSCRIBE_OVERRIDE_KEY) == 0) {
/* this can only occur once, so just store it */
if (ORTE_SUCCESS != (rc = orte_dss.get((void**)&bptr, kval->value, ORTE_BOOL))) {
ORTE_ERROR_LOG(rc);
return rc;
}
override_oversubscribed = *bptr;
continue;
}
} /* end for loop to process global data */
} else {
/* this must have come from one of the process containers, so it must
@ -1030,6 +1052,14 @@ int orte_odls_default_launch_local_procs(orte_gpr_notify_data_t *data, char **ba
child->app_idx = *sptr; /* save the index into the app_context objects */
continue;
}
if(strcmp(kval->key, ORTE_NODE_OVERSUBSCRIBED_KEY) == 0) {
if (ORTE_SUCCESS != (rc = orte_dss.get((void**)&bptr, kval->value, ORTE_BOOL))) {
ORTE_ERROR_LOG(rc);
return rc;
}
oversubscribed = *bptr;
continue;
}
} /* kv2 */
/* protect operation on the global list of children */
OPAL_THREAD_LOCK(&orte_odls_default.mutex);
@ -1044,13 +1074,52 @@ int orte_odls_default_launch_local_procs(orte_gpr_notify_data_t *data, char **ba
} /* for j */
}
/* determine if we are oversubscribed */
oversubscribed = false; /* default to being a hog */
opal_paffinity_base_get_num_processors(&rc);
num_processors = (size_t)rc;
if (opal_list_get_size(&orte_odls_default.children) > num_processors) { /* oversubscribed */
oversubscribed = true;
/* setup for processor affinity. If there are enough physical processors on this node, then
* we indicate which processor each process should be assigned to, IFF the user has requested
* processor affinity be used - the paffinity subsystem will make that final determination. All
* we do here is indicate that we should do the definitions just in case paffinity is active
*/
if (ORTE_SUCCESS != opal_paffinity_base_get_num_processors(&num_processors)) {
/* if we cannot find the number of local processors, then default to conservative
* settings
*/
want_processor = false; /* default to not being a hog */
/* leave oversubscribed alone */
opal_output(orte_odls_globals.output,
"odls: could not get number of processors - using conservative settings");
} else {
/* only do this if we can actually get info on the number of processors */
if (opal_list_get_size(&orte_odls_default.children) > (size_t)num_processors) {
want_processor = false;
} else {
want_processor = true;
}
/* now let's deal with the oversubscribed flag - and the use-case where a hostfile or some
* other non-guaranteed-accurate method was used to inform us about our allocation. Since
* the information on the number of slots on this node could have been incorrect, we need
* to check it against the local number of processors to ensure we don't overload them
*/
if (override_oversubscribed) {
opal_output(orte_odls_globals.output, "odls: overriding oversubscription");
if (opal_list_get_size(&orte_odls_default.children) > (size_t)num_processors) {
/* if the #procs > #processors, declare us oversubscribed regardless
* of what the mapper claimed - the user may have told us something
* incorrect
*/
oversubscribed = true;
} else {
/* likewise, if there are more processors here than we were told,
* declare us to not be oversubscribed so we can be aggressive. This
* covers the case where the user didn't tell us anything about the
* number of available slots, so we defaulted to a value of 1
*/
oversubscribed = false;
}
}
}
opal_output(orte_odls_globals.output, "odls: oversubscribed set to %s want_processor set to %s",
oversubscribed ? "true" : "false", want_processor ? "true" : "false");
/* okay, now let's launch our local procs using a fork/exec */
i = 0;

1
orte/mca/ras/base/Makefile.am
Просмотреть файл

@ -30,6 +30,7 @@ libmca_ras_la_SOURCES += \
base/ras_base_no_ops.c \
base/ras_base_open.c \
base/ras_base_receive.c \
base/ras_base_support_fns.c \
base/data_type_support/ras_data_type_compare_fns.c \
base/data_type_support/ras_data_type_copy_fns.c \
base/data_type_support/ras_data_type_packing_fns.c \

