cleanup

This commit was SVN r7414.

orte/mca/pls/bproc/pls_bproc.c

@@ -275,36 +275,6 @@ static void orte_pls_bproc_waitpid_cb(pid_t wpid, int status, void *data) {
         opal_output(0, "in orte_pls_bproc_waitpid_cb, %d processes left\n", 
                 mca_pls_bproc_component.num_procs);
     }
-    if(0 == mca_pls_bproc_component.num_procs && 
-            mca_pls_bproc_component.done_launching) {
-        orte_buffer_t ack;
-        size_t i;
-
-        OBJ_CONSTRUCT(&ack, orte_buffer_t);
-        rc = orte_dps.pack(&ack, &i, 1, ORTE_BYTE);
-        if(rc != ORTE_SUCCESS) {
-            ORTE_ERROR_LOG(rc);
-        }
-        for(i = 0; i < mca_pls_bproc_component.num_daemons; i++) {
-            proc = orte_pointer_array_get_item(mca_pls_bproc_component.daemon_names, i);
-            if(NULL == proc) {
-                ORTE_ERROR_LOG(rc);
-                continue;
-            }
-            rc = mca_oob_send_packed(proc, &ack, MCA_OOB_TAG_BPROC, 0);
-            if (0 > rc) {
-                ORTE_ERROR_LOG(rc);
-            }
-            free(proc);
-        }
-        OBJ_DESTRUCT(&ack);
-        while(0 < mca_pls_bproc_component.num_daemons) {
-             opal_condition_wait(&mca_pls_bproc_component.condition,
-                                 &mca_pls_bproc_component.lock);
-        } 
-        mca_pls_bproc_component.done_launching = false; 
-        orte_pointer_array_clear(mca_pls_bproc_component.daemon_names); 
-    }
     OPAL_THREAD_UNLOCK(&mca_pls_bproc_component.lock);
 }
  
@@ -586,10 +556,12 @@ static int orte_pls_bproc_launch_daemons(orte_cellid_t cellid, char *** envp,
     }
     
     /* launch the daemons */
+    mca_pls_bproc_component.num_daemons = num_daemons;
     rc = bproc_vexecmove(num_daemons, daemon_list, pids, orted_path, argv, *envp);
     if(rc != num_daemons) {
         opal_show_help("help-pls-bproc.txt", "daemon-launch-number", true, 
                        num_daemons, rc, orted_path);
+        mca_pls_bproc_component.num_daemons = 0;
         rc = ORTE_ERROR;
         goto cleanup;
     }
@@ -637,7 +609,6 @@ static int orte_pls_bproc_launch_daemons(orte_cellid_t cellid, char *** envp,
             }
         }
     }
-    mca_pls_bproc_component.num_daemons = num_daemons;
 
 cleanup:
     if(NULL != argv) {
@@ -649,7 +620,6 @@ cleanup:
     if(NULL != orted_path) {
         free(orted_path);
     }
-    
     return rc;
 }
 
@@ -798,7 +768,6 @@ cleanup:
 int orte_pls_bproc_launch(orte_jobid_t jobid) {
     opal_list_item_t* item;
     opal_list_t mapping;
-    orte_buffer_t ack;
     orte_cellid_t cellid;
     orte_rmaps_base_map_t* map;
     orte_vpid_t vpid_launch;
@@ -812,8 +781,6 @@ int orte_pls_bproc_launch(orte_jobid_t jobid) {
     int context = 0;
     size_t idx, j;
 
