Changeset 33 for trunk

Timestamp:

10/27/11 14:10:01 (13 years ago)

Author:

mmamonski

Message:

PBS DRMAA log reader - on site fixes

Location:

trunk/pbs_drmaa

Files:

• job.c (modified) (1 diff)
• log_reader.c (modified) (4 diffs)
• session.c (modified) (1 diff)

trunk/pbs_drmaa/job.c

@@ -438 +438 @@
         pbsdrmaa_session_t *pbssession = (pbsdrmaa_session_t*)self->session;
 
-        if( pbssession->pbs_home == NULL || pbssession->super.wait_thread_started )
+        if( pbssession->pbs_home != NULL && pbssession->super.wait_thread_started )
+                fsd_log_debug(("Job on missing but WT is running. Skipping...")); /* TODO: try to provide implementation that uses accounting/server log files */
+        else
                 pbsdrmaa_job_on_missing_standard( self );       
-        else
-                pbsdrmaa_job_on_missing_standard( self ); /* TODO: try to provide implementation that uses accounting/server log files */
 }
 

trunk/pbs_drmaa/log_reader.c

@@ -321 +321 @@
                                                          }
 #ifndef PBS_PBS_PROFESSIONAL
-                                                        else if (event_type == pbsdrmaa_event_0008 && strncmp(msg, "Job deleted", 11))
+                                                        else if (event_type == pbsdrmaa_event_0008 && strncmp(msg, "Job deleted", 11) == 0)
 #else
-                                                        else if (event_type == pbsdrmaa_event_0008 && strncmp(msg, "Job to be deleted", 17))
+                                                        else if (event_type == pbsdrmaa_event_0008 && strncmp(msg, "Job to be deleted", 17) == 0)
 #endif
                                                          {
@@ -338 +338 @@
                                                                 if (job->state < DRMAA_PS_RUNNING)
                                                                  {
-                                                                        fsd_log_info(("Job %s killed before entering running state (%d).", job->job_id, job->state));
+                                                                        fsd_log_info(("WT - Job %s killed before entering running state (%d).", job->job_id, job->state));
                                                                         attribs = pbsdrmaa_add_attr(attribs, PBSDRMAA_JOB_STATE, "C");
                                                                         attribs = pbsdrmaa_add_attr(attribs, PBSDRMAA_MTIME, timestamp_unix);
@@ -368 +368 @@
                                                                          {
                                                                                 token[11] = '\0';
-                                                                                attribs = pbsdrmaa_add_attr(attribs, token, token + 12);
+                                                                                attribs = pbsdrmaa_add_attr(attribs, PBSDRMAA_EXIT_STATUS, token + 12);
                                                                                 fsd_log_info(("WT - Completion of job %s (Exit_status=%s) detected after %d seconds", job->job_id, token+12, (int)(time(NULL) - timestamp_time_t) ));
                                                                          }
@@ -451 +451 @@
 
 
-                                 { /* poll on log file */
-                                        struct timeval timeout_tv;
-                                        fd_set log_fds;
-
-                                        fsd_mutex_unlock( &self->session->mutex );
-                                        
-                                        FD_ZERO(&log_fds);
-                                        FD_SET(fileno(self->fhandle), &log_fds);
-
-                                        timeout_tv.tv_sec = 1;
-                                        timeout_tv.tv_usec = 0;
-                                        fsd_log_debug(("Polling log file for %d seconds", timeout_tv.tv_sec));
-                                        /* ignore return value - the next get line call will handle IO errors */
-                                        (void)select(1, &log_fds, NULL, NULL, &timeout_tv);
-
-                                        fsd_mutex_lock( &self->session->mutex );
-
-                                        self->run_flag = self->session->wait_thread_run_flag;
-                                 }
+
+                                fsd_mutex_unlock( &self->session->mutex );
+                                
+                                usleep(1500000); /* 500 ms - consider using inotify - but this would not work with NFS */                               
+
+                                fsd_mutex_lock( &self->session->mutex );
+
+                                self->run_flag = self->session->wait_thread_run_flag;
                         }
                         EXCEPT_DEFAULT

trunk/pbs_drmaa/session.c

@@ -239 +239 @@
                                 char errbuf[256] = "InternalError";
                                 (void)strerror_r(errno, errbuf, sizeof(errbuf));
-                                fsd_exc_raise_fmt(FSD_ERRNO_INTERNAL_ERROR,"stat error: %s",errbuf);
+                                fsd_exc_raise_fmt(FSD_ERRNO_INTERNAL_ERROR,"stat error on file %s: %s", log_path, errbuf);
                           }