66
orte/mca/ras/base/ras_base_support_fns.c Обычный файл
Просмотреть файл

@ -0,0 +1,66 @@
/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/orte_constants.h"
#include "orte/dss/dss_types.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/gpr/gpr.h"
#include "orte/mca/schema/schema_types.h"
#include "orte/mca/ras/base/ras_private.h"
/*
* Indicate that the allocation for this job is uncertain - therefore,
* the oversubscribed condition that might results should be overridden
* locally based on the actual available hardware on the node
*/
int orte_ras_base_set_oversubscribe_override(orte_jobid_t job)
{
orte_gpr_addr_mode_t addr_mode;
char *segment;
char *tokens[] = {
ORTE_JOB_GLOBALS,
NULL
};
orte_data_value_t val = ORTE_DATA_VALUE_EMPTY;
bool trueval = true;
int rc;
addr_mode = ORTE_GPR_OVERWRITE | ORTE_GPR_TOKENS_OR | ORTE_GPR_KEYS_OR;
/* get the job segment name */
if (ORTE_SUCCESS != (rc = orte_schema.get_job_segment_name(&segment, job))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* setup the value to be true */
val.type = ORTE_BOOL;
val.data = &trueval;
if (ORTE_SUCCESS != (rc = orte_gpr.put_1(addr_mode, segment, tokens,
ORTE_JOB_OVERSUBSCRIBE_OVERRIDE_KEY, &val))) {
ORTE_ERROR_LOG(rc);
}
free(segment);
return rc;
}

2
orte/mca/ras/base/ras_private.h
Просмотреть файл

@ -80,6 +80,8 @@ ORTE_DECLSPEC int orte_ras_base_allocate_nodes(orte_jobid_t jobid,
ORTE_DECLSPEC int orte_ras_base_reallocate(orte_jobid_t parent_jobid,
orte_jobid_t child_jobid);
ORTE_DECLSPEC int orte_ras_base_set_oversubscribe_override(orte_jobid_t job);
/*
* Query the registry for all available nodes
*/

9
orte/mca/ras/dash_host/ras_dash_host_module.c
Просмотреть файл

@ -166,6 +166,15 @@ static int orte_ras_dash_host_allocate(orte_jobid_t jobid, opal_list_t *attribut
(rc = orte_ras_base_allocate_nodes(jobid, &nodes))) {
goto cleanup;
}
/* now indicate that there is uncertainty about the number of slots here,
* so the launcher should use knowledge of the local number of processors to
* override any oversubscription flags
*/
rc = orte_ras_base_set_oversubscribe_override(jobid);
if (ORTE_SUCCESS != rc) {
goto cleanup;
}
}
cleanup:

11
orte/mca/ras/localhost/ras_localhost_module.c
Просмотреть файл

@ -119,7 +119,16 @@ static int orte_ras_localhost_allocate(orte_jobid_t jobid, opal_list_t *attribut
goto cleanup;
}
cleanup:
/* now indicate that there is uncertainty about the number of slots here,
* so the launcher should use knowledge of the local number of processors to
* override any oversubscription flags
*/
ret = orte_ras_base_set_oversubscribe_override(jobid);
if (ORTE_SUCCESS != ret) {
goto cleanup;
}
cleanup:
item = opal_list_remove_first(&nodes);
OBJ_RELEASE(item);
OBJ_DESTRUCT(&nodes);

2
orte/mca/rds/base/rds_base_no_ops.c
Просмотреть файл

@ -28,7 +28,7 @@
#include "orte/mca/rds/base/rds_private.h"
int orte_rds_base_no_op_query(void)
int orte_rds_base_no_op_query(orte_jobid_t job)
{
return ORTE_ERR_NOT_SUPPORTED;
}

