BF: CS-649 an invalid epilog does not set a queue error state

// + fixed typos + cleanup
hpc-gridware · Oct 4, 2024 · 78935a9 · 78935a9
1 parent 7dce9b3
commit 78935a9
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 13 deletions.
diff --git a/source/daemons/execd/reaper_execd.cc b/source/daemons/execd/reaper_execd.cc
@@ -711,6 +711,10 @@ static int clean_up_job(lListElem *jr, int failed, int shepherd_exit_status,
    case SSTATE_PROLOG_FAILED:
    case SSTATE_BEFORE_PESTART:
    case SSTATE_PESTART_FAILED:
+   case SSTATE_BEFORE_PESTOP:
+   case SSTATE_PESTOP_FAILED:
+   case SSTATE_BEFORE_EPILOG:
+   case SSTATE_EPILOG_FAILED:
       general_failure = GFSTATE_QUEUE;
       lSetUlong(jr, JR_general_failure, general_failure);
       job_related_adminmail(EXECD, jr, is_array, job_owner);
@@ -770,10 +774,6 @@ static int clean_up_job(lListElem *jr, int failed, int shepherd_exit_status,
    ** if an error occurred after the job has been run
    ** it is not as serious
    */
-   case SSTATE_BEFORE_PESTOP:
-   case SSTATE_PESTOP_FAILED:
-   case SSTATE_BEFORE_EPILOG:
-   case SSTATE_EPILOG_FAILED:
    case SSTATE_PROCSET_NOTFREED:
       general_failure = GFSTATE_NO_HALT;
       lSetUlong(jr, JR_general_failure, general_failure);

diff --git a/source/daemons/shepherd/builtin_starter.cc b/source/daemons/shepherd/builtin_starter.cc
@@ -1420,7 +1420,7 @@ int use_starter_method /* If this flag is set the shellpath contains the
             execvp(filename, args);
          }
 
-         /* Aaaah - execvp() failed */
+         // execvp() failed
          {
             char failed_str[2048+128];
             snprintf(failed_str, sizeof(failed_str), "%s failed: %s", err_str, strerror(errno));

diff --git a/source/daemons/shepherd/shepherd.cc b/source/daemons/shepherd/shepherd.cc
@@ -453,10 +453,9 @@ static int do_epilog(int timeout, int ckpt_type)
                      prolog_epilog_variables);
       exit_status = start_child("epilog", command, nullptr, timeout, ckpt_type);
       if (n_exit_status<(i=count_exit_status())) {
-         shepherd_trace("exit states increased from %d to %d", 
-                                n_exit_status, i);
+         shepherd_trace("exit states increased from %d to %d", n_exit_status, i);
          /*
-         ** in this case the child didnt get to the exec call or it failed
+         ** in this case the child didn't get to the exec call, or it failed
          ** the status that waitpid and finally start_child returns is
          ** reserved for the exit status of the job
          */
@@ -465,7 +464,7 @@ static int do_epilog(int timeout, int ckpt_type)
       }
 
       if (exit_status) {
-         switch( exit_status ) {
+         switch(exit_status) {
             case RESCHEDULE_EXIT_STATUS:
                shepherd_state = SSTATE_AGAIN;
                break;
@@ -478,10 +477,10 @@ static int do_epilog(int timeout, int ckpt_type)
          shepherd_error(0, "exit_status of epilog = %d", exit_status);
          return SSTATE_EPILOG_FAILED;
       }
-   }
-   else
+   } else {
       shepherd_trace("no epilog script to start");
-
+   }
+
    return 0;
 }
 
@@ -1348,7 +1347,7 @@ int ckpt_type
    */
    if (!SGE_STAT("exit_status", &buf) && buf.st_size) {
       /*
-      ** in this case the child didnt get to the exec call or it failed
+      ** in this case the child didn't get to the exec call, or it failed
       ** the status that waitpid and finally start_child returns is
       ** reserved for the exit status of the job
       */