Merge pull request #1602 from rhc54/topic/psm

Enable PSM to support dynamic processes

orte/mca/plm/base/plm_base_launch_support.c

@@ -70,6 +70,7 @@
 #include "orte/runtime/orte_quit.h"
 #include "orte/util/name_fns.h"
 #include "orte/util/nidmap.h"
+#include "orte/util/pre_condition_transports.h"
 #include "orte/util/proc_info.h"
 #include "orte/util/regex.h"
 #include "orte/mca/state/state.h"
@@ -272,6 +273,9 @@ void orte_plm_base_setup_job(int fd, short args, void *cbdata)
     int i;
     orte_app_context_t *app;
     orte_state_caddy_t *caddy = (orte_state_caddy_t*)cbdata;
+    char *key;
+    orte_job_t *parent;
+    orte_process_name_t name, *nptr;
 
     OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
                          "%s plm:base:setup_job",
@@ -308,6 +312,50 @@ void orte_plm_base_setup_job(int fd, short args, void *cbdata)
         ORTE_FLAG_SET(caddy->jdata, ORTE_JOB_FLAG_RECOVERABLE);
     }
 
+    /* setup transport keys in case the MPI layer needs them. If
+     * this is a dynamic spawn, then use the same keys as the
+     * parent process had so the new/old procs can communicate.
+     * Otherwise we can use the jobfam and stepid as unique keys
+     * because they are unique values assigned by the RM
+     */
+     nptr = &name;
+     if (orte_get_attribute(&caddy->jdata->attributes, ORTE_JOB_LAUNCH_PROXY, (void**)&nptr, OPAL_NAME)) {
+        /* get the parent jdata */
+        if (NULL == (parent = orte_get_job_data_object(name.jobid))) {
+            ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+            ORTE_FORCED_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
+            OBJ_RELEASE(caddy);
+            return;
+        }
+        key = NULL;
+        if (!orte_get_attribute(&parent->attributes, ORTE_JOB_TRANSPORT_KEY, (void**)&key, OPAL_STRING) ||
+            NULL == key) {
+            ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+            ORTE_FORCED_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
+            OBJ_RELEASE(caddy);
+            return;
+        }
+        /* record it */
+        orte_set_attribute(&caddy->jdata->attributes, ORTE_JOB_TRANSPORT_KEY, ORTE_ATTR_LOCAL, key, OPAL_STRING);
+        /* add the transport key envar to each app */
+        for (i=0; i < caddy->jdata->apps->size; i++) {
+            if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(caddy->jdata->apps, i))) {
+                continue;
+            }
+            opal_setenv(OPAL_MCA_PREFIX"orte_precondition_transports", key, true, &app->env);
+        }
+        free(key);
+    } else {
+        /* this will also record the transport key attribute in the job object, and
+         * adds the key envar to each app */
+        if (ORTE_SUCCESS != (rc = orte_pre_condition_transports(caddy->jdata))) {
+            ORTE_ERROR_LOG(rc);
+            ORTE_FORCED_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
+            OBJ_RELEASE(caddy);
+            return;
+        }
+    }
+
     /* if app recovery is not defined, set apps to defaults */
     for (i=0; i < caddy->jdata->apps->size; i++) {
         if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(caddy->jdata->apps, i))) {

orte/mca/plm/base/plm_base_receive.c

@@ -127,7 +127,7 @@ void orte_plm_base_recv(int status, orte_process_name_t* sender,
     orte_exit_code_t exit_code;
     int32_t rc=ORTE_SUCCESS, ret;
     orte_app_context_t *app, *child_app;
-    orte_process_name_t name;
+    orte_process_name_t name, *nptr;
     pid_t pid;
     bool running;
     int i, room;
@@ -162,8 +162,17 @@ void orte_plm_base_recv(int status, orte_process_name_t* sender,
         jdata->originator.jobid = sender->jobid;
         jdata->originator.vpid = sender->vpid;
 
+        /* get the name of the actual spawn parent - i.e., the proc that actually
+         * requested the spawn */
+        nptr = &name;
+        if (!orte_get_attribute(&jdata->attributes, ORTE_JOB_LAUNCH_PROXY, (void**)&nptr, OPAL_NAME)) {
+            ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+            rc = ORTE_ERR_NOT_FOUND;
+            goto ANSWER_LAUNCH;
+        }
+
         /* get the parent's job object */
-        if (NULL != (parent = orte_get_job_data_object(sender->jobid))) {
+        if (NULL != (parent = orte_get_job_data_object(name.jobid))) {
             /* if the prefix was set in the parent's job, we need to transfer
              * that prefix to the child's app_context so any further launch of
              * orteds can find the correct binary. There always has to be at

orte/orted/orted_submit.c

@@ -1010,14 +1010,6 @@ int orte_submit_job(char *argv[], int *index,
             free(param);
         }
 