4
orte/mca/rds/base/rds_base_query.c
Просмотреть файл

@ -30,7 +30,7 @@
/**
* Function for querying all loaded components.
*/
int orte_rds_base_query(void)
int orte_rds_base_query(orte_jobid_t job)
{
opal_list_item_t* item;
@ -39,7 +39,7 @@ int orte_rds_base_query(void)
item != opal_list_get_end(&orte_rds_base.rds_selected);
item = opal_list_get_next(item)) {
orte_rds_base_selected_t* selected = (orte_rds_base_selected_t*)item;
int rc = selected->module->query();
int rc = selected->module->query(job);
if(rc != ORTE_SUCCESS)
return rc;
}

11
orte/mca/rds/base/rds_base_receive.c
Просмотреть файл

@ -92,6 +92,7 @@ void orte_rds_base_recv(int status, orte_process_name_t* sender,
{
orte_buffer_t answer;
orte_rds_cmd_flag_t command;
orte_jobid_t job;
orte_std_cntr_t count;
int rc;
@ -105,11 +106,17 @@ void orte_rds_base_recv(int status, orte_process_name_t* sender,
switch (command) {
case ORTE_RDS_QUERY_CMD:
if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, &command, 1, ORTE_RDS_CMD))) {
count = 1;
if (ORTE_SUCCESS != (rc = orte_dss.unpack(buffer, &job, &count, ORTE_JOBID))) {
ORTE_ERROR_LOG(rc);
return;
}
if (ORTE_SUCCESS != (rc = orte_dss.pack(&answer, &command, 1, ORTE_RDS_CMD))) {
ORTE_ERROR_LOG(rc);
}
if (ORTE_SUCCESS != (rc = orte_rds_base_query())) {
if (ORTE_SUCCESS != (rc = orte_rds_base_query(job))) {
ORTE_ERROR_LOG(rc);
}

4
orte/mca/rds/base/rds_private.h
Просмотреть файл

@ -46,7 +46,7 @@ typedef uint8_t orte_rds_cmd_flag_t;
/*
* API function definitions
*/
ORTE_DECLSPEC int orte_rds_base_query(void);
ORTE_DECLSPEC int orte_rds_base_query(orte_jobid_t job);
/*
* oob interface
@ -65,7 +65,7 @@ ORTE_DECLSPEC int orte_rds_base_store_resource(opal_list_t *resource_list);
/*
* the "null" component functions
*/
int orte_rds_base_no_op_query(void);
int orte_rds_base_no_op_query(orte_jobid_t job);
int orte_rds_base_no_op_store_resource(opal_list_t *resource_list);
#if defined(c_plusplus) || defined(__cplusplus)

14
orte/mca/rds/hostfile/rds_hostfile.c
Просмотреть файл

@ -372,7 +372,7 @@ unlock:
* rds_hostfile_path, and add the nodes to the registry.
*/
static int orte_rds_hostfile_query(void)
static int orte_rds_hostfile_query(orte_jobid_t job)
{
opal_list_t existing;
opal_list_t updates, rds_updates;
@ -501,6 +501,18 @@ static int orte_rds_hostfile_query(void)
if (ORTE_SUCCESS != rc) {
goto cleanup;
}
/* and now, indicate that ORTE should override any oversubscribed conditions
* based on local hardware limits since the user (a) might not have
* provided us any info on the #slots for a node, and (b) the user
* might have been wrong! If we don't check the number of local physical
* processors, then we could be too aggressive on our sched_yield setting
* and cause performance problems.
*/
rc = orte_ras_base_set_oversubscribe_override(job);
if (ORTE_SUCCESS != rc) {
goto cleanup;
}
}
cleanup:

