Do not retry tasks by default anymore

Author: sawenzel

MC/bin/o2_dpg_workflow_runner.py

@@ -51,7 +51,7 @@
 parser.add_argument('--webhook', help=argparse.SUPPRESS) # log some infos to this webhook channel
 parser.add_argument('--checkpoint-on-failure', help=argparse.SUPPRESS) # debug option making a debug-tarball and sending to specified address
                                                                        # argument is alien-path
-parser.add_argument('--retry-on-failure', help=argparse.SUPPRESS, default=2) # number of times a failing task is retried
+parser.add_argument('--retry-on-failure', help=argparse.SUPPRESS, default=0) # number of times a failing task is retried
 parser.add_argument('--rootinit-speedup', help=argparse.SUPPRESS, action='store_true') # enable init of ROOT environment vars to speedup init/startup
 parser.add_argument('--action-logfile', help='Logfilename for action logs. If none given, pipeline_action_#PID.log will be used')
 parser.add_argument('--metric-logfile', help='Logfilename for metric logs. If none given, pipeline_metric_#PID.log will be used')
@@ -796,10 +796,10 @@ def waitforany(self, process_list, finished, failingtasks):
             finished.append(tid)
             process_list.remove(p)
             if returncode != 0:
-               print (str(tid) + ' failed ... checking retry')
+               print (str(self.idtotask[tid]) + ' failed ... checking retry')
                # we inspect if this is something "unlucky" which could be resolved by a simple resubmit
                if self.is_worth_retrying(tid) and self.retry_counter[tid] < int(args.retry_on_failure):
-                 print (str(tid) + ' to be retried')
+                 print (str(self.idtotask[tid]) + ' to be retried')
                  actionlogger.info ('Task ' + str(self.idtotask[tid]) + ' failed but marked to be retried ')
                  self.tids_marked_toretry.append(tid)
                  self.retry_counter[tid] += 1