Cleanup some missed updates to orte_wait_cb as params have changed

Refs trac:4717

This commit was SVN r32025.

The following Trac tickets were found above:
  Ticket 4717 --> https://svn.open-mpi.org/trac/ompi/ticket/4717

orte/mca/plm/alps/plm_alps_module.c

@@ -107,7 +107,7 @@ orte_plm_base_module_t orte_plm_alps_module = {
 /*
  * Local variables
  */
-static pid_t alps_pid = 0;
+static orte_proc_t *alpsrun = NULL;
 static bool failed_launch;
 static void launch_daemons(int fd, short args, void *cbdata);
 
@@ -462,7 +462,9 @@ static int plm_alps_terminate_orteds(void)
      * do NOT ERROR_LOG any return code to avoid confusing, duplicate error
      * messages
      */
-    orte_wait_cb_cancel(alps_pid);
+    if (NULL != alpsrun) {
+        orte_wait_cb_cancel(alpsrun);
+    }
 
     /* now tell them to die */
     if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
@@ -478,8 +480,8 @@ static int plm_alps_terminate_orteds(void)
  */
 static int plm_alps_signal_job(orte_jobid_t jobid, int32_t signal)
 {
-    if (0 != alps_pid) {
+    if (NULL != alpsrun && 0 != alpsrun->pid) {
-        kill(alps_pid, (int)signal);
+        kill(alpsrun->pid, (int)signal);
    }
     return ORTE_SUCCESS;
 }
@@ -489,6 +491,10 @@ static int plm_alps_finalize(void)
 {
     int rc;
     
+    if (NULL != alpsrun) {
+        OBJ_RELEASE(alpsrun);
+    }
+
     /* cleanup any pending recvs */
     if (ORTE_SUCCESS != (rc = orte_plm_base_comm_stop())) {
         ORTE_ERROR_LOG(rc);
@@ -498,7 +504,7 @@ static int plm_alps_finalize(void)
 }
 
 
-static void alps_wait_cb(pid_t pid, int status, void* cbdata){
+static void alps_wait_cb(orte_proc_t *proc, void* cbdata){
     orte_job_t *jdata;
 
     /* According to the ALPS folks, alps always returns the highest exit
@@ -518,7 +524,7 @@ static void alps_wait_cb(pid_t pid, int status, void* cbdata){
     */
     jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
     
-    if (0 != status) {
+    if (0 != proc->exit_code) {
         if (failed_launch) {
             /* report that the daemon has failed so we break out of the daemon
              * callback receive and exit
@@ -531,7 +537,6 @@ static void alps_wait_cb(pid_t pid, int status, void* cbdata){
             ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_ABORTED);
         }
     }
-    
 }
 
 
@@ -551,6 +556,11 @@ static int plm_alps_start_proc(int argc, char **argv, char **env,
         return ORTE_ERR_SYS_LIMITS_CHILDREN;
     }
     
+    alpsrun = OBJ_NEW(orte_proc_t);
+    alpsrun->pid = alps_pid;
+    /* setup the waitpid so we can find out if alps succeeds! */
+    orte_wait_cb(alpsrun, alps_wait_cb, NULL);
+
     if (0 == alps_pid) {  /* child */
         char *bin_base = NULL, *lib_base = NULL;
 
@@ -635,8 +645,6 @@ static int plm_alps_start_proc(int argc, char **argv, char **env,
         sides of the fork... */
         setpgid(alps_pid, alps_pid);
         
-        /* setup the waitpid so we can find out if alps succeeds! */
-        orte_wait_cb(alps_pid, alps_wait_cb, NULL);
         free(exec_argv);
     }
 

orte/mca/plm/slurm/plm_slurm_module.c

@@ -508,7 +508,7 @@ static int plm_slurm_finalize(void)
 }
 
 
-static void srun_wait_cb(pid_t pid, int status, void* cbdata){
+static void srun_wait_cb(orte_proc_t *proc, void* cbdata){
     orte_job_t *jdata;
     