8
orte/mca/rds/proxy/rds_proxy.c
Просмотреть файл

@ -41,7 +41,7 @@
* functions
*/
int orte_rds_proxy_query(void)
int orte_rds_proxy_query(orte_jobid_t job)
{
orte_buffer_t* cmd;
orte_buffer_t* answer;
@ -63,6 +63,12 @@ int orte_rds_proxy_query(void)
return rc;
}
if (ORTE_SUCCESS != (rc = orte_dss.pack(cmd, &job, 1, ORTE_JOBID))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(cmd);
return rc;
}
if (0 > orte_rml.send_buffer(orte_rds_proxy_replica, cmd, ORTE_RML_TAG_RDS, 0)) {
ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
OBJ_RELEASE(cmd);

2
orte/mca/rds/proxy/rds_proxy.h
Просмотреть файл

@ -47,7 +47,7 @@ int orte_rds_proxy_finalize(void);
/*
* proxy function prototypes
*/
int orte_rds_proxy_query(void);
int orte_rds_proxy_query(orte_jobid_t job);
#if defined(c_plusplus) || defined(__cplusplus)

2
orte/mca/rds/rds.h
Просмотреть файл

@ -147,7 +147,7 @@ extern "C" {
* return_value = orte_rmgr.query();
* @endcode
*/
typedef int (*orte_rds_base_module_query_fn_t)(void);
typedef int (*orte_rds_base_module_query_fn_t)(orte_jobid_t job);
/**
* Add a list of resources to the Resource Segment

2
orte/mca/rds/resfile/rds_resfile.c
Просмотреть файл

@ -223,7 +223,7 @@ static int orte_rds_resfile_parse_site(char *site, FILE *fp)
}
int orte_rds_resfile_query(void)
int orte_rds_resfile_query(orte_jobid_t job)
{
int fileid, rc;
FILE *fp;

2
orte/mca/rds/resfile/rds_resfile.h
Просмотреть файл

@ -36,7 +36,7 @@ extern "C" {
/*
* RDS Resource file functions
*/
int orte_rds_resfile_query(void);
int orte_rds_resfile_query(orte_jobid_t job);
int orte_rds_resfile_finalize(void);

15
orte/mca/rmgr/proxy/rmgr_proxy.c
Просмотреть файл

@ -384,14 +384,6 @@ static int orte_rmgr_proxy_spawn_job(
flags = 0xff;
}
/*
* Perform resource discovery.
*/
if (ORTE_SUCCESS != (rc = orte_rds.query())) {
ORTE_ERROR_LOG(rc);
return rc;
}
/*
* Setup job and allocate resources
*/
@ -403,6 +395,13 @@ static int orte_rmgr_proxy_spawn_job(
}
}
if (flags & ORTE_RMGR_RES_DISC) {
if (ORTE_SUCCESS != (rc = orte_rds.query(*jobid))) {
ORTE_ERROR_LOG(rc);
return rc;
}
}
if (flags & ORTE_RMGR_ALLOC) {
if (ORTE_SUCCESS != (rc = orte_ras.allocate_job(*jobid, attributes))) {
ORTE_ERROR_LOG(rc);

9
orte/mca/rmgr/rmgr_types.h
Просмотреть файл

@ -56,10 +56,11 @@ ORTE_DECLSPEC OBJ_CLASS_DECLARATION(orte_attribute_t);
* procedure.
*/
#define ORTE_RMGR_SETUP 0x01
#define ORTE_RMGR_ALLOC 0x02
#define ORTE_RMGR_MAP 0x04
#define ORTE_RMGR_SETUP_TRIGS 0x08
#define ORTE_RMGR_LAUNCH 0x10
#define ORTE_RMGR_RES_DISC 0x02
#define ORTE_RMGR_ALLOC 0x04
#define ORTE_RMGR_MAP 0x08
#define ORTE_RMGR_SETUP_TRIGS 0x10
#define ORTE_RMGR_LAUNCH 0x20
/* direct the RMGR spawn procedure to use the provided jobid
* instead of getting a new one

