Changeset 26 for trunk/ll_drmaa/session.c

Timestamp:

11/11/12 15:06:16 (13 years ago)

Author:

mmamonski

Message:

SupMUC on site fixes: 1. Polling mode 2. Handling missing jobs 3. monitor -> drmaa_monitor 4. force stderr file creation

File:

• trunk/ll_drmaa/session.c (modified) (6 diffs)

trunk/ll_drmaa/session.c

@@ -132 +132 @@
         LL_job job_info;
 
-        char *monitor_program = LL_DRMAA_BIN_DIR"/monitor";
+        char *monitor_program = LL_DRMAA_BIN_DIR"/lldrmaa_monitor";
 
         TRY
@@ -147 +147 @@
 
                 connection_lock = fsd_mutex_lock( &self->drm_connection_mutex );
-                status = llsubmit(cmd_path, monitor_program, llself->unix_socket_name, &job_info, LL_JOB_VERSION);
+                if (self->wait_thread_run_flag) 
+                  {
+                        fsd_log_info(("llsubmit(%s, %s, %s, %p, %d)",cmd_path, monitor_program, llself->unix_socket_name, (void*)&job_info, LL_JOB_VERSION));
+                        status = llsubmit(cmd_path, monitor_program, llself->unix_socket_name, &job_info, LL_JOB_VERSION);
+                  }
+                else
+                  {
+                        fsd_log_info(("llsubmit(%s, NULL, NULL, %p, %d)",cmd_path, (void*)&job_info, LL_JOB_VERSION));
+                        status = llsubmit(cmd_path, NULL, NULL, &job_info, LL_JOB_VERSION);
+                  }
                 connection_lock = fsd_mutex_unlock( &self->drm_connection_mutex );
 
-                if(remove(cmd_path) == -1)
+                if(getenv("LLDRMAA_KEEP_CMD") == NULL && remove(cmd_path) == -1)
                         fsd_log_warning(("Can't delete cmd file: %s", cmd_path));
 
@@ -161 +170 @@
                  }
                 else /* 0 */
+                 {
                         fsd_log_debug(("llsubmit: %s",lldrmaa_err_submit(status)));
+                 }
 
                 if( start != end )
@@ -170 +181 @@
                          {
                                 job_ids[i] = fsd_asprintf("%s.%d.%d", job_info.step_list[i]->id.from_host, job_info.step_list[i]->id.cluster, job_info.step_list[i]->id.proc);
-
+                                fsd_log_info((" new array job id: %s", job_ids[i]));
                                 job = lldrmaa_job_new( fsd_strdup(job_ids[i]) );
                                 job->session = self;
@@ -185 +196 @@
                         job_ids[0] = fsd_asprintf( "%s.%d.0", job_info.step_list[0]->id.from_host, job_info.step_list[0]->id.cluster);
 
+                        fsd_log_info((" new job id: %s", job_ids[0]));
                         job = lldrmaa_job_new( fsd_strdup(job_ids[0]) );
                         job->session = self;
@@ -247 +259 @@
          }
 
-        if ( !self->wait_thread_started )
-                fsd_exc_raise_msg(FSD_ERRNO_INTERNAL_ERROR, "DRMAA for LL requires that wait thread is enable. Don't disable it in configuration file!" );
 }