Extend the daemon reuse functionality to *most* of the other environments.

Note that Bproc won't support this operation, so we just ignore the --reuse-daemons directive.

I'm afraid I don't understand the POE and XGrid environments well enough to attempt the necessary modifications.

Also, please note that XGrid support has been broken on the trunk. I don't understand the code syntax well enough to make the required changes to that PLS component, so it won't compile at the moment. I'm hoping Brian has a few minutes to fix it after SC.

This commit was SVN r12614.

orte/mca/pls/base/pls_base_dmn_registry_fns.c

@@ -149,8 +149,19 @@ static int get_daemons(opal_list_t *daemons, orte_jobid_t job)
     orte_pls_daemon_info_t *dmn, *dmn2;
     bool found_name, found_node, found_cell;
     opal_list_item_t *item;
+    bool check_dups;
     int rc;
 
+    /* check the list to see if there is anything already on it. If there is, then
+     * we will need to check for duplicate entries before we add something. If not,
+     * then this can go a lot faster
+     */
+    if (0 < opal_list_get_size(daemons)) {
+        check_dups = true;
+    } else {
+        check_dups = false;
+    }
+    
     /* setup the key */
     if (ORTE_SUCCESS != (rc = orte_ns.convert_jobid_to_string(&jobid_string, job))) {
         ORTE_ERROR_LOG(rc);
@@ -208,15 +219,17 @@ static int get_daemons(opal_list_t *daemons, orte_jobid_t job)
         }
         /* if we found everything, then this is a valid entry */
         if (found_name && found_node && found_cell) {
-            /* see if this daemon is already on the list - if so, then we don't add it */
-            for (item = opal_list_get_first(daemons);
-                 item != opal_list_get_end(daemons);
-                 item = opal_list_get_next(item)) {
-                dmn2 = (orte_pls_daemon_info_t*)item;
-                
-                if (ORTE_EQUAL == orte_dss.compare(dmn2->name, name, ORTE_NAME)) {
-                    /* already on list - ignore it */
-                    goto MOVEON;
+            if (check_dups) {
+                /* see if this daemon is already on the list - if so, then we don't add it */
+                for (item = opal_list_get_first(daemons);
+                     item != opal_list_get_end(daemons);
+                     item = opal_list_get_next(item)) {
+                    dmn2 = (orte_pls_daemon_info_t*)item;
+                    
+                    if (ORTE_EQUAL == orte_dss.compare(dmn2->name, name, ORTE_NAME)) {
+                        /* already on list - ignore it */
+                        goto MOVEON;
+                    }
                 }
             }
             dmn = OBJ_NEW(orte_pls_daemon_info_t);
@@ -316,22 +329,40 @@ int orte_pls_base_remove_daemon(orte_pls_daemon_info_t *info)
 int orte_pls_base_check_avail_daemons(opal_list_t *daemons,
                                       orte_jobid_t job)
 {
-    orte_jobid_t parent;
+    orte_jobid_t root, *descendants;
+    orte_std_cntr_t i, ndesc;
     int rc;
     
-    /* check for daemons belonging to the parent job */
-    if (ORTE_SUCCESS != (rc = orte_ns.get_parent_job(&parent, job))) {
+    /* check for daemons belonging to any job in this job's family.
+     * Since the jobs in any family must exit together, it is reasonable
+     * for us to reuse any daemons that were spawned by any member
+     * of our extended family. We can find all of our family members
+     * by first finding our root job, and then getting all of its
+     * descendants
+     */
+    if (ORTE_SUCCESS != (rc = orte_ns.get_root_job(&root, job))) {
         ORTE_ERROR_LOG(rc);
         return rc;
     }
     
-    if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(daemons, parent, NULL))) {
+    if (ORTE_SUCCESS != (rc = orte_ns.get_job_descendants(&descendants, &ndesc, root))) {
         ORTE_ERROR_LOG(rc);
         return rc;
     }
     
+    /* loop through the descendants, adding to the daemon list as we go */
+    for (i=0; i < ndesc; i++) {
+        if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(daemons, descendants[i], NULL))) {
+            ORTE_ERROR_LOG(rc);
+            free(descendants);
+            return rc;
+        }
+    }
+    free(descendants);  /* all done with these */
+    
     /* now add in any persistent daemons - they are tagged as bootproxies
-     * for jobid = 0 */
+     * for jobid = 0
+     */
     if (ORTE_SUCCESS != (rc = orte_pls_base_get_active_daemons(daemons, 0, NULL))) {
         ORTE_ERROR_LOG(rc);
         return rc;

orte/mca/pls/gridengine/pls_gridengine_module.c

@@ -85,6 +85,7 @@
 #include "orte/mca/smr/smr.h"
 
 #include "orte/mca/pls/pls.h"
+#include "orte/mca/pls/base/base.h"
 #include "orte/mca/pls/base/pls_private.h"
 #include "orte/mca/pls/gridengine/pls_gridengine.h"
 
@@ -226,19 +227,33 @@ int orte_pls_gridengine_launch_job(orte_jobid_t jobid)
     rc = orte_rmaps.get_job_map(&map, jobid);
     if (ORTE_SUCCESS != rc) {
         ORTE_ERROR_LOG(rc);
-        goto cleanup;
+        OBJ_DESTRUCT(&daemons);
+        return rc;
     }
 
