7313: crunch-job reports an error when a task doesn't record state.

[arvados.git] / sdk / cli / bin / crunch-job

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index 555c4d19a613037e36395338a9de32e15ee53c6a..28da66d0ca74296ef8475a5d96c0cb6fb3d41764 100755 (executable)
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -467,7 +467,7 @@ fi
     }
     srun(["srun", "--nodelist=" . $node[0]],
          ["/bin/sh", "-ec",
-          "a=`$docker_bin run --rm $try_user_arg $docker_hash id --user` && " .
+          "a=`$docker_bin run $try_user_arg $docker_hash id --user` && " .
           " test \$a -ne 0"],
          {fork => 1});
     if ($? == 0) {
@@ -820,6 +820,9 @@ update_progress_stats();
 THISROUND:
 for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
 {
+  # Don't create new tasks if we already know the job's final result.
+  last if defined($main::success);
+
   my $id = $jobstep_todo[$todo_ptr];
   my $Jobstep = $jobstep[$id];
   if ($Jobstep->{level} != $level)
@@ -893,12 +896,13 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
         .q{&& SWAP=$(awk '($1 == "SwapTotal:"){print $2}' </proc/meminfo) }
         ."&& MEMLIMIT=\$(( (\$MEM * 95) / ($ENV{CRUNCH_NODE_SLOTS} * 100) )) "
         ."&& let SWAPLIMIT=\$MEMLIMIT+\$SWAP ";
-    $command .= "&& exec arv-mount --by-id --allow-other $ENV{TASK_KEEPMOUNT} --exec ";
+    $command .= "&& exec arv-mount --by-pdh --allow-other $ENV{TASK_KEEPMOUNT} --exec ";
     if ($docker_hash)
     {
-      my $cidfile = "$ENV{CRUNCH_TMP}/$Jobstep->{arvados_task}->{uuid}-$Jobstep->{failures}.cid";
+      my $containername = "$Jobstep->{arvados_task}->{uuid}-$Jobstep->{failures}";
+      my $cidfile = "$ENV{CRUNCH_TMP}/$containername.cid";
       $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -cgroup-parent=docker -cgroup-cid=$cidfile -poll=10000 ";
-      $command .= "$docker_bin run --rm=true --attach=stdout --attach=stderr --attach=stdin -i \Q$dockeruserarg\E --cidfile=$cidfile --sig-proxy ";
+      $command .= "$docker_bin run --name=$containername --attach=stdout --attach=stderr --attach=stdin -i \Q$dockeruserarg\E --cidfile=$cidfile --sig-proxy ";
       # We only set memory limits if Docker lets us limit both memory and swap.
       # Memory limits alone have been supported longer, but subprocesses tend
       # to get SIGKILL if they exceed that without any swap limit set.
@@ -1017,7 +1021,7 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
         ||
         ($round_num_freeslots > @freeslot && $todo_ptr+1 > $#jobstep_todo))
   {
-    last THISROUND if $main::please_freeze || defined($main::success);
+    last THISROUND if $main::please_freeze;
     if ($main::please_info)
     {
       $main::please_info = 0;
@@ -1178,6 +1182,9 @@ sub reapchildren
 
   if (!defined $task_success) {
     # task did not indicate one way or the other --> fail
+    Log($jobstepid, sprintf(
+          "ERROR: Task process exited %d, but never updated its task record to indicate success and record its output.",
+          exit_status_s($childstatus)));
     $Jobstep->{'arvados_task'}->{success} = 0;
     $Jobstep->{'arvados_task'}->save;
     $task_success = 0;
@@ -1694,20 +1701,24 @@ sub log_writer_finish()
 
   close($log_pipe_in);
 
+  my $logger_failed = 0;
   my $read_result = log_writer_read_output(120);
   if ($read_result == -1) {
+    $logger_failed = -1;
     Log (undef, "timed out reading from 'arv-put'");
   } elsif ($read_result != 0) {
+    $logger_failed = -2;
     Log(undef, "failed to read arv-put log manifest to EOF");
   }
 
   waitpid($log_pipe_pid, 0);
   if ($?) {
+    $logger_failed ||= $?;
     Log(undef, "log_writer_finish: arv-put exited " . exit_status_s($?))
   }
 
   close($log_pipe_out);
-  my $arv_put_output = $log_pipe_out_buf;
+  my $arv_put_output = $logger_failed ? undef : $log_pipe_out_buf;
   $log_pipe_pid = $log_pipe_in = $log_pipe_out = $log_pipe_out_buf =
       $log_pipe_out_select = undef;
 
@@ -1773,13 +1784,13 @@ sub save_meta
   my $justcheckpoint = shift; # false if this will be the last meta saved
   return if $justcheckpoint;  # checkpointing is not relevant post-Warehouse.pm
   return unless log_writer_is_active();
+  my $log_manifest = log_writer_finish();
+  return unless defined($log_manifest);
 
-  my $log_manifest = "";
   if ($Job->{log}) {
     my $prev_log_coll = api_call("collections/get", uuid => $Job->{log});
-    $log_manifest .= $prev_log_coll->{manifest_text};
+    $log_manifest = $prev_log_coll->{manifest_text} . $log_manifest;
   }
-  $log_manifest .= log_writer_finish();
 
   my $log_coll = api_call(
     "collections/create", ensure_unique_name => 1, collection => {