-        /* pre-condition any network transports that require it */
-        if (ORTE_SUCCESS != (rc = orte_pre_condition_transports(jdata))) {
-            ORTE_ERROR_LOG(rc);
-            orte_show_help("help-orterun.txt", "orterun:precondition", false,
-                           orte_basename, NULL, NULL, rc);
-            ORTE_UPDATE_EXIT_STATUS(ORTE_ERROR_DEFAULT_EXIT_CODE);
-            return rc;
-        }
         /* setup for debugging */
         orte_debugger_init_before_spawn(jdata);
 

orte/tools/orterun/orterun.c

@@ -187,21 +187,21 @@ int orterun(int argc, char *argv[])
         }
     }
 
-      // wait for response and unpack the status, jobid
-      while (orte_event_base_active && launchst.active) {
-          opal_event_loop(orte_event_base, OPAL_EVLOOP_ONCE);
-      }
-      if (orte_debug_flag) {
-          opal_output(0, "Job %s has launched",
-                      (NULL == launchst.jdata) ? "UNKNOWN" : ORTE_JOBID_PRINT(launchst.jdata->jobid));
-      }
-      if (!orte_event_base_active || ORTE_SUCCESS != launchst.status) {
-          goto DONE;
-      }
+    // wait for response and unpack the status, jobid
+    while (orte_event_base_active && launchst.active) {
+        opal_event_loop(orte_event_base, OPAL_EVLOOP_ONCE);
+    }
+    if (orte_debug_flag) {
+        opal_output(0, "Job %s has launched",
+                   (NULL == launchst.jdata) ? "UNKNOWN" : ORTE_JOBID_PRINT(launchst.jdata->jobid));
+    }
+    if (!orte_event_base_active || ORTE_SUCCESS != launchst.status) {
+        goto DONE;
+    }
 
-      while (orte_event_base_active && completest.active) {
-          opal_event_loop(orte_event_base, OPAL_EVLOOP_ONCE);
-      }
+    while (orte_event_base_active && completest.active) {
+        opal_event_loop(orte_event_base, OPAL_EVLOOP_ONCE);
+    }
 
     if (ORTE_PROC_IS_HNP) {
         /* ensure all local procs are dead */

orte/util/attr.c

@@ -14,6 +14,7 @@
 #include "orte/constants.h"
 
 #include "opal/dss/dss.h"
+#include "opal/util/output.h"
 
 #include "orte/mca/errmgr/errmgr.h"
 
@@ -279,6 +280,8 @@ const char *orte_attr_key_to_str(orte_attribute_key_t key)
             return "ORTE_JOB_MULTI_DAEMON_SIM";
         case ORTE_JOB_NOTIFY_COMPLETION:
             return "ORTE_JOB_NOTIFY_COMPLETION";
+        case ORTE_JOB_TRANSPORT_KEY:
+            return "ORTE_JOB_TRANSPORT_KEY";
 
         case ORTE_PROC_NOBARRIER:
             return "PROC-NOBARRIER";

orte/util/attr.h

@@ -139,6 +139,7 @@ typedef uint16_t orte_job_flags_t;
 #define ORTE_JOB_TIMESTAMP_OUTPUT       (ORTE_JOB_START_KEY + 48)    // bool - timestamp stdout/stderr
 #define ORTE_JOB_MULTI_DAEMON_SIM       (ORTE_JOB_START_KEY + 49)    // bool - multiple daemons/node to simulate large cluster
 #define ORTE_JOB_NOTIFY_COMPLETION      (ORTE_JOB_START_KEY + 50)    // bool - notify parent proc when spawned job terminates
+#define ORTE_JOB_TRANSPORT_KEY          (ORTE_JOB_START_KEY + 51)    // string - transport keys assigned to this job
 
 #define ORTE_JOB_MAX_KEY   300
 

orte/util/pre_condition_transports.c

@@ -12,6 +12,7 @@
  * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved.
  * Copyright (c) 2015      Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
+ * Copyright (c) 2016      Intel, Inc. All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -43,6 +44,7 @@
 #include "orte/constants.h"
 #include "orte/types.h"
 #include "orte/mca/errmgr/errmgr.h"
+#include "orte/util/attr.h"
 
 #include "orte/util/pre_condition_transports.h"
 
@@ -161,6 +163,9 @@ int orte_pre_condition_transports(orte_job_t *jdata)
         return ORTE_ERR_OUT_OF_RESOURCE;
     }
 
+    /* record it in case this job executes a dynamic spawn */
+    orte_set_attribute(&jdata->attributes, ORTE_JOB_TRANSPORT_KEY, ORTE_ATTR_LOCAL, string_key, OPAL_STRING);
+
     if (OPAL_SUCCESS != mca_base_var_env_name ("orte_precondition_transports", &cs_env)) {
         ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
         free(string_key);