X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/1e3f8ceebd90058e902494fae84b1fd57ac6693b..febdebbb58592be73dcf7d4bd4b2c7ff96657741:/sdk/cli/bin/crunch-job

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index f2e9fc2878..b8afe638ac 100755
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -1,10 +1,9 @@
 #!/usr/bin/env perl
+# -*- mode: perl; perl-indent-level: 2; indent-tabs-mode: nil; -*-
 # Copyright (C) The Arvados Authors. All rights reserved.
 #
 # SPDX-License-Identifier: AGPL-3.0
 
-# -*- mode: perl; perl-indent-level: 2; indent-tabs-mode: nil; -*-
-
 =head1 NAME
 
 crunch-job: Execute job steps, save snapshots as requested, collate output.
@@ -133,6 +132,7 @@ my $resume_stash;
 my $cgroup_root = "/sys/fs/cgroup";
 my $docker_bin = "docker.io";
 my $docker_run_args = "";
+my $srun_sync_timeout = 15*60;
 GetOptions('force-unlock' => \$force_unlock,
            'git-dir=s' => \$git_dir,
            'job=s' => \$jobspec,
@@ -142,6 +142,7 @@ GetOptions('force-unlock' => \$force_unlock,
            'cgroup-root=s' => \$cgroup_root,
            'docker-bin=s' => \$docker_bin,
            'docker-run-args=s' => \$docker_run_args,
+           'srun-sync-timeout=i' => \$srun_sync_timeout,
     );
 
 if (defined $job_api_token) {
@@ -1188,6 +1189,8 @@ sub reapchildren
                     . $slot[$proc{$pid}->{slot}]->{cpu});
     my $jobstepidx = $proc{$pid}->{jobstepidx};
 
+    readfrompipes_after_exit ($jobstepidx);
+
     $children_reaped++;
     my $elapsed = time - $proc{$pid}->{time};
     my $Jobstep = $jobstep[$jobstepidx];
@@ -1259,7 +1262,6 @@ sub reapchildren
     $Jobstep->{finishtime} = time;
     $Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
     retry_op(sub { $Jobstep->{'arvados_task'}->save; }, "job_tasks.update API");
-    process_stderr_final ($jobstepidx);
     Log ($jobstepidx, sprintf("task output (%d bytes): %s",
                               length($Jobstep->{'arvados_task'}->{output}),
                               $Jobstep->{'arvados_task'}->{output}));
@@ -1562,9 +1564,27 @@ sub preprocess_stderr
 }
 
 
-sub process_stderr_final
+# Read whatever is still available on its stderr+stdout pipes after
+# the given child process has exited.
+sub readfrompipes_after_exit
 {
   my $jobstepidx = shift;
+
+  # The fact that the child has exited allows some convenient
+  # simplifications: (1) all data must have already been written, so
+  # there's no need to wait for more once sysread returns 0; (2) the
+  # total amount of data available is bounded by the pipe buffer size,
+  # so it's safe to read everything into one string.
+  my $buf;
+  while (0 < sysread ($reader{$jobstepidx}, $buf, 65536)) {
+    $jobstep[$jobstepidx]->{stderr_at} = time;
+    $jobstep[$jobstepidx]->{stderr} .= $buf;
+  }
+  if ($jobstep[$jobstepidx]->{stdout_r}) {
+    while (0 < sysread ($jobstep[$jobstepidx]->{stdout_r}, $buf, 65536)) {
+      $jobstep[$jobstepidx]->{stdout_captured} .= $buf;
+    }
+  }
   preprocess_stderr ($jobstepidx);
 
   map {
@@ -1989,6 +2009,8 @@ sub srun_sync
   my ($stdout_r, $stdout_w);
   pipe $stdout_r, $stdout_w or croak("pipe() failed: $!");
 
+  my $started_srun = scalar time;
+
   my $srunpid = fork();
   if ($srunpid == 0)
   {
@@ -2032,12 +2054,17 @@ sub srun_sync
     if (!$busy) {
       select(undef, undef, undef, 0.1);
     }
+    if (($started_srun + $srun_sync_timeout) < scalar time) {
+      # Exceeded general timeout for "srun_sync" operations, likely
+      # means something got stuck on the remote node.
+      Log(undef, "srun_sync exceeded timeout, will fail.");
+      $main::please_freeze = 1;
+    }
     killem(keys %proc) if $main::please_freeze;
   }
   my $exited = $?;
 
-  1 while readfrompipes();
-  process_stderr_final ($jobstepidx);
+  readfrompipes_after_exit ($jobstepidx);
 
   Log (undef, "$label: exit ".exit_status_s($exited));