Merge branch 'master' into 3187-pipeline-instance-page

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

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index 0ffd8a9a17ace37a5137013074980db467ef93e1..70f379e53fd9cc307bd933bc1b21276097863e4a 100755 (executable)
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -74,6 +74,7 @@ behavior (e.g., cancel job if cancelled_at becomes non-nil).
 
 use strict;
 use POSIX ':sys_wait_h';
+use POSIX qw(strftime);
 use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK);
 use Arvados;
 use Digest::MD5 qw(md5_hex);
@@ -729,6 +730,9 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
   delete $Jobstep->{stderr};
   delete $Jobstep->{finishtime};
 
+  $Jobstep->{'arvados_task'}->{started_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{starttime});
+  $Jobstep->{'arvados_task'}->save;
+
   splice @jobstep_todo, $todo_ptr, 1;
   --$todo_ptr;
 
@@ -844,12 +848,6 @@ release_allocation();
 freeze();
 my $collated_output = &collate_output();
 
-if ($job_has_uuid) {
-  $Job->update_attributes('running' => 0,
-                          'success' => $collated_output && $main::success,
-                          'finished_at' => scalar gmtime)
-}
-
 if (!$collated_output) {
   Log(undef, "output undef");
 }
@@ -876,6 +874,13 @@ else {
 Log (undef, "finish");
 
 save_meta();
+
+if ($job_has_uuid) {
+  $Job->update_attributes('running' => 0,
+                          'success' => $collated_output && $main::success,
+                          'finished_at' => scalar gmtime)
+}
+
 exit ($Job->{'success'} ? 1 : 0);
 
 
@@ -979,6 +984,8 @@ sub reapchildren
   }
   $Jobstep->{exitcode} = $childstatus;
   $Jobstep->{finishtime} = time;
+  $Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
+  $Jobstep->{'arvados_task'}->save;
   process_stderr ($jobstepid, $task_success);
   Log ($jobstepid, "output " . $Jobstep->{'arvados_task'}->{output});
 
@@ -1200,7 +1207,8 @@ sub collate_output
   Log (undef, "collate");
 
   my ($child_out, $child_in);
-  my $pid = open2($child_out, $child_in, 'arv-put', '--raw');
+  my $pid = open2($child_out, $child_in, 'arv-put', '--raw',
+                  '--retries', put_retry_count());
   my $joboutput;
   for (@jobstep)
   {
@@ -1340,8 +1348,9 @@ sub save_meta
   return if $justcheckpoint;  # checkpointing is not relevant post-Warehouse.pm
 
   $local_logfile->flush;
-  my $cmd = "arv-put --portable-data-hash --filename ''\Q$keep_logfile\E "
-      . quotemeta($local_logfile->filename);
+  my $retry_count = put_retry_count();
+  my $cmd = "arv-put --portable-data-hash --retries $retry_count " .
+      "--filename ''\Q$keep_logfile\E " . quotemeta($local_logfile->filename);
   my $loglocator = `$cmd`;
   die "system $cmd failed: $?" if $?;
   chomp($loglocator);
@@ -1487,6 +1496,20 @@ sub find_docker_image {
   }
 }
 
+sub put_retry_count {
+  # Calculate a --retries argument for arv-put that will have it try
+  # approximately as long as this Job has been running.
+  my $stoptime = shift || time;
+  my $starttime = $jobstep[0]->{starttime};
+  my $timediff = defined($starttime) ? ($stoptime - $starttime) : 1;
+  my $retries = 0;
+  while ($timediff >= 2) {
+    $retries++;
+    $timediff /= 2;
+  }
+  return ($retries > 3) ? $retries : 3;
+}
+
 __DATA__
 #!/usr/bin/perl