15
orte/mca/rmgr/urm/rmgr_urm.c
Просмотреть файл

@ -336,14 +336,6 @@ static int orte_rmgr_urm_spawn_job(
flags = 0xff;
}
/*
* Perform resource discovery.
*/
if (ORTE_SUCCESS != (rc = orte_rds.query())) {
ORTE_ERROR_LOG(rc);
return rc;
}
/*
* Initialize job segment and allocate resources
*/ /* JJH Insert C/N mapping stuff here */
@ -359,6 +351,13 @@ static int orte_rmgr_urm_spawn_job(
}
}
if (flags & ORTE_RMGR_RES_DISC) {
if (ORTE_SUCCESS != (rc = orte_rds.query(*jobid))) {
ORTE_ERROR_LOG(rc);
return rc;
}
}
if (flags & ORTE_RMGR_ALLOC) {
if (ORTE_SUCCESS != (rc = orte_ras.allocate_job(*jobid, attributes))) {
ORTE_ERROR_LOG(rc);

81
orte/mca/schema/schema_types.h
Просмотреть файл

@ -58,46 +58,47 @@
* ORTE-wide key names for storing/retrieving data from the registry.
* Subsystem-specific keys will be defined in each=/ subsystem's xxx_types.h file.
*/
#define ORTE_CELLID_KEY "orte-cellid"
#define ORTE_JOBGRP_KEY "orte-jobgrp"
#define ORTE_JOBID_KEY "orte-jobid"
#define ORTE_VPID_KEY "orte-vpid"
#define ORTE_NODE_NAME_KEY "orte-node-name"
#define ORTE_NODE_ARCH_KEY "orte-node-arch"
#define ORTE_NODE_STATE_KEY "orte-node-state"
#define ORTE_NODE_SLOTS_KEY "orte-node-slots"
#define ORTE_NODE_SLOTS_ALLOC_KEY "orte-node-slots-alloc"
#define ORTE_NODE_SLOTS_IN_USE_KEY "orte-node-slots-in-use"
#define ORTE_NODE_SLOTS_MAX_KEY "orte-node-slots-max"
#define ORTE_NODE_ALLOC_KEY "orte-node-alloc"
#define ORTE_NODE_BOOTPROXY_KEY "orte-node-bootproxy"
#define ORTE_NODE_USERNAME_KEY "orte-node-username"
#define ORTE_NODE_OVERSUBSCRIBED_KEY "orte-node-oversubscribed"
#define ORTE_JOB_APP_CONTEXT_KEY "orte-job-app-context"
#define ORTE_JOB_SLOTS_KEY "orte-job-slots" /**< number of procs in job */
#define ORTE_JOB_VPID_START_KEY "orte-job-vpid-start"
#define ORTE_JOB_VPID_RANGE_KEY "orte-job-vpid-range"
#define ORTE_JOB_IOF_KEY "orte-job-iof"
#define ORTE_JOB_STATE_KEY "orte-job-state"
#define ORTE_PROC_NAME_KEY "orte-proc-name"
#define ORTE_PROC_RANK_KEY "orte-proc-rank"
#define ORTE_PROC_PID_KEY "orte-proc-pid"
#define ORTE_PROC_LOCAL_PID_KEY "orte-proc-local-pid"
#define ORTE_PROC_STATE_KEY "orte-proc-state"
#define ORTE_PROC_APP_CONTEXT_KEY "orte-proc-app-context"
#define ORTE_PROC_EXIT_CODE_KEY "orte-proc-exit-code"
#define ORTE_PROC_NUM_ALIVE "orte-proc-num-alive"
#define ORTE_PROC_NUM_ABORTED "orte-proc-num-aborted"
#define ORTE_PROC_NUM_FAILED_START "orte-proc-num-failed-start"
#define ORTE_PROC_NUM_AT_INIT "orte-proc-num-init"
#define ORTE_PROC_NUM_LAUNCHED "orte-proc-num-launched"
