Changeset 52

Timestamp:

01/07/12 21:21:29 (13 years ago)

Author:

mmamonski

Message:

store job exit status in file

Location:

trunk/pbs_drmaa

Files:

• session.c (modified) (4 diffs)
• session.h (modified) (1 diff)
• submit.c (modified) (2 diffs)

trunk/pbs_drmaa/session.c

@@ -135 +135 @@
                                 pbsdrmaa_exc_raise_pbs( "pbs_connect" );
                  }
+
+                 self->job_exit_status_file_prefix = NULL;
          }
         EXCEPT_DEFAULT
@@ -159 +161 @@
                 pbs_disconnect( pbsself->pbs_conn );
         fsd_free( pbsself->status_attrl );
+        fsd_free( pbsself->job_exit_status_file_prefix );
+
         pbsself->super_destroy( self ); 
 }
@@ -220 +224 @@
         fsd_conf_option_t *wait_thread_sleep_time = NULL;
         fsd_conf_option_t *max_retries_count = NULL;
+        fsd_conf_option_t *user_state_dir = NULL;
 
         pbs_home = fsd_conf_dict_get(self->configuration, "pbs_home" );
         wait_thread_sleep_time = fsd_conf_dict_get(self->configuration, "wait_thread_sleep_time" );
         max_retries_count = fsd_conf_dict_get(self->configuration, "max_retries_count" );
+        user_state_dir = fsd_conf_dict_get(self->configuration, "user_state_dir" );
 
         if( pbs_home && pbs_home->type == FSD_CONF_STRING )
@@ -269 +275 @@
                 fsd_log_info(("Wait thread sleep time: %d", pbsself->wait_thread_sleep_time));
           }
+
+        if( user_state_dir && user_state_dir->type == FSD_CONF_STRING )
+          {
+                struct passwd *pw = NULL;
+                uid_t uid;
+
+                uid = geteuid();
+                pw = getpwuid(uid); /* drmaa_init is always called in thread safely fashion */
+
+                if (!pw)
+                        fsd_exc_raise_fmt(FSD_ERRNO_INTERNAL_ERROR,"Failed to get pw_name of the user %d", uid);
+
+                pbsself->job_exit_status_file_prefix = fsd_asprintf(user_state_dir->val.string, pw->pw_name);
+          }
+        else
+          {
+                pbsself->job_exit_status_file_prefix = fsd_asprintf("%s/.drmaa", getenv("HOME"));
+          }
+
+        if (mkdir(pbsself->job_exit_status_file_prefix, 0600) == -1 && errno != EEXIST) /* TODO it would be much better to do stat before */
+          {
+                fsd_log_warning("Failed to create job state directory: %s.  Valid job exit status may not be available in some cases.", pbsself->job_exit_status_file_prefix)
+          }
+
+
+        /* TODO purge old exit statuses files */
 
         pbsself->super_apply_configuration(self); /* call method from the superclass */

trunk/pbs_drmaa/session.h

@@ -84 +84 @@
          */
         int wait_thread_sleep_time;
+
+        /*
+         * The prefix of the directory where the job exit status information is stored (default: %HOME/.pbs-drmaa/
+         */
+        char *job_exit_status_file_prefix;
 };
 

trunk/pbs_drmaa/submit.c

@@ -342 +342 @@
                 if( input_path != NULL )
                         script_len += strlen(" <") + strlen(input_path);
+                if (((pbsdrmaa_session_t)self->session)->job_exit_status_file_prefix)
+                        script_len += strlen("; echo $? >") + strlen(((pbsdrmaa_session_t)self->session)->job_exit_status_file_prefix) + strlen("/$PBS_JOBID.exitcode");
          }
 
@@ -357 +359 @@
                 if( input_path != NULL )
                         s += sprintf( s, " <%s", input_path );
+                if (((pbsdrmaa_session_t)self->session)->job_exit_status_file_prefix)
+                        s += sprintf( s, "; echo $? >%s/$PBS_JOBID.exitcode", ((pbsdrmaa_session_t)self->session)->job_exit_status_file_prefix);
+
                 fsd_assert( s == script+script_len );
          }