Cleanup our handling of VMs to ensure daemons don't get mapped when operating with a VM.

Have each mapper flag it did the map so we can see who did it later.

Ensure procs are flagged as "ready to launch".

This commit was SVN r24406.

orte/mca/rmaps/base/rmaps_base_map_job.c

@@ -120,8 +120,10 @@ int orte_rmaps_base_map_job(orte_job_t *jdata)
                 return rc;
             }
         }
-        /* if we get here without doing the map, then that's an error */
+        /* if we get here without doing the map, or with zero procs in
-        if (!did_map) {
+         * the map, then that's an error
+         */
+        if (!did_map || 0 == jdata->num_procs) {
             orte_show_help("help-orte-rmaps-base.txt", "failed-map", true);
             return ORTE_ERR_FAILED_TO_MAP;
         }

orte/mca/rmaps/base/rmaps_base_support_fns.c

@@ -372,6 +372,8 @@ int orte_rmaps_base_claim_slot(orte_job_t *jdata,
         }
         /* set the jobid */
         proc->name.jobid = jdata->jobid;
+        /* flag the proc as ready for launch */
+        proc->state = ORTE_PROC_STATE_INIT;
         /* we do not set the vpid here - this will be done
          * during a second phase
          */
@@ -673,8 +675,9 @@ retry_lr:
 }
 
 
-int orte_rmaps_base_define_daemons(orte_job_map_t *map)
+int orte_rmaps_base_define_daemons(orte_job_t *jdata)
 {
+    orte_job_map_t *map;
     orte_node_t *node;
     orte_proc_t *proc;
     orte_job_t *daemons;
@@ -685,6 +688,13 @@ int orte_rmaps_base_define_daemons(orte_job_map_t *map)
                          "%s rmaps:base:define_daemons",
                          ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
 
+    if (ORTE_MAPPING_USE_VM & jdata->map->policy) {
+        /* nothing for us to do - all daemons are
+         * defined by definition!
+         */
+        return ORTE_SUCCESS;
+    }
+
     /* get the daemon job data struct */
     if (NULL == (daemons = orte_get_job_data_object(ORTE_PROC_MY_HNP->jobid))) {
         /* bad news */
@@ -693,6 +703,7 @@ int orte_rmaps_base_define_daemons(orte_job_map_t *map)
     }
     
     /* initialize the #new daemons */
+    map = jdata->map;
     map->num_new_daemons = 0;
     
     /* go through the nodes in the map, checking each one's daemon name

orte/mca/rmaps/base/rmaps_private.h

@@ -88,7 +88,7 @@ ORTE_DECLSPEC void orte_rmaps_base_update_local_ranks(orte_job_t *jdata, orte_no
 
 ORTE_DECLSPEC int orte_rmaps_base_rearrange_map(orte_app_context_t *app, orte_job_map_t *map, opal_list_t *procs);
 
-ORTE_DECLSPEC int orte_rmaps_base_define_daemons(orte_job_map_t *map);
+ORTE_DECLSPEC int orte_rmaps_base_define_daemons(orte_job_t *jdata);
 
 ORTE_DECLSPEC int orte_rmaps_base_setup_virtual_machine(orte_job_t *jdata);
 

orte/mca/rmaps/load_balance/rmaps_lb.c

@@ -71,10 +71,13 @@ static int switchyard(orte_job_t *jdata)
                             ORTE_JOBID_PRINT(jdata->jobid));
         return ORTE_ERR_TAKE_NEXT_OPTION;
     }
+
     opal_output_verbose(5, orte_rmaps_base.rmaps_output,
                         "mca:rmaps:loadbalance: mapping job %s",
                         ORTE_JOBID_PRINT(jdata->jobid));
  
+    /* flag that I did the mapping */
+    jdata->map->mapper = ORTE_RMAPS_LOADBALANCE;
 
     if (0 < orte_rmaps_base.npernode) {
         rc = npernode(jdata);
@@ -97,7 +100,7 @@ static int switchyard(orte_job_t *jdata)
     }
     