#define ORTE_PROC_NUM_RUNNING "orte-proc-num-running"
#define ORTE_PROC_NUM_AT_STG1 "orte-proc-num-stg1"
#define ORTE_PROC_NUM_AT_STG2 "orte-proc-num-stg2"
#define ORTE_PROC_NUM_AT_STG3 "orte-proc-num-stg3"
#define ORTE_PROC_NUM_FINALIZED "orte-proc-num-finalized"
#define ORTE_PROC_NUM_TERMINATED "orte-proc-num-terminated"
#define ORTE_PROC_RML_IP_ADDRESS_KEY "orte-proc-rml-ip-addr"
#define ORTE_CELLID_KEY "orte-cellid"
#define ORTE_JOBGRP_KEY "orte-jobgrp"
#define ORTE_JOBID_KEY "orte-jobid"
#define ORTE_VPID_KEY "orte-vpid"
#define ORTE_NODE_NAME_KEY "orte-node-name"
#define ORTE_NODE_ARCH_KEY "orte-node-arch"
#define ORTE_NODE_STATE_KEY "orte-node-state"
#define ORTE_NODE_SLOTS_KEY "orte-node-slots"
#define ORTE_NODE_SLOTS_ALLOC_KEY "orte-node-slots-alloc"
#define ORTE_NODE_SLOTS_IN_USE_KEY "orte-node-slots-in-use"
#define ORTE_NODE_SLOTS_MAX_KEY "orte-node-slots-max"
#define ORTE_NODE_ALLOC_KEY "orte-node-alloc"
#define ORTE_NODE_BOOTPROXY_KEY "orte-node-bootproxy"
#define ORTE_NODE_USERNAME_KEY "orte-node-username"
#define ORTE_NODE_OVERSUBSCRIBED_KEY "orte-node-oversubscribed"
#define ORTE_JOB_APP_CONTEXT_KEY "orte-job-app-context"
#define ORTE_JOB_SLOTS_KEY "orte-job-slots" /**< number of procs in job */
#define ORTE_JOB_VPID_START_KEY "orte-job-vpid-start"
#define ORTE_JOB_VPID_RANGE_KEY "orte-job-vpid-range"
#define ORTE_JOB_OVERSUBSCRIBE_OVERRIDE_KEY "orte-job-override-oversubscribe"
#define ORTE_JOB_IOF_KEY "orte-job-iof"
#define ORTE_JOB_STATE_KEY "orte-job-state"
#define ORTE_PROC_NAME_KEY "orte-proc-name"
#define ORTE_PROC_RANK_KEY "orte-proc-rank"
#define ORTE_PROC_PID_KEY "orte-proc-pid"
#define ORTE_PROC_LOCAL_PID_KEY "orte-proc-local-pid"
#define ORTE_PROC_STATE_KEY "orte-proc-state"
#define ORTE_PROC_APP_CONTEXT_KEY "orte-proc-app-context"
#define ORTE_PROC_EXIT_CODE_KEY "orte-proc-exit-code"
#define ORTE_PROC_NUM_ALIVE "orte-proc-num-alive"
#define ORTE_PROC_NUM_ABORTED "orte-proc-num-aborted"
#define ORTE_PROC_NUM_FAILED_START "orte-proc-num-failed-start"
#define ORTE_PROC_NUM_AT_INIT "orte-proc-num-init"
#define ORTE_PROC_NUM_LAUNCHED "orte-proc-num-launched"
#define ORTE_PROC_NUM_RUNNING "orte-proc-num-running"
#define ORTE_PROC_NUM_AT_STG1 "orte-proc-num-stg1"
#define ORTE_PROC_NUM_AT_STG2 "orte-proc-num-stg2"
#define ORTE_PROC_NUM_AT_STG3 "orte-proc-num-stg3"
#define ORTE_PROC_NUM_FINALIZED "orte-proc-num-finalized"
#define ORTE_PROC_NUM_TERMINATED "orte-proc-num-terminated"
#define ORTE_PROC_RML_IP_ADDRESS_KEY "orte-proc-rml-ip-addr"
/*
* ORTE-wide names for specific system triggers and subscriptions