-    OBJ_CONSTRUCT(&ack, orte_buffer_t);
-
     /* query for the application context and allocated nodes */
     OBJ_CONSTRUCT(&mapping, opal_list_t);
     if(ORTE_SUCCESS != (rc = orte_rmaps_base_get_map(jobid, &mapping))) {
@@ -871,9 +838,12 @@ int orte_pls_bproc_launch(orte_jobid_t jobid) {
      * sucessfully set up the pty/pipes and IO forwarding which the user apps
      * will use  */
     for(j = 0; j < mca_pls_bproc_component.num_daemons; j++) {
+        orte_buffer_t ack;
+        OBJ_CONSTRUCT(&ack, orte_buffer_t);
         rc = mca_oob_recv_packed(MCA_OOB_NAME_ANY, &ack, MCA_OOB_TAG_BPROC);
         if(0 > rc) {
             ORTE_ERROR_LOG(rc);
+            OBJ_DESTRUCT(&ack);
             goto cleanup;
         }
         idx = 4;
@@ -881,6 +851,8 @@ int orte_pls_bproc_launch(orte_jobid_t jobid) {
         if(ORTE_SUCCESS != rc) {
             ORTE_ERROR_LOG(rc);
         }
+        OBJ_DESTRUCT(&ack);
+
         if(-1 == src[0]) { 
             /* one of the daemons has failed to properly launch. The error is sent
              * by orte_pls_bproc_waitpid_daemon_cb  */
@@ -929,7 +901,6 @@ cleanup:
         free(node_array_len);
     }
     OBJ_DESTRUCT(&mapping);
-    OBJ_DESTRUCT(&ack);
     return rc;
 }
 

orte/mca/pls/bproc_orted/pls_bproc_orted.c

@@ -68,8 +68,6 @@ static int pls_bproc_orted_link_pipes(int proc_rank, orte_jobid_t jobid, int * f
                                       bool connect_stdin, size_t app_context);
 static void pls_bproc_orted_delete_dir_tree(char * path);
 static int pls_bproc_orted_remove_dir(void);
-static void pls_bproc_orted_kill_cb(int status, orte_process_name_t * peer,
-                                    orte_buffer_t* buffer, int tag, void* cbdata);
 static void pls_bproc_orted_send_cb(int status, orte_process_name_t * peer,
                                     orte_buffer_t* buffer, int tag, void* cbdata);
 
@@ -357,22 +355,6 @@ static int pls_bproc_orted_remove_dir() {
     return ORTE_SUCCESS;
 }
 
-/**
- * Callback function for when mpirun sends us a message saying all the child 
- * procs are done.
- * @param status
- * @param peer
- * @param buffer
- * @param tag
- * @param cbdata 
- */
-static void pls_bproc_orted_kill_cb(int status, orte_process_name_t * peer,
-                                    orte_buffer_t* buffer, int tag, void* cbdata) {
-    OPAL_THREAD_LOCK(&mca_pls_bproc_orted_component.lock);
-    opal_condition_signal(&mca_pls_bproc_orted_component.condition);
-    OPAL_THREAD_UNLOCK(&mca_pls_bproc_orted_component.lock);
-}
-
 /**
  * Callback function for when we tell mpirun we are ready
  * @param status
@@ -382,7 +364,9 @@ static void pls_bproc_orted_kill_cb(int status, orte_process_name_t * peer,
  * @param cbdata 
  */
 static void pls_bproc_orted_send_cb(int status, orte_process_name_t * peer,
-                                    orte_buffer_t* buffer, int tag, void* cbdata) {
+                                    orte_buffer_t* buffer, int tag, void* cbdata) 
+{
+    OBJ_RELEASE(buffer);
 }
 
 /**
@@ -402,7 +386,7 @@ int orte_pls_bproc_orted_launch(orte_jobid_t jobid) {
     int num_procs = 0;
     size_t i;
     int src = 0;
-    orte_buffer_t ack;
+    orte_buffer_t *ack;
     char * param;
     bool connect_stdin;
     char * pty_name = NULL;
@@ -413,7 +397,6 @@ int orte_pls_bproc_orted_launch(orte_jobid_t jobid) {
         goto cleanup;
     }
 #endif
-    OBJ_CONSTRUCT(&ack, orte_buffer_t);
 
     rc = bproc_currnode();
     if(0 > rc) {
@@ -503,20 +486,14 @@ int orte_pls_bproc_orted_launch(orte_jobid_t jobid) {
         }
     }
     