     /* define the daemons that we will use for this job */
-    if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(jdata->map))) {
+    if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(jdata))) {
         ORTE_ERROR_LOG(rc);
     }
     

orte/mca/rmaps/rank_file/rmaps_rank_file.c

@@ -308,10 +308,13 @@ static int orte_rmaps_rf_map(orte_job_t *jdata)
                             ORTE_JOBID_PRINT(jdata->jobid));
         return ORTE_ERR_TAKE_NEXT_OPTION;
     }
+
     opal_output_verbose(5, orte_rmaps_base.rmaps_output,
                         "mca:rmaps:rank_file: mapping job %s",
                         ORTE_JOBID_PRINT(jdata->jobid));
  
+    /* flag that I did the mapping */
+    jdata->map->mapper = ORTE_RMAPS_RF;
 
     /* convenience def */
     map = jdata->map;
@@ -597,7 +600,7 @@ static int orte_rmaps_rf_map(orte_job_t *jdata)
     }
 
     /* define the daemons that we will use for this job */
-    if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(map))) {
+    if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(jdata))) {
         ORTE_ERROR_LOG(rc);
         return rc;
     }

orte/mca/rmaps/resilient/rmaps_resilient.c

@@ -86,6 +86,8 @@ static int orte_rmaps_resilient_map(orte_job_t *jdata)
                         "mca:rmaps:resilient: mapping job %s",
                         ORTE_JOBID_PRINT(jdata->jobid));
  
+    /* flag that I did the mapping */
+    jdata->map->mapper = ORTE_RMAPS_RESILIENT;
 
     /* have we already constructed the fault group list? */
     if (!have_ftgrps) {
@@ -217,12 +219,10 @@ static int orte_rmaps_resilient_map(orte_job_t *jdata)
          */
         orte_rmaps_base_update_local_ranks(jdata, oldnode, nd, proc);
     }
-    if (!(ORTE_MAPPING_USE_VM & jdata->map->policy)) {
+    /* define the daemons that we will use for this job */
-        /* define the daemons that we will use for this job */
+    if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(jdata))) {
-        if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(jdata->map))) {
+        ORTE_ERROR_LOG(rc);
-            ORTE_ERROR_LOG(rc);
+        return rc;
-            return rc;
-        }
     }
 
  error:
@@ -746,12 +746,10 @@ static int map_to_ftgrps(orte_job_t *jdata)
         return rc;
     }
 
-    if (!(ORTE_MAPPING_USE_VM & jdata->map->policy)) {
+    /* define the daemons that we will use for this job */
-        /* define the daemons that we will use for this job */
+    if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(jdata))) {
-        if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(jdata->map))) {
+        ORTE_ERROR_LOG(rc);
-            ORTE_ERROR_LOG(rc);
+        return rc;
-            return rc;
-        }
     }
     
     return ORTE_SUCCESS;

orte/mca/rmaps/round_robin/rmaps_rr.c

@@ -77,6 +77,9 @@ static int orte_rmaps_rr_map(orte_job_t *jdata)
                         "mca:rmaps:rr: mapping job %s",
                         ORTE_JOBID_PRINT(jdata->jobid));
  
+    /* flag that I did the mapping */
+    jdata->map->mapper = ORTE_RMAPS_RR;
+
     /* start at the beginning... */
     jdata->num_procs = 0;
     
@@ -155,7 +158,7 @@ static int orte_rmaps_rr_map(orte_job_t *jdata)
     }
 
     /* define the daemons that we will use for this job */
-    if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(jdata->map))) {
+    if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(jdata))) {
         ORTE_ERROR_LOG(rc);
         return rc;
     }

orte/mca/rmaps/seq/rmaps_seq.c

@@ -93,10 +93,13 @@ static int orte_rmaps_seq_map(orte_job_t *jdata)
                             ORTE_JOBID_PRINT(jdata->jobid));
         return ORTE_ERR_TAKE_NEXT_OPTION;
     }
+
     opal_output_verbose(5, orte_rmaps_base.rmaps_output,
                         "mca:rmaps:seq: mapping job %s",
                         ORTE_JOBID_PRINT(jdata->jobid));
- 
+
+    /* flag that I did the mapping */
+    jdata->map->mapper = ORTE_RMAPS_SEQ;
 
     /* conveniece def */
     map = jdata->map;
@@ -247,7 +250,7 @@ static int orte_rmaps_seq_map(orte_job_t *jdata)
     }
 
     /* define the daemons that we will use for this job */
-    if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(map))) {
+    if (ORTE_SUCCESS != (rc = orte_rmaps_base_define_daemons(jdata))) {
         ORTE_ERROR_LOG(rc);
         return rc;
     }