Changeset 11055 for trunk/Ohana/src/opihi/pantasks/CheckJobs.c

Timestamp:

Jan 11, 2007, 4:51:41 PM (20 years ago)

Author:

eugene

Message:

added timing stats, npending limit

File:

• trunk/Ohana/src/opihi/pantasks/CheckJobs.c (modified) (11 diffs)

trunk/Ohana/src/opihi/pantasks/CheckJobs.c

@@ -5 +5 @@
   FILE *f;
   Job *job;
+  Task *task;
   Macro *macro;
   int i, status;
@@ -13 +14 @@
   while ((job = NextJob ()) != NULL) {
 
+    task = job[0].task;
+
     /* check poll period (ready to ask for status?) */
-    if (GetTaskTimer(job[0].last) < job[0].task[0].poll_period) continue;
+    if (GetTaskTimer(job[0].last) < task[0].poll_period) continue;
 
     /* check current status */
@@ -20 +23 @@
     switch (status) {
       case JOB_PENDING:
-        /* if (VerboseMode()) gprint (GP_LOG, "job %s (%d) pending\n", job[0].task[0].name, job[0].JobID); */
+        /* if (VerboseMode()) gprint (GP_LOG, "job %s (%d) pending\n", task[0].name, job[0].JobID); */
         break;
 
       case JOB_BUSY:
-        /* if (VerboseMode()) gprint (GP_LOG, "job %s (%d) busy\n", job[0].task[0].name, job[0].JobID); */
+        /* if (VerboseMode()) gprint (GP_LOG, "job %s (%d) busy\n", task[0].name, job[0].JobID); */
         break;
 
@@ -35 +38 @@
 
         /* save the stdout and stderr if desired */
-        if (job[0].stdout_dump != NULL) {
+        if ((job[0].stdout_dump != NULL) && strcasecmp(job[0].stdout_dump, "NULL")) {
           f = fopen (job[0].stdout_dump, "a");
           if (f == NULL) {
@@ -44 +47 @@
           }
         }
-        if (job[0].stderr_dump != NULL) {
+        if ((job[0].stderr_dump != NULL) && strcasecmp(job[0].stderr_dump, "NULL")) {
           f = fopen (job[0].stderr_dump, "a");
           if (f == NULL) {
@@ -68 +71 @@
         set_int_variable ("options:n", job[0].optc);
 
+        set_variable ("JOB_DTIME", job[0].dtime);
+
         if (status == JOB_CRASH) {
           /* XXX add an Ncrash element? */
-          job[0].task[0].Nfailure ++;
+          task[0].Nfailure ++;
+          UpdateTaskTimerStats (task, TIMER_FAILURE, job[0].dtime);
 
           /* run task[0].crash macro, if it exists */
           /* perhaps define PushNamedQueueBuffer */
 
-          if (VerboseMode()) gprint (GP_LOG, "job %s (%d) crash\n", job[0].task[0].name, job[0].JobID);
-          if (job[0].task[0].crash != NULL) {
-            exec_loop (job[0].task[0].crash);
+          set_str_variable ("JOB_STATUS", "CRASH");
+
+          if (VerboseMode()) gprint (GP_LOG, "job %s (%d) crash\n", task[0].name, job[0].JobID);
+          if (task[0].crash != NULL) {
+            exec_loop (task[0].crash);
           }
         }
@@ -83 +91 @@
           /* update the exit status counters */
           if (job[0].exit_status) {
-            job[0].task[0].Nfailure ++;
+            task[0].Nfailure ++;
+            UpdateTaskTimerStats (task, TIMER_FAILURE, job[0].dtime);
           } else {
-            job[0].task[0].Nsuccess ++;
-          }
+            task[0].Nsuccess ++;
+            UpdateTaskTimerStats (task, TIMER_SUCCESS, job[0].dtime);
+          }
+
+          set_int_variable ("JOB_STATUS", job[0].exit_status);
 
           /* run corresponding task[0].exit macro, if it exists */
-          if (VerboseMode()) gprint (GP_LOG, "job %s (%d) exit\n", job[0].task[0].name, job[0].JobID);
-          macro = job[0].task[0].defexit;
-          for (i = 0; i < job[0].task[0].Nexit; i++) {
-            if (job[0].exit_status == atoi(job[0].task[0].exit[i][0].name)) {
-              macro = job[0].task[0].exit[i];
+          if (VerboseMode()) gprint (GP_LOG, "job %s (%d) exit\n", task[0].name, job[0].JobID);
+          macro = task[0].defexit;
+          for (i = 0; i < task[0].Nexit; i++) {
+            if (job[0].exit_status == atoi(task[0].exit[i][0].name)) {
+              macro = task[0].exit[i];
               break;
             }
@@ -106 +118 @@
         if (queue) InitQueue (queue);
 
+        UpdateTaskTimerStats (task, TIMER_ALLJOBS, job[0].dtime);
+
+        task[0].Npending --;
         DeleteJob (job);
         continue;
@@ -120 +135 @@
      */
     if (job[0].mode == JOB_LOCAL) {
-      if (GetTaskTimer(job[0].start) < job[0].task[0].timeout_period) continue;
-      if (VerboseMode()) gprint (GP_LOG, "timeout on %s\n", job[0].task[0].name);
+      if (GetTaskTimer(job[0].start) < task[0].timeout_period) continue;
+      if (VerboseMode()) gprint (GP_LOG, "timeout on %s\n", task[0].name);
 
       // XXX harvest STDERR and STDOUT from timeout job (should be available...)
@@ -127 +142 @@
 
       /* update the timeout counter */
-      job[0].task[0].Ntimeout ++;
+      task[0].Ntimeout ++;
 
       if (!KillLocalJob (job)) {
@@ -150 +165 @@
 
       /* run task[0].timeout macro, if it exists */
-      if (job[0].task[0].timeout != NULL) {
-        exec_loop (job[0].task[0].timeout);
+      if (task[0].timeout != NULL) {
+        exec_loop (task[0].timeout);
       }
       DeleteJob (job);