Revise the launch timeout detection so we don't mistakenly declare "failed to start". Recognize that timeout is at the per-job level, and define the timeout param as a total value instead of seconds/daemon as it otherwise can get to be an enormous (and useless) number.

Resolves problems in loop_spawn where the timer was incorrectly firing and killing the overall job.

cmr=v1.7.4:reviewer=hjelmn

This commit was SVN r29661.

orte/mca/plm/base/plm_base_launch_support.c

@@ -460,16 +460,13 @@ void orte_plm_base_complete_setup(int fd, short args, void *cbdata)
 /* catch timeout to allow cmds to progress */
 static void timer_cb(int fd, short event, void *cbdata)
 {
-    orte_timer_t *tm = (orte_timer_t*)cbdata;
+    orte_job_t *jdata = (orte_job_t*)cbdata;
-    orte_job_t *jdata = (orte_job_t*)tm->payload;
 
-    if (NULL == jdata || jdata->state < ORTE_JOB_STATE_RUNNING) {
     /* declare launch failed */
     ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_FAILED_TO_START);
-    }
 
     /* free event */
-    OBJ_RELEASE(tm);
+    OBJ_RELEASE(jdata->failure_timer);
 }
 
 void orte_plm_base_launch_apps(int fd, short args, void *cbdata)
@@ -532,11 +529,24 @@ void orte_plm_base_launch_apps(int fd, short args, void *cbdata)
      */
     caddy->jdata->num_daemons_reported++;
 
-    /* setup a timer - if we don't launch within the
+    /* if requested, setup a timer - if we don't launch within the
      * defined time, then we know things have failed
      */
     if (0 < orte_startup_timeout) {
-        ORTE_DETECT_TIMEOUT(orte_startup_timeout, 1000000, 0, timer_cb, jdata);
+        OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
+                             "%s plm:base:launch defining timeout for job %s",
+                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
+                             ORTE_JOBID_PRINT(jdata->jobid)));
+        jdata->failure_timer = OBJ_NEW(orte_timer_t);
+        jdata->failure_timer->payload = jdata;
+        opal_event_evtimer_set(orte_event_base,
+                               jdata->failure_timer->ev, timer_cb,
+                               jdata);
+        opal_event_set_priority(jdata->failure_timer->ev, ORTE_ERROR_PRI);
+        jdata->failure_timer->tv.tv_sec = orte_startup_timeout;
+        jdata->failure_timer->tv.tv_usec = 0;
+        opal_event_evtimer_add(jdata->failure_timer->ev,
+                               &jdata->failure_timer->tv);
     }
 
     /* cleanup */
@@ -553,6 +563,16 @@ void orte_plm_base_post_launch(int fd, short args, void *cbdata)
     /* convenience */
     jdata = caddy->jdata;
 
+    /* if a timer was defined, cancel it */
+    if (NULL != jdata->failure_timer) {
+        opal_event_evtimer_del(jdata->failure_timer->ev);
+        OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
+                             "%s plm:base:launch deleting timeout for job %s",
+                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
+                             ORTE_JOBID_PRINT(jdata->jobid)));
+        OBJ_RELEASE(jdata->failure_timer);
+    }
+
     if (ORTE_JOB_STATE_RUNNING != caddy->job_state) {
         ORTE_FORCED_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
         OBJ_RELEASE(caddy);
@@ -594,8 +614,9 @@ void orte_plm_base_registered(int fd, short args, void *cbdata)
     jdata = caddy->jdata;
 
     OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
-                         "%s plm:base:launch registered event",
+                         "%s plm:base:launch %s registered",
-                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
+                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
+                         ORTE_JOBID_PRINT(jdata->jobid)));
 
     if (ORTE_JOB_STATE_REGISTERED != caddy->job_state) {
         OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,

orte/runtime/orte_globals.c

@@ -702,6 +702,7 @@ static void orte_job_construct(orte_job_t* job)
                             2);
     job->num_apps = 0;
     job->controls = ORTE_JOB_CONTROL_FORWARD_OUTPUT;
+    job->failure_timer = NULL;
     job->gang_launched = true;
     job->stdin_target = ORTE_VPID_INVALID;
     job->stdout_target = ORTE_JOBID_INVALID;
@@ -767,6 +768,10 @@ static void orte_job_destruct(orte_job_t* job)
     }
     OBJ_RELEASE(job->apps);
     
+    if (NULL != job->failure_timer) {
+        OBJ_RELEASE(job->failure_timer);
+    }
+
     if (NULL != job->map) {
         OBJ_RELEASE(job->map);
         job->map = NULL;

orte/runtime/orte_globals.h

@@ -388,6 +388,8 @@ typedef struct {
      * for description of supported flags
      */
     orte_job_controls_t controls;
+    /* timer event for failure detect/response if fails to launch */
+    orte_timer_t *failure_timer;
     /* flag to indicate that MPI is allowed on this job - i.e.,
      * that all members of the job are being simultaneously
      * launched

orte/runtime/orte_mca_params.c

@@ -314,7 +314,7 @@ int orte_register_params(void)
 
     orte_startup_timeout = 0;
     (void) mca_base_var_register ("orte", "orte", NULL, "startup_timeout",
-                                  "Seconds/daemon to wait for startup before declaring failed_to_start (default: 0 => do not check)",
+                                  "Seconds to wait for startup or job launch before declaring failed_to_start (default: 0 => do not check)",
                                   MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
                                   OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_READONLY,
                                   &orte_startup_timeout);