Ensure that the session directory fields of orte_process_info have been initialized prior to cleaning up those directories as part of the initialization process that deals with stale session directory trees.

Fixes trac:4534

cmr=v1.8.1:reviewer=jsquyres

This commit was SVN r31421.

The following Trac tickets were found above:
  Ticket 4534 --> https://svn.open-mpi.org/trac/ompi/ticket/4534

orte/mca/ess/base/ess_base_std_orted.c

@@ -112,11 +112,6 @@ int orte_ess_base_orted_setup(char **hosts)
 
     plm_in_use = false;
 
-    /* clear the session directory just in case there are
-     * stale directories laying around
-     */
-    orte_session_dir_cleanup(ORTE_JOBID_WILDCARD);
-
     /* setup callback for SIGPIPE */
     setup_sighandler(SIGPIPE, &epipe_handler, epipe_signal_callback);
     /* Set signal handlers to catch kill signals so we can properly clean up
@@ -385,6 +380,23 @@ int orte_ess_base_orted_setup(char **hosts)
                              (NULL == orte_process_info.tmpdir_base) ? "UNDEF" : orte_process_info.tmpdir_base,
                              orte_process_info.nodename));
         
+        /* take a pass thru the session directory code to fillin the
+         * tmpdir names - don't create anything yet
+         */
+        if (ORTE_SUCCESS != (ret = orte_session_dir(false,
+                                                    orte_process_info.tmpdir_base,
+                                                    orte_process_info.nodename, NULL,
+                                                    ORTE_PROC_MY_NAME))) {
+            ORTE_ERROR_LOG(ret);
+            error = "orte_session_dir define";
+            goto error;
+        }
+        /* clear the session directory just in case there are
+         * stale directories laying around
+         */
+        orte_session_dir_cleanup(ORTE_JOBID_WILDCARD);
+
+        /* now actually create the directory tree */
         if (ORTE_SUCCESS != (ret = orte_session_dir(true,
                                                     orte_process_info.tmpdir_base,
                                                     orte_process_info.nodename, NULL,

orte/mca/ess/hnp/ess_hnp_module.c

@@ -301,11 +301,6 @@ static int rte_init(void)
     }
     /* Setup the communication infrastructure */
     
-    /* clear the session directory just in case there are
-     * stale directories laying around
-     */
-    orte_session_dir_cleanup(ORTE_JOBID_WILDCARD);
-
     /*
      * OOB Layer
      */
@@ -596,6 +591,23 @@ static int rte_init(void)
                              (NULL == orte_process_info.tmpdir_base) ? "UNDEF" : orte_process_info.tmpdir_base,
                              orte_process_info.nodename));
         
+        /* take a pass thru the session directory code to fillin the
+         * tmpdir names - don't create anything yet
+         */
+        if (ORTE_SUCCESS != (ret = orte_session_dir(false,
+                                                    orte_process_info.tmpdir_base,
+                                                    orte_process_info.nodename, NULL,
+                                                    ORTE_PROC_MY_NAME))) {
+            ORTE_ERROR_LOG(ret);
+            error = "orte_session_dir define";
+            goto error;
+        }
+        /* clear the session directory just in case there are
+         * stale directories laying around
+         */
+        orte_session_dir_cleanup(ORTE_JOBID_WILDCARD);
+
+        /* now actually create the directory tree */
         if (ORTE_SUCCESS != (ret = orte_session_dir(true,
                                                     orte_process_info.tmpdir_base,
                                                     orte_process_info.nodename, NULL,

orte/mca/ess/singleton/ess_singleton_module.c

@@ -167,6 +167,16 @@ static int rte_init(void)
     /* flag that we are not routing since we have no HNP */
     orte_routing_is_enabled = false;
 
+    /* take a pass thru the session directory code to fillin the
+     * tmpdir names - don't create anything yet
+     */
+    if (ORTE_SUCCESS != (rc = orte_session_dir(false,
+                                               orte_process_info.tmpdir_base,
+                                               orte_process_info.nodename, NULL,
+                                               ORTE_PROC_MY_NAME))) {
+        ORTE_ERROR_LOG(rc);
+        return rc;
+    }
     /* clear the session directory just in case there are
      * stale directories laying around
      */

orte/util/session_dir.c

@@ -379,11 +379,11 @@ int orte_session_dir(bool create,
     /*
      * Get the session directory full name
      */
-    if( ORTE_SUCCESS != ( rc = orte_session_dir_get_name(&fulldirpath, 
+    if (ORTE_SUCCESS != (rc = orte_session_dir_get_name(&fulldirpath, 
                                                          &local_prefix,
                                                          &frontend,
                                                          hostid, 
-                                                         batchid, proc) ) ) {
+                                                         batchid, proc))) {
         if (ORTE_ERR_FATAL == rc) {
             /* this indicates we should abort quietly */
             rc = ORTE_ERR_SILENT;
@@ -403,45 +403,17 @@ int orte_session_dir(bool create,
             goto cleanup;
         }
     }
-    /*
-     * if we are not creating, then just verify that the path is OK
-     */
-    else {
-        if( ORTE_SUCCESS != (rc = opal_os_dirpath_access(fulldirpath, 0) )) {
-            /* it is okay for the path not to be found - don't error
-             * log that case, but do error log others
-             */
-            if (ORTE_ERR_NOT_FOUND != rc) {
-                ORTE_ERROR_LOG(rc);
-            }
-            goto cleanup;
-        }
-    }
     