-    /* post recieve for termination signal */
+    /* message to indicate that we are ready */
-    rc = mca_oob_recv_packed_nb(MCA_OOB_NAME_SEED, MCA_OOB_TAG_BPROC, 0, 
+    ack = OBJ_NEW(orte_buffer_t);
-                                pls_bproc_orted_kill_cb, NULL);
+    rc = orte_dps.pack(ack, &src, 1, ORTE_INT);
-    if (0 > rc) {
-        ORTE_ERROR_LOG(rc);
-        goto cleanup;
-    }
-    /* do callback to say we are ready */
-    rc = orte_dps.pack(&ack, &src, 1, ORTE_INT);
     if(ORTE_SUCCESS != rc) {
         ORTE_ERROR_LOG(rc);
     }
-    rc = mca_oob_send_packed_nb(MCA_OOB_NAME_SEED, &ack, MCA_OOB_TAG_BPROC, 0, 
+    rc = mca_oob_send_packed_nb(MCA_OOB_NAME_SEED, ack, MCA_OOB_TAG_BPROC, 0, 
-                                pls_bproc_orted_send_cb, NULL);
+        pls_bproc_orted_send_cb, NULL);
     if (0 > rc) {
         ORTE_ERROR_LOG(rc);
         goto cleanup;
@@ -531,7 +508,6 @@ cleanup:
         free(pty_name);
     }
     OBJ_DESTRUCT(&map);
-    OBJ_DESTRUCT(&ack);
     return rc;
 }
 
@@ -542,7 +518,8 @@ cleanup:
  * @param jobid The job to terminate
  * @retval ORTE_SUCCESS
  */
-int orte_pls_bproc_orted_terminate_job(orte_jobid_t jobid) {
+int orte_pls_bproc_orted_terminate_job(orte_jobid_t jobid) 
+{
     orte_iof.iof_flush();
     return ORTE_SUCCESS;
 }
@@ -554,28 +531,21 @@ int orte_pls_bproc_orted_terminate_job(orte_jobid_t jobid) {
  * @param proc the process's name
  * @retval ORTE_SUCCESS
  */
-int orte_pls_bproc_orted_terminate_proc(const orte_process_name_t* proc) {
+int orte_pls_bproc_orted_terminate_proc(const orte_process_name_t* proc) 
+{
     orte_iof.iof_flush();
     return ORTE_SUCCESS;
 }
 
 /**
- * Finalizes the bproc_orted module. This function simply blocks on a condition
+ * Finalizes the bproc_orted module. Cleanup tmp directory/files
- * until we get a message from the seed telling us the user processes are done
+ * used for I/O forwarding.
- * so we can now terminate.
  * @retval ORTE_SUCCESS
  */
-int orte_pls_bproc_orted_finalize(void) {
+int orte_pls_bproc_orted_finalize(void) 
-    OPAL_THREAD_LOCK(&mca_pls_bproc_orted_component.lock);
+{
-    opal_condition_wait(&mca_pls_bproc_orted_component.condition, 
-                        &mca_pls_bproc_orted_component.lock);
-    orte_iof.iof_flush();
     pls_bproc_orted_remove_dir();
     orte_session_dir_finalize(orte_process_info.my_name);
-    if(mca_pls_bproc_orted_component.debug) {
-        opal_output(0, "pls_bproc_orted: all cleanup routines called\n");
-    }
-    OPAL_THREAD_UNLOCK(&mca_pls_bproc_orted_component.lock);
     return ORTE_SUCCESS;
 }
 

orte/mca/pls/bproc_orted/pls_bproc_orted.h

@@ -67,9 +67,6 @@ struct orte_pls_bproc_orted_component_t {
      *   that bproc is available and running on this node, */
     opal_mutex_t lock;
     /**< Lock used to prevent some race conditions */
-    opal_condition_t condition;
-    /**< Condition that is signaled when we have recieved a message from
-     *   the seed that it is time to terminate. */
 };
 /**
  * Convenience typedef

orte/mca/pls/bproc_orted/pls_bproc_orted_component.c

@@ -64,7 +64,6 @@ int orte_pls_bproc_orted_component_open(void)
 {
     /* initialize globals */
     OBJ_CONSTRUCT(&mca_pls_bproc_orted_component.lock, opal_mutex_t);
-    OBJ_CONSTRUCT(&mca_pls_bproc_orted_component.condition, opal_condition_t);
 
     /* lookup parameters */
     mca_base_param_reg_int(&mca_pls_bproc_orted_component.super.pls_version, 
@@ -110,7 +109,6 @@ orte_pls_base_module_t *orte_pls_bproc_orted_init(int *priority)
 int orte_pls_bproc_orted_component_close(void)
 {
     OBJ_DESTRUCT(&mca_pls_bproc_orted_component.lock);
-    OBJ_DESTRUCT(&mca_pls_bproc_orted_component.condition);
     return ORTE_SUCCESS;
 }