When directly launched by an RM, flag that a process is operating without daemons - i.e., standalone. Provide an error string for the new socket_not_available error. Use errmgr.abort to exit when we cannot get a socket, and ensure that the slurmd module returns the proper exit status for slurm 2.0

This commit was SVN r21868.

orte/mca/ess/slurmd/ess_slurmd_module.c

@@ -59,6 +59,7 @@
 
 static int rte_init(void);
 static int rte_finalize(void);
+static void rte_abort(int error_code, bool report) __opal_attribute_noreturn__;
 static uint8_t proc_get_locality(orte_process_name_t *proc);
 static orte_vpid_t proc_get_daemon(orte_process_name_t *proc);
 static char* proc_get_hostname(orte_process_name_t *proc);
@@ -70,7 +71,7 @@ static int update_nidmap(opal_byte_object_t *bo);
 orte_ess_base_module_t orte_ess_slurmd_module = {
     rte_init,
     rte_finalize,
-    orte_ess_base_app_abort,
+    rte_abort,
     proc_get_locality,
     proc_get_daemon,
     proc_get_hostname,
@@ -124,6 +125,9 @@ static int rte_init(void)
      * provided that info in our environment, so get it from there
      */
     
+    /* declare ourselves to be standalone - i.e., not launched by orted */
+    orte_standalone_operation = true;
+    
     /* get the slurm jobid - this will be our job family */
     envar = getenv("SLURM_JOBID");
     /* don't need to check this for NULL - if it was, we would
@@ -353,11 +357,6 @@ static int rte_init(void)
     return ORTE_SUCCESS;
     
 error:
-    if (ORTE_ERR_SOCKET_NOT_AVAILABLE == ret && slurm20) {
-        /* exit silently with a special error code for slurm 2.0 */
-        exit(108);
-    }
-    
     orte_show_help("help-orte-runtime.txt",
                    "orte_init:startup:internal-failure",
                    true, error, ORTE_ERROR_NAME(ret), ret);
@@ -384,6 +383,16 @@ static int rte_finalize(void)
     return ret;    
 }
 
+static void rte_abort(int error_code, bool report)
+{
+    if (ORTE_ERR_SOCKET_NOT_AVAILABLE == error_code && slurm20) {
+        /* exit silently with a special error code for slurm 2.0 */
+        orte_ess_base_app_abort(108, false);
+    } else {
+        orte_ess_base_app_abort(error_code, report);
+    }
+}
+
 static uint8_t proc_get_locality(orte_process_name_t *proc)
 {
     orte_nid_t *nid;

orte/mca/oob/tcp/oob_tcp.c

@@ -546,7 +546,6 @@ mca_oob_tcp_create_listen(int *target_sd, unsigned short *target_port, uint16_t
     struct sockaddr_storage inaddr;
     opal_socklen_t addrlen;
     char **ports=NULL;
-    char *ctmp;
 
     /* create a listen socket for incoming connections */
     *target_sd = socket(af_family, SOCK_STREAM, 0);
@@ -786,17 +785,15 @@ mca_oob_tcp_create_listen(int *target_sd, unsigned short *target_port, uint16_t
 
     }
     
-    /* if we reach this point, then no socket could be found in the specified
+    /* cleanup */
-     * range that was available to us, so report the error
-     */
-    ctmp = opal_argv_join(ports, ',');
-    opal_output(0, "%s oob:tcp:bind() failed - no port available in specified list:\n\t%s",
-                ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ctmp);
-    free(ctmp);
-    
-    /* cleanup and return the error */
     CLOSE_THE_SOCKET(*target_sd);
     opal_argv_free(ports);
+    if (orte_standalone_operation) {
+        /* if we are running as a standalone app - i.e., one
+         * not launched by orteds - then abort
+         */
+        orte_errmgr.abort(ORTE_ERR_SOCKET_NOT_AVAILABLE, NULL);
+    }
     return ORTE_ERR_SOCKET_NOT_AVAILABLE;
 
 

orte/runtime/orte_globals.c

@@ -60,6 +60,7 @@ bool orted_spin_flag = false;
 /* ORTE OOB port flags */
 bool orte_static_ports = false;
 char *orte_oob_static_ports = NULL;
+bool orte_standalone_operation = false;
 
 bool orte_keep_fqdn_hostnames = false;
 bool orte_show_resolved_nodenames;

orte/runtime/orte_globals.h

@@ -491,6 +491,7 @@ ORTE_DECLSPEC extern bool orted_spin_flag;
 /* ORTE OOB port flags */
 ORTE_DECLSPEC extern bool orte_static_ports;
 ORTE_DECLSPEC extern char *orte_oob_static_ports;
+ORTE_DECLSPEC extern bool orte_standalone_operation;
 
 ORTE_DECLSPEC extern bool orte_keep_fqdn_hostnames;
 ORTE_DECLSPEC extern bool orte_show_resolved_nodenames;

orte/tools/orterun/help-orterun.txt

@@ -252,6 +252,14 @@ while attempting to start process rank %lu.
 Error name: %s
 Node: %s
 
+when attempting to start process rank %lu.
+#
+[orterun:proc-socket-not-avail]
+%s was unable to start the specified application as it encountered an error:
+
+Error name: %s
+Node: %s
+
 when attempting to start process rank %lu.
 #
 [orterun:proc-failed-to-start-no-status]

orte/tools/orterun/orterun.c

@@ -1004,6 +1004,10 @@ static void dump_aborted_procs(void)
                     orte_show_help("help-orterun.txt", "orterun:proc-failed-to-start", true,
                                    orterun_basename, ORTE_ERROR_NAME(proc->exit_code), proc->node->name,
                                    (unsigned long)proc->name.vpid);
+                } else if (ORTE_ERR_SOCKET_NOT_AVAILABLE == proc->exit_code) {
+                    orte_show_help("help-orterun.txt", "orterun:proc-socket-not-avail", true,
+                                   orterun_basename, ORTE_ERROR_NAME(proc->exit_code), proc->node->name,
+                                   (unsigned long)proc->name.vpid);
                 } else {
                     orte_show_help("help-orterun.txt", "orterun:proc-failed-to-start-no-status", true,
                                    orterun_basename, proc->node->name);

orte/util/error_strings.c

@@ -118,6 +118,9 @@ orte_err2str(int errnum)
     case ORTE_ERR_SYS_LIMITS_SOCKETS:
         retval = "The system limit on number of network connections a process can open was reached";
         break;
+    case ORTE_ERR_SOCKET_NOT_AVAILABLE:
+        retval = "Unable to open a TCP socket for out-of-band communications";
+        break;
     default:
         retval = NULL;
     }