-    /*
+    /* update global structure fields */
-     * If we are creating the directory tree, then force overwrite of the
-     * global structure fields
-     */
-    if (create) {
     if (NULL != orte_process_info.tmpdir_base) {
         free(orte_process_info.tmpdir_base);
-    	    orte_process_info.tmpdir_base = NULL;
     }
+    orte_process_info.tmpdir_base = strdup(local_prefix);
     if (NULL != orte_process_info.top_session_dir) {
         free(orte_process_info.top_session_dir);
         orte_process_info.top_session_dir = NULL;
     }
-    }
+    if (NULL != frontend) {
-    
-    /* 
-     * Update some of the global structures if they are empty
-     */
-    if (NULL == orte_process_info.tmpdir_base) {
-        orte_process_info.tmpdir_base = strdup(local_prefix);
-    }
-    
-    if (NULL == orte_process_info.top_session_dir &&
-        NULL != frontend) {
         orte_process_info.top_session_dir = strdup(frontend);
     }
 
@@ -449,15 +421,10 @@ int orte_session_dir(bool create,
      * Set the process session directory
      */
     if (ORTE_VPID_INVALID != proc->vpid) {
-    	if (create) { /* overwrite if creating */
         if (NULL != orte_process_info.proc_session_dir) {
             free(orte_process_info.proc_session_dir);
-                orte_process_info.proc_session_dir = NULL;
     	}
-    	}
-    	if (NULL == orte_process_info.proc_session_dir) {
         orte_process_info.proc_session_dir = strdup(fulldirpath);
-    	}
         
         /* Strip off last part of directory structure */
         sav = opal_dirname(fulldirpath);
@@ -470,16 +437,11 @@ int orte_session_dir(bool create,
      * Set the job session directory
      */
     if (ORTE_JOBID_INVALID != proc->jobid) {
-    	if (create) { /* overwrite if creating */
         if (NULL != orte_process_info.job_session_dir) {
             free(orte_process_info.job_session_dir);
-                orte_process_info.job_session_dir = NULL;
         }
-    	}
-    	if (NULL == orte_process_info.job_session_dir) {
         orte_process_info.job_session_dir = strdup(fulldirpath);
     }
-    }
     
     if (orte_debug_flag) {
     	opal_output(0, "procdir: %s", 
@@ -551,24 +513,32 @@ orte_session_dir_cleanup(orte_jobid_t jobid)
 
     if (NULL != job_session_dir && opal_os_dirpath_is_empty(job_session_dir)) {
     	if (orte_debug_flag) {
-    	    opal_output(0, "sess_dir_finalize: found job session dir empty - deleting");
+    	    opal_output(0, "sess_dir_cleanup: found job session dir empty - deleting");
     	}
     	rmdir(job_session_dir);
     } else {
     	if (orte_debug_flag) {
-    	    opal_output(0, "sess_dir_finalize: job session dir not empty - leaving");
+            if (OPAL_ERR_NOT_FOUND == opal_os_dirpath_access(job_session_dir, 0)) {
+                opal_output(0, "sess_dir_cleanup: job session dir does not exist");
+            } else {
+                opal_output(0, "sess_dir_cleanup: job session dir not empty - leaving");
+            }
     	}
         goto CLEANUP;
     }
 
     if (opal_os_dirpath_is_empty(tmp)) {
     	if (orte_debug_flag) {
-    	    opal_output(0, "sess_dir_finalize: found top session dir empty - deleting");
+    	    opal_output(0, "sess_dir_cleanup: found top session dir empty - deleting");
     	}
     	rmdir(tmp);
     } else {
     	if (orte_debug_flag) {
-    	    opal_output(0, "sess_dir_finalize: top session dir not empty - leaving");
+            if (OPAL_ERR_NOT_FOUND == opal_os_dirpath_access(tmp, 0)) {
+                opal_output(0, "sess_dir_cleanup: top session dir does not exist");
+            } else {
+                opal_output(0, "sess_dir_cleanup: top session dir not empty - leaving");
+            }
     	}
     }
 
@@ -631,8 +601,12 @@ orte_session_dir_finalize(orte_process_name_t *proc)
     	rmdir(proc_session_dir);
     } else {
     	if (orte_debug_flag) {
+            if (OPAL_ERR_NOT_FOUND == opal_os_dirpath_access(proc_session_dir, 0)) {
+                opal_output(0, "sess_dir_finalize: proc session dir does not exist");
+            } else {
                 opal_output(0, "sess_dir_finalize: proc session dir not empty - leaving");
             }
+    	}
         goto CLEANUP;
     }
 
@@ -643,8 +617,12 @@ orte_session_dir_finalize(orte_process_name_t *proc)
     	rmdir(job_session_dir);
     } else {
     	if (orte_debug_flag) {
+            if (OPAL_ERR_NOT_FOUND == opal_os_dirpath_access(job_session_dir, 0)) {
+                opal_output(0, "sess_dir_finalize: job session dir does not exist");
+            } else {
                 opal_output(0, "sess_dir_finalize: job session dir not empty - leaving");
             }
+    	}
         goto CLEANUP;
     }
 
@@ -655,9 +633,13 @@ orte_session_dir_finalize(orte_process_name_t *proc)
     	rmdir(tmp);
     } else {
     	if (orte_debug_flag) {
+            if (OPAL_ERR_NOT_FOUND == opal_os_dirpath_access(tmp, 0)) {
+                opal_output(0, "sess_dir_finalize: top session dir does not exist");
+            } else {
                 opal_output(0, "sess_dir_finalize: top session dir not empty - leaving");
             }
     	}
+    }
 
 CLEANUP:
     free(tmp);