     /* According to the SLURM folks, srun always returns the highest exit
@@ -545,7 +545,7 @@ static void srun_wait_cb(pid_t pid, int status, void* cbdata){
         /* if this is after launch, then we need to abort only if the status
          * returned is non-zero - i.e., if the orteds exited with an error
          */
-        if (0 != status) {
+        if (0 != proc->exit_code) {
             /* an orted must have died unexpectedly after launch - report
              * that the daemon has failed so we exit
              */
@@ -555,7 +555,7 @@ static void srun_wait_cb(pid_t pid, int status, void* cbdata){
             ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_ABORTED);
         }
         /* otherwise, check to see if this is the primary pid */
-        if (primary_srun_pid == pid) {
+        if (primary_srun_pid == proc->pid) {
             /* in this case, we just want to fire the proper trigger so
              * mpirun can exit
              */
@@ -567,6 +567,8 @@ static void srun_wait_cb(pid_t pid, int status, void* cbdata){
             ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DAEMONS_TERMINATED);
         }
     }
+    /* done with this dummy */
+    OBJ_RELEASE(proc);
 }
 
 
@@ -576,6 +578,7 @@ static int plm_slurm_start_proc(int argc, char **argv, char **env,
     int fd;
     int srun_pid;
     char *exec_argv = opal_path_findv(argv[0], 0, env, NULL);
+    orte_proc_t *dummy;
 
     if (NULL == exec_argv) {
         return ORTE_ERR_NOT_FOUND;
@@ -588,6 +591,12 @@ static int plm_slurm_start_proc(int argc, char **argv, char **env,
         return ORTE_ERR_SYS_LIMITS_CHILDREN;
     }
     
+    /* setup a dummy proc object to track the srun */
+    dummy = OBJ_NEW(orte_proc_t);
+    dummy->pid = srun_pid;
+    /* setup the waitpid so we can find out if srun succeeds! */
+    orte_wait_cb(dummy, srun_wait_cb, NULL);
+
     if (0 == srun_pid) {  /* child */
         char *bin_base = NULL, *lib_base = NULL;
 
@@ -677,8 +686,6 @@ static int plm_slurm_start_proc(int argc, char **argv, char **env,
             primary_pid_set = true;
         }
         
-        /* setup the waitpid so we can find out if srun succeeds! */
-        orte_wait_cb(srun_pid, srun_wait_cb, NULL);
         free(exec_argv);
     }
 

orte/mca/sstore/stage/sstore_stage_local.c

@@ -1490,6 +1490,7 @@ static int start_compression(orte_sstore_stage_local_snapshot_info_t *handle_inf
 {
     int ret, exit_status = ORTE_SUCCESS;
     char * postfix = NULL;
+    orte_proc_t *proc;
 
     /* Sanity Check */
     if( !orte_sstore_stage_enabled_compression ) {
@@ -1531,11 +1532,10 @@ static int start_compression(orte_sstore_stage_local_snapshot_info_t *handle_inf
                          app_info->compress_pid,
                          ORTE_NAME_PRINT(&(app_info->name)) ));
 
-    if( ORTE_SUCCESS != (ret = orte_wait_cb(app_info->compress_pid, sstore_stage_local_compress_waitpid_cb, app_info) ) ) {
+    proc = OBJ_NEW(orte_proc_t);
-        ORTE_ERROR_LOG(ret);
+    proc->pid = app_info->compress_pid;
-        exit_status = ret;
+
-        goto cleanup;
+    orte_wait_cb(proc, sstore_stage_local_compress_waitpid_cb, app_info);
-    }
 
  cleanup:
     if( NULL != postfix ) {
@@ -1546,7 +1546,7 @@ static int start_compression(orte_sstore_stage_local_snapshot_info_t *handle_inf
     return exit_status;
 }
 
-static void sstore_stage_local_compress_waitpid_cb(pid_t pid, int status, void* cbdata)
+static void sstore_stage_local_compress_waitpid_cb(orte_proc_t *proc, void* cbdata)
 {
     orte_sstore_stage_local_app_snapshot_info_t *app_info = NULL;
 
@@ -1558,6 +1558,7 @@ static void sstore_stage_local_compress_waitpid_cb(pid_t pid, int status, void*
                          ORTE_NAME_PRINT(&(app_info->name)) ));
 
     app_info->compress_pid = 0;
+    OBJ_RELEASE(proc);
 }
 
 static int wait_all_compressed(orte_sstore_stage_local_snapshot_info_t *handle_info)