Check the return status when we forward stdin and remove the recipient when they are no longer alive

This commit was SVN r22786.

orte/mca/iof/hnp/iof_hnp.h

@@ -79,7 +79,7 @@ void orte_iof_hnp_read_local_handler(int fd, short event, void *cbdata);
 void orte_iof_hnp_stdin_cb(int fd, short event, void *cbdata);
 bool orte_iof_hnp_stdin_check(int fd);
 
-void orte_iof_hnp_send_data_to_endpoint(orte_process_name_t *host,
+int orte_iof_hnp_send_data_to_endpoint(orte_process_name_t *host,
                                        orte_process_name_t *target,
                                        orte_iof_tag_t tag,
                                        unsigned char *data, int numbytes);

orte/mca/iof/hnp/iof_hnp_read.c

@@ -85,7 +85,7 @@ void orte_iof_hnp_read_local_handler(int fd, short event, void *cbdata)
     orte_iof_read_event_t *rev = (orte_iof_read_event_t*)cbdata;
     unsigned char data[ORTE_IOF_BASE_MSG_MAX];
     int32_t numbytes;
-    opal_list_item_t *item;
+    opal_list_item_t *item, *prev_item;
     orte_iof_proc_t *proct;
     int rc;
     
@@ -182,7 +182,15 @@ void orte_iof_hnp_read_local_handler(int fd, short event, void *cbdata)
                  * sent - this will tell the daemon to close
                  * the fd for stdin to that proc
                  */
-                orte_iof_hnp_send_data_to_endpoint(&sink->daemon, &sink->name, ORTE_IOF_STDIN, data, numbytes);
+                if( ORTE_SUCCESS != (rc = orte_iof_hnp_send_data_to_endpoint(&sink->daemon, &sink->name, ORTE_IOF_STDIN, data, numbytes))) {
+                    /* if the addressee is unknown, remove the sink from the list */
+                    if( ORTE_ERR_ADDRESSEE_UNKNOWN == rc ) {
+                        prev_item = opal_list_get_prev(item);
+                        opal_list_remove_item(&mca_iof_hnp_component.sinks, item);
+                        OBJ_RELEASE(item);
+                        item = prev_item;
+                    }
+                }
             }
         }
         /* if num_bytes was zero, then we need to terminate the event */

orte/mca/iof/hnp/iof_hnp_send.c

@@ -51,7 +51,7 @@ static void send_cb(int status, orte_process_name_t *peer,
     OBJ_RELEASE(buf);
 }
 
-void orte_iof_hnp_send_data_to_endpoint(orte_process_name_t *host,
+int orte_iof_hnp_send_data_to_endpoint(orte_process_name_t *host,
                                        orte_process_name_t *target,
                                        orte_iof_tag_t tag,
                                        unsigned char *data, int numbytes)
@@ -67,7 +67,7 @@ void orte_iof_hnp_send_data_to_endpoint(orte_process_name_t *host,
     if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, &tag, 1, ORTE_IOF_TAG))) {
         ORTE_ERROR_LOG(rc);
         OBJ_RELEASE(buf);
-        return;
+        return rc;
     }
     /* pack the name of the target - this is either the intended
      * recipient (if the tag is stdin and we are sending to a daemon),
@@ -76,7 +76,7 @@ void orte_iof_hnp_send_data_to_endpoint(orte_process_name_t *host,
     if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, target, 1, ORTE_NAME))) {
         ORTE_ERROR_LOG(rc);
         OBJ_RELEASE(buf);
-        return;
+        return rc;
     }
     
     /* if data is NULL, then we are done */
@@ -85,7 +85,7 @@ void orte_iof_hnp_send_data_to_endpoint(orte_process_name_t *host,
         if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, data, numbytes, OPAL_BYTE))) {
             ORTE_ERROR_LOG(rc);
             OBJ_RELEASE(buf);
-            return;
+            return rc;
         }
     }
     
@@ -95,12 +95,17 @@ void orte_iof_hnp_send_data_to_endpoint(orte_process_name_t *host,
         /* xcast this to everyone - the local daemons will know how to handle it */
         orte_grpcomm.xcast(ORTE_PROC_MY_NAME->jobid, buf, ORTE_RML_TAG_IOF_PROXY);
         OBJ_RELEASE(buf);
-        return;
+        return ORTE_SUCCESS;
     }
     
     /* send the buffer to the host - this is either a daemon or
      * a tool that requested IOF
      */
-    orte_rml.send_buffer_nb(host, buf, ORTE_RML_TAG_IOF_PROXY,
+    if( ORTE_SUCCESS != (rc = orte_rml.send_buffer_nb(host, buf, ORTE_RML_TAG_IOF_PROXY,
-                            0, send_cb, NULL);
+                                                      0, send_cb, NULL))) {
+        ORTE_ERROR_LOG(rc);
+        return rc;
+    }
+
+    return ORTE_SUCCESS;
 }