+    /* if the user requested that we re-use daemons,
+     * launch the procs on any existing, re-usable daemons
+     */
+    if (orte_pls_base.reuse_daemons) {
+        if (ORTE_SUCCESS != (rc = orte_pls_base_launch_on_existing_daemons(map, jobid))) {
+            ORTE_ERROR_LOG(rc);
+            OBJ_RELEASE(map);
+            OBJ_DESTRUCT(&daemons);
+            return rc;
+        }
+    }
+    
     num_nodes = (orte_std_cntr_t)opal_list_get_size(&map->nodes);
-
+    if (num_nodes == 0) {
+        /* job must have been launched on existing daemons - just return */
+        OBJ_RELEASE(map);
+        OBJ_DESTRUCT(&daemons);
+        return ORTE_SUCCESS;
+    }
+    
     /*
      * Allocate a range of vpids for the daemons.
      */
-    if (num_nodes == 0) {
-        rc = ORTE_ERR_BAD_PARAM;
-        ORTE_ERROR_LOG(rc);
-        goto cleanup;
-    }
     rc = orte_ns.reserve_range(0, num_nodes, &vpid);
     if (ORTE_SUCCESS != rc) {
         ORTE_ERROR_LOG(rc);

orte/mca/pls/rsh/pls_rsh_module.c

@@ -479,23 +479,29 @@ int orte_pls_rsh_launch(orte_jobid_t jobid)
      */
     rc = orte_rmaps.get_job_map(&map, jobid);
     if (ORTE_SUCCESS != rc) {
-        goto cleanup;
+        ORTE_ERROR_LOG(rc);
+        OBJ_DESTRUCT(&active_daemons);
+        return rc;
     }
 
     /* if the user requested that we re-use daemons,
-     * launch the procs on any existing, re-usable daemons */
+     * launch the procs on any existing, re-usable daemons
+     */
     if (orte_pls_base.reuse_daemons) {
         if (ORTE_SUCCESS != (rc = orte_pls_base_launch_on_existing_daemons(map, jobid))) {
             ORTE_ERROR_LOG(rc);
-            goto cleanup;
+            OBJ_RELEASE(map);
+            OBJ_DESTRUCT(&active_daemons);
+            return rc;
         }
     }
     
     num_nodes = (orte_std_cntr_t)opal_list_get_size(&map->nodes);
-    if (0 >= num_nodes) {
+    if (0 == num_nodes) {
         /* nothing left to do - just return */
-        rc = ORTE_SUCCESS;
-        goto cleanup;
+        OBJ_RELEASE(map);
+        OBJ_DESTRUCT(&active_daemons);
+        return ORTE_SUCCESS;
     }
 
     if (mca_pls_rsh_component.debug_daemons &&

orte/mca/pls/slurm/pls_slurm_module.c

@@ -65,6 +65,7 @@
 #include "orte/mca/rmaps/rmaps.h"
 
 #include "orte/mca/pls/pls.h"
+#include "orte/mca/pls/base/base.h"
 #include "orte/mca/pls/base/pls_private.h"
 #include "pls_slurm.h"
 
@@ -156,15 +157,34 @@ static int pls_slurm_launch_job(orte_jobid_t jobid)
      */
     rc = orte_rmaps.get_job_map(&map, jobid);
     if (ORTE_SUCCESS != rc) {
-        goto cleanup;
+        ORTE_ERROR_LOG(rc);
+        OBJ_DESTRUCT(&daemons);
+        return rc;
     }
 
+    /* if the user requested that we re-use daemons,
+     * launch the procs on any existing, re-usable daemons
+     */
+    if (orte_pls_base.reuse_daemons) {
+        if (ORTE_SUCCESS != (rc = orte_pls_base_launch_on_existing_daemons(map, jobid))) {
+            ORTE_ERROR_LOG(rc);
+            OBJ_RELEASE(map);
+            OBJ_DESTRUCT(&daemons);
+            return rc;
+        }
+    }
+    
     /*
      * Allocate a range of vpids for the daemons.
      */
     num_nodes = opal_list_get_size(&map->nodes);
     if (num_nodes == 0) {
-        return ORTE_ERR_BAD_PARAM;
+        /* nothing further to do - job must have been launched
+         * on existing daemons, so we can just return
+         */
+        OBJ_RELEASE(map);
+        OBJ_DESTRUCT(&daemons);
+        return ORTE_SUCCESS;
     }
     rc = orte_ns.reserve_range(0, num_nodes, &vpid);
     if (ORTE_SUCCESS != rc) {

orte/mca/pls/tm/pls_tm_module.c

@@ -69,7 +69,7 @@
 #include "orte/mca/rml/rml.h"
 #include "orte/mca/ns/ns.h"
 
-
+#include "orte/mca/pls/base/base.h"
 #include "orte/mca/pls/base/pls_private.h"
 #include "pls_tm.h"
 
@@ -164,17 +164,31 @@ static int pls_tm_launch_job(orte_jobid_t jobid)
      */
     rc = orte_rmaps.get_job_map(&map, jobid);
     if (ORTE_SUCCESS != rc) {
-        goto cleanup;
+        ORTE_ERROR_LOG(rc);
+        return rc;
     }
 
+    /* if the user requested that we re-use daemons,
+     * launch the procs on any existing, re-usable daemons
+     */
+    if (orte_pls_base.reuse_daemons) {
+        if (ORTE_SUCCESS != (rc = orte_pls_base_launch_on_existing_daemons(map, jobid))) {
+            ORTE_ERROR_LOG(rc);
+            OBJ_RELEASE(map);
+            return rc;
+        }
+    }
+    
     num_nodes = opal_list_get_size(&map->nodes);
-
+    if (0 == num_nodes) {
+        /* must have been launched on existing daemons - just return */
+        OBJ_RELEASE(map);
+        return ORTE_SUCCESS;
+    }
+    
     /*
      * Allocate a range of vpids for the daemons.
      */
-    if (0 == num_nodes) {
-        return ORTE_ERR_BAD_PARAM;
-    }
     rc = orte_ns.reserve_range(0, num_nodes, &vpid);
     if (ORTE_SUCCESS != rc) {
         goto cleanup;