Create a more specific error code for when specific sockets are not available. Ensure that slurm 2.0 gets the expected error return if the process can't start for that reason so it can take corrective action.

This commit was SVN r21867.

orte/include/orte/constants.h

@@ -98,7 +98,8 @@ enum {
     ORTE_ERR_FAILED_TO_START                = (ORTE_ERR_BASE - 26),
     ORTE_ERR_FILE_NOT_EXECUTABLE            = (ORTE_ERR_BASE - 27),
     ORTE_ERR_HNP_COULD_NOT_START            = (ORTE_ERR_BASE - 28),
-    ORTE_ERR_SYS_LIMITS_SOCKETS             = (ORTE_ERR_BASE - 29)
+    ORTE_ERR_SYS_LIMITS_SOCKETS             = (ORTE_ERR_BASE - 29),
+    ORTE_ERR_SOCKET_NOT_AVAILABLE           = (ORTE_ERR_BASE - 30)
 };
 
 #define ORTE_ERR_MAX                      (ORTE_ERR_BASE - 100)

orte/mca/ess/slurmd/ess_slurmd_module.c

@@ -83,6 +83,7 @@ orte_ess_base_module_t orte_ess_slurmd_module = {
 
 /* Local globals */
 static bool app_init_complete;
+static bool slurm20;
 
 /****    MODULE FUNCTIONS    ****/
 
@@ -108,6 +109,7 @@ static int rte_init(void)
     
     /* init flag */
     app_init_complete = false;
+    slurm20 = false;
     
     /* run the prolog */
     if (ORTE_SUCCESS != (ret = orte_ess_base_std_prolog())) {
@@ -171,6 +173,7 @@ static int rte_init(void)
          * picked up by the OOB
          */
         orte_oob_static_ports = strdup(envar);
+        slurm20 = true;
         OPAL_OUTPUT_VERBOSE((1, orte_ess_base_output,
                              "%s using SLURM-reserved ports %s",
                              ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
@@ -350,10 +353,14 @@ 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);
-    
     return ret;
 }
 

orte/mca/oob/tcp/oob_tcp.c

@@ -797,7 +797,7 @@ mca_oob_tcp_create_listen(int *target_sd, unsigned short *target_port, uint16_t
     /* cleanup and return the error */
     CLOSE_THE_SOCKET(*target_sd);
     opal_argv_free(ports);
-    return ORTE_ERROR;
+    return ORTE_ERR_SOCKET_NOT_AVAILABLE;
 
 
 socket_binded: