[svn-r6975] Added code to disable auto-cleanup of processes when requested by the

user.

src/pm/hydra/include/hydra_base.h

@@ -312,6 +312,8 @@ struct HYD_user_global {
     int enablex;
     int debug;
 
+    int auto_cleanup;
+
     struct HYD_env_global global_env;
 };
 

src/pm/hydra/pm/pmiserv/pmip_cb.c

@@ -273,15 +273,20 @@ static HYD_status pmi_cb(int fd, HYD_event_t events, void *userp)
     }
 
     if (closed) {
-        /* This is a hack to improve user-friendliness. If a PMI
-         * application terminates, we clean up the remaining
+        /* If a PMI application terminates, we clean up the remaining
          * processes. For a correct PMI application, we should never
          * get closed socket event as we deregister this socket as
          * soon as we get the finalize message. For non-PMI
          * applications, this is harder to identify, so we just let
          * the user cleanup the processes on a failure. */
-        if (using_pmi_port || HYD_pmcd_pmip.downstream.pmi_fd_active[i])
-            HYD_pmcd_pmip_killjob();
+        if (using_pmi_port || HYD_pmcd_pmip.downstream.pmi_fd_active[i]) {
+            if (HYD_pmcd_pmip.user_global.auto_cleanup)
+                HYD_pmcd_pmip_killjob();
+            else {
+                /* If the user doesn't want to automatically cleanup,
+                 * just deregister the FD and ignore this error */
+            }
+        }
         goto fn_exit;
     }
 

src/pm/hydra/pm/pmiserv/pmip_utils.c

@@ -98,6 +98,11 @@ static HYD_status enable_stdin_fn(char *arg, char ***argv)
     return HYDU_set_int_and_incr(arg, argv, &HYD_pmcd_pmip.system_global.enable_stdin);
 }
 
+static HYD_status auto_cleanup_fn(char *arg, char ***argv)
+{
+    return HYDU_set_int_and_incr(arg, argv, &HYD_pmcd_pmip.user_global.auto_cleanup);
+}
+
 static HYD_status pmi_port_fn(char *arg, char ***argv)
 {
     return HYDU_set_str_and_incr(arg, argv, &HYD_pmcd_pmip.system_global.pmi_port);
@@ -393,6 +398,7 @@ struct HYD_arg_match_table HYD_pmcd_pmip_match_table[] = {
     {"iface", iface_fn, NULL},
     {"prepend-rank", prepend_rank_fn, NULL},
     {"enable-stdin", enable_stdin_fn, NULL},
+    {"auto-cleanup", auto_cleanup_fn, NULL},
 
     /* Executable parameters */
     {"pmi-port", pmi_port_fn, NULL},

src/pm/hydra/pm/pmiserv/pmiserv_utils.c

@@ -167,6 +167,8 @@ HYD_status HYD_pmcd_pmi_fill_in_exec_launch_info(struct HYD_pg *pg)
         proxy->exec_launch_info[arg++] = HYDU_strdup("--global-process-count");
         proxy->exec_launch_info[arg++] = HYDU_int_to_str(pg->pg_process_count);
 
+        proxy->exec_launch_info[arg++] = HYDU_strdup("--auto-cleanup");
+        proxy->exec_launch_info[arg++] = HYDU_int_to_str(HYD_handle.user_global.auto_cleanup);
 
         /* Check if we are running in embedded mode */
         ret = MPL_env2str("PMI_FD", (const char **) &pmi_fd);

src/pm/hydra/ui/mpich/mpiexec.c

@@ -105,6 +105,7 @@ static void usage(void)
     printf("    -prepend-rank                    prepend rank to output\n");
     printf
         ("    -nameserver                      name server information (host:port format)\n");
+    printf("    -disable-auto-cleanup            don't cleanup processes on error\n");
 }
 
 static void signal_cb(int sig)

src/pm/hydra/ui/mpich/utils.c

@@ -840,6 +840,19 @@ static HYD_status nameserver_fn(char *arg, char ***argv)
     return HYDU_set_str_and_incr(arg, argv, &HYD_handle.nameserver);
 }
 
+static void auto_cleanup_help_fn(void)
+{
+    printf("\n");
+    printf("-disable-auto-cleanup: Don't auto-cleanup of processes when the app aborts\n");
+    printf("-enable-auto-cleanup: Auto-cleanup processes when the app aborts (default)\n\n");
+}
+
+static HYD_status auto_cleanup_fn(char *arg, char ***argv)
+{
+    return HYDU_set_int(arg, argv, &HYD_handle.user_global.auto_cleanup,
+                        !strcmp(arg, "enable-auto-cleanup"));
+}
+
 static struct HYD_arg_match_table match_table[] = {
     /* Global environment options */
     {"genv", genv_fn, genv_help_fn},
@@ -910,6 +923,8 @@ static struct HYD_arg_match_table match_table[] = {
     {"print-all-exitcodes", print_all_exitcodes_fn, print_all_exitcodes_help_fn},
     {"iface", iface_fn, iface_help_fn},
     {"nameserver", nameserver_fn, nameserver_help_fn},
+    {"disable-auto-cleanup", auto_cleanup_fn, auto_cleanup_help_fn},
+    {"enable-auto-cleanup", auto_cleanup_fn, auto_cleanup_help_fn},
 
     {"\0", NULL}
 };
@@ -961,6 +976,9 @@ static HYD_status set_default_values(void)
         HYD_handle.user_global.global_env.prop =
             !strcmp(tmp, "all") ? HYDU_strdup("all") : HYDU_strdup("none");
 
+    if (HYD_handle.user_global.auto_cleanup == -1)
+        HYD_handle.user_global.auto_cleanup = 1;
+
   fn_exit:
     return status;
 

src/pm/hydra/utils/alloc/alloc.c

@@ -24,6 +24,8 @@ void HYDU_init_user_global(struct HYD_user_global *user_global)
     user_global->debug = -1;
     user_global->prepend_rank = -1;
 
+    user_global->auto_cleanup = -1;
+
     HYDU_init_global_env(&user_global->global_env);
 }