LSF support is now working. W00t! May be subject to a further tweak

or two.

 * checking lsb_init() is not sufficient to know whether you're in an
   LSF job or not; you also need to check for environment variable
   markers 
 * remove lots of debugging output
 * no need for the sds lsf to call lsb_init()
 * remove some slurm-like dead code and a copy-n-paste error in the
   sds lsf

This commit was SVN r15644.

orte/mca/pls/lsf/pls_lsf_component.c

@@ -142,7 +142,7 @@ static orte_pls_base_module_t *pls_lsf_init(int *priority)
     }
     
     /* check if lsf is running here */
-    if (lsb_init("ORTE launcher") < 0) {
+    if (NULL == getenv("LSB_JOBID") || lsb_init("ORTE launcher") < 0) {
         /* nope, not here */
         opal_output_verbose(10, orte_pls_base.pls_output,
                             "pls:lsf: NOT available for selection");

orte/mca/pls/lsf/pls_lsf_module.c

@@ -133,7 +133,6 @@ static int pls_lsf_launch_job(orte_jobid_t jobid)
     int proc_name_index = 0;
     bool failed_launch = true;
 
-    opal_output(0, "pls lsf being used to launch!\n");
     if (mca_pls_lsf_component.timing) {
         if (0 != gettimeofday(&joblaunchstart, NULL)) {
             opal_output(0, "pls_lsf: could not obtain job start time");
@@ -179,7 +178,6 @@ static int pls_lsf_launch_job(orte_jobid_t jobid)
 #endif
     
     num_nodes = map->num_new_daemons;
-    opal_output(0, "pls lsf num new daemons: %d!\n", num_nodes);
     if (num_nodes == 0) {
         /* no new daemons required - just launch apps */
         goto launch_apps;
@@ -231,7 +229,7 @@ static int pls_lsf_launch_job(orte_jobid_t jobid)
 
     /* force orted to use the lsf sds */
     opal_argv_append(&argc, &argv, "--ns-nds");
-    opal_argv_append(&argc, &argv, "lsf");
+    opal_argv_append(&argc, &argv, "env");
 
     /* tell the new daemons the base of the name list so they can compute
      * their own name on the other end
@@ -306,16 +304,12 @@ static int pls_lsf_launch_job(orte_jobid_t jobid)
      * orterun can do the rest of its stuff. Instead, we'll catch any
      * failures and deal with them elsewhere
      */
-    opal_output(0, "launching on: %s", opal_argv_join(nodelist_argv, ' '));
-    opal_output(0, "launching: %s", opal_argv_join(argv, ' '));
     if (lsb_launch(nodelist_argv, argv, LSF_DJOB_NOWAIT, env) < 0) {
         ORTE_ERROR_LOG(ORTE_ERR_FAILED_TO_START);
         opal_output(0, "lsb_launch failed: %d", rc);
         rc = ORTE_ERR_FAILED_TO_START;
         goto cleanup;
     }
-    opal_output(0, "lsb_launch launched ok; waiting for %d daemons\n",
-                map->num_new_daemons);
     
     /* wait for daemons to callback */
     if (ORTE_SUCCESS != 
@@ -325,12 +319,6 @@ static int pls_lsf_launch_job(orte_jobid_t jobid)
     }
 
 launch_apps:
-    {
-        int i = 0;
-        opal_output(0, "waiting for attach");
-        while (i == 0) sleep(5);
-    }
-    opal_output(0, "laounching apps using lsf");
     if (ORTE_SUCCESS != (rc = orte_pls_base_launch_apps(map))) {
         ORTE_ERROR_LOG(rc);
         goto cleanup;
@@ -338,7 +326,6 @@ launch_apps:
     
     /* declare the launch a success */
     failed_launch = false;
-    opal_output(0, "launched apps with lsf ok");
     
     if (mca_pls_lsf_component.timing) {
         if (0 != gettimeofday(&launchstop, NULL)) {
@@ -382,7 +369,6 @@ cleanup:
         }        
     }
 
-    opal_output(0, "lsf pls returning: %d\n", rc);
     return rc;
 }
 

orte/mca/ras/lsf/ras_lsf_component.c

@@ -97,7 +97,7 @@ static orte_ras_base_module_t *orte_ras_lsf_init(int* priority)
     }
     
     /* check if lsf is running here */
-    if (lsb_init("ORTE launcher") < 0) {
+    if (NULL == getenv("LSB_JOBID") || lsb_init("ORTE launcher") < 0) {
         /* nope, not here */
         return NULL;
     }

orte/mca/ras/lsf/ras_lsf_module.c

@@ -77,24 +77,16 @@ static int allocate(orte_jobid_t jobid, opal_list_t *attributes)
     
     /* step through the list */
     for (count = i = 0; i < num_nodes; i++) {
-        opal_output(0, "lsf got node: %s", nodelist[i]);
         /* is this a repeat of the current node? */
         if (NULL != node && 0 == strcmp(nodelist[i], node->node_name)) {
             /* it is a repeat - just bump the slot count */
             ++node->node_slots;
-            opal_output(0, "lsf ras repeat -- slot count now %d",
-                        node->node_slots);
             continue;
         }
         
-        opal_output(0, "lsf ras new node");
         /* not a repeat - create a node entry for it */
         node = OBJ_NEW(orte_ras_node_t);
         node->node_name = strdup(nodelist[i]);
-        /* RHC: LSF does not use launch_id, so leave it invalid */
-#if 0
-        node->launch_id = count++;
-#endif
         node->node_slots_inuse = 0;
         node->node_slots_max = 0;
         node->node_slots = 1;

orte/mca/sds/lsf/sds_lsf_component.c

@@ -71,12 +71,6 @@ orte_sds_base_module_t *orte_sds_lsf_component_init(int *priority)
     int id;
     char *mode;
 
-    /* check if lsf is running here */
-    if (lsb_init("ORTE launcher") < 0) {
-        /* nope, not here */
-        return NULL;
-    }
-    
     id = mca_base_param_register_string("ns", "nds", NULL, NULL, NULL);
     mca_base_param_lookup_string(id, &mode);
 

orte/mca/sds/lsf/sds_lsf_module.c

@@ -49,15 +49,12 @@ orte_sds_base_module_t orte_sds_lsf_module = {
     orte_sds_lsf_finalize,
 };
 
-static char *get_lsf_nodename(int nodeid);
-
 
 int orte_sds_lsf_set_name(void)
 {
     int rc;
     int id;
     char* name_string = NULL;
-    int lsf_nodeid;
 
     /* start by getting our jobid, and vpid (which is the
        starting vpid for the list of daemons) */
@@ -76,17 +73,9 @@ int orte_sds_lsf_set_name(void)
     } else {
         orte_jobid_t jobid;
         orte_vpid_t vpid;
-        char* cellid_string;
         char* jobid_string;
         char* vpid_string;
       
-        id = mca_base_param_register_string("ns", "nds", "cellid", NULL, NULL);
-        mca_base_param_lookup_string(id, &cellid_string);
-        if (NULL == cellid_string) {
-            ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
-            return ORTE_ERR_NOT_FOUND;
-        }
-            
         id = mca_base_param_register_string("ns", "nds", "jobid", NULL, NULL);
         mca_base_param_lookup_string(id, &jobid_string);
         if (NULL == jobid_string) {
@@ -119,18 +108,6 @@ int orte_sds_lsf_set_name(void)
         }
     }
 
-    /* fix up the base name and make it the "real" name */
-    lsf_nodeid = atoi(getenv("LSF_PM_TASKID"));
-    orte_process_info.my_name->vpid += lsf_nodeid;
-
-#if 0
-    /* fix up the system info nodename to match exactly what lsf returned */
-    if (NULL != orte_system_info.nodename) {
-        free(orte_system_info.nodename);
-    }
-    orte_system_info.nodename = get_lsf_nodename(lsf_nodeid);
-#endif
-
     /* get the non-name common environmental variables */
     if (ORTE_SUCCESS != (rc = orte_sds_env_get())) {
         ORTE_ERROR_LOG(rc);
@@ -145,35 +122,3 @@ int orte_sds_lsf_finalize(void)
 {
     return ORTE_SUCCESS;
 }
-
-
-static char *get_lsf_nodename(int nodeid)
-{
-    char **names = NULL;
-    char *lsf_nodelist;
-    char *ret;
-
-    lsf_nodelist = getenv("OMPI_MCA_orte_lsf_nodelist");
-    
-    if (NULL == lsf_nodelist) {
-        return NULL;
-    }
-    
-    /* split the node list into an argv array */
-    names = opal_argv_split(lsf_nodelist, ',');
-    if (NULL == names) {  /* got an error */
-        return NULL;
-    }
-
-    /* check to see if there are enough entries */
-    if (nodeid > opal_argv_count(names)) {
-        return NULL;
-    }
-
-    ret = strdup(names[nodeid]);
-
-    opal_argv_free(names);
-
-    /* All done */
-    return ret;
-}