X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/b269c28f1d54e8609f36c8aeb77a2b6025172066..3cbdb14acde0092ae4f33e41a4bf10c6a1fce052:/sdk/cli/bin/crunch-job

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index b38efdc53e..5539012c49 100755
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -1,4 +1,4 @@
-#!/usr/bin/perl
+#!/usr/bin/env perl
 # -*- mode: perl; perl-indent-level: 2; indent-tabs-mode: nil; -*-
 
 =head1 NAME
@@ -126,7 +126,7 @@ my $jobspec;
 my $job_api_token;
 my $no_clear_tmp;
 my $resume_stash;
-my $docker_bin = "/usr/bin/docker.io";
+my $docker_bin = "docker.io";
 GetOptions('force-unlock' => \$force_unlock,
            'git-dir=s' => \$git_dir,
            'job=s' => \$jobspec,
@@ -169,8 +169,7 @@ if ($jobspec =~ /^[-a-z\d]+$/)
 }
 else
 {
-  $Job = JSON::decode_json($jobspec);
-  $local_job = 1;
+  $local_job = JSON::decode_json($jobspec);
 }
 
 
@@ -178,7 +177,7 @@ else
 # at least able to run basic commands: they aren't down or severely
 # misconfigured.
 my $cmd = ['true'];
-if ($Job->{docker_image_locator}) {
+if (($Job || $local_job)->{docker_image_locator}) {
   $cmd = [$docker_bin, 'ps', '-q'];
 }
 Log(undef, "Sanity check is `@$cmd`");
@@ -208,15 +207,15 @@ else
 {
   if (!$resume_stash)
   {
-    map { croak ("No $_ specified") unless $Job->{$_} }
+    map { croak ("No $_ specified") unless $local_job->{$_} }
     qw(script script_version script_parameters);
   }
 
-  $Job->{'is_locked_by_uuid'} = $User->{'uuid'};
-  $Job->{'started_at'} = gmtime;
-  $Job->{'state'} = 'Running';
+  $local_job->{'is_locked_by_uuid'} = $User->{'uuid'};
+  $local_job->{'started_at'} = gmtime;
+  $local_job->{'state'} = 'Running';
 
-  $Job = api_call("jobs/create", job => $Job);
+  $Job = api_call("jobs/create", job => $local_job);
 }
 $job_id = $Job->{'uuid'};
 
@@ -391,12 +390,12 @@ if (!defined $no_clear_tmp) {
   my $cleanpid = fork();
   if ($cleanpid == 0)
   {
-    # Find FUSE mounts that look like Keep mounts (the mount path has the
-    # word "keep") and unmount them.  Then clean up work directories.
-    # TODO: When #5036 is done and widely deployed, we can get rid of the
-    # regular expression and just unmount everything with type fuse.keep.
+    # Find FUSE mounts under $CRUNCH_TMP and unmount them.
+    # Then clean up work directories.
+    # TODO: When #5036 is done and widely deployed, we can limit mount's
+    # -t option to simply fuse.keep.
     srun (["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
-          ['bash', '-ec', 'mount -t fuse,fuse.keep | awk \'($3 ~ /\ykeep\y/){print $3}\' | xargs -r -n 1 fusermount -u -z; sleep 1; rm -rf $JOB_WORK $CRUNCH_INSTALL $CRUNCH_TMP/task $CRUNCH_TMP/src* $CRUNCH_TMP/*.cid']);
+          ['bash', '-ec', '-o', 'pipefail', 'mount -t fuse,fuse.keep | awk "(index(\$3, \"$CRUNCH_TMP\") == 1){print \$3}" | xargs -r -n 1 fusermount -u -z; sleep 1; rm -rf $JOB_WORK $CRUNCH_INSTALL $CRUNCH_TMP/task $CRUNCH_TMP/src* $CRUNCH_TMP/*.cid']);
     exit (1);
   }
   while (1)
@@ -405,7 +404,10 @@ if (!defined $no_clear_tmp) {
     freeze_if_want_freeze ($cleanpid);
     select (undef, undef, undef, 0.1);
   }
-  Log (undef, "Cleanup command exited ".exit_status_s($?));
+  if ($?) {
+    Log(undef, "Clean work dirs: exit ".exit_status_s($?));
+    exit(EX_RETRY_UNLOCKED);
+  }
 }
 
 # If this job requires a Docker image, install that.
@@ -596,7 +598,7 @@ else {
   unless ($? == 0 && $sha1 =~ /^([0-9a-f]{40})$/) {
     croak("`$gitcmd rev-list` exited "
           .exit_status_s($?)
-          .", '$treeish' not found. Giving up.");
+          .", '$treeish' not found, giving up");
   }
   $commit = $1;
   Log(undef, "Version $treeish is commit $commit");
@@ -866,13 +868,6 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
         $command .= "--memory=\${MEMLIMIT}k --memory-swap=\${SWAPLIMIT}k ";
       }
 
-      # Dynamically configure the container to use the host system as its
-      # DNS server.  Get the host's global addresses from the ip command,
-      # and turn them into docker --dns options using gawk.
-      $command .=
-          q{$(ip -o address show scope global |
-              gawk 'match($4, /^([0-9\.:]+)\//, x){print "--dns", x[1]}') };
-
       # The source tree and $destdir directory (which we have
       # installed on the worker host) are available in the container,
       # under the same path.
@@ -985,14 +980,14 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
     my $gotsome
 	= readfrompipes ()
 	+ reapchildren ();
-    if (!$gotsome)
+    if (!$gotsome || ($latest_refresh + 2 < scalar time))
     {
       check_refresh_wanted();
       check_squeue();
       update_progress_stats();
       select (undef, undef, undef, 0.1);
     }
-    elsif (time - $progress_stats_updated >= 30)
+    elsif (time - $progress_stats_updated >= 30 || $progress_is_dirty)
     {
       update_progress_stats();
     }
@@ -1099,8 +1094,8 @@ sub update_progress_stats
   $progress_stats_updated = time;
   return if !$progress_is_dirty;
   my ($todo, $done, $running) = (scalar @jobstep_todo,
-				 scalar @jobstep_done,
-				 scalar @slot - scalar @freeslot - scalar @holdslot);
+                                 scalar @jobstep_done,
+                                 scalar keys(%proc));
   $Job->{'tasks_summary'} ||= {};
   $Job->{'tasks_summary'}->{'todo'} = $todo;
   $Job->{'tasks_summary'}->{'done'} = $done;
@@ -2042,7 +2037,7 @@ sub set_nonblocking {
 }
 
 __DATA__
-#!/usr/bin/perl
+#!/usr/bin/env perl
 #
 # This is crunch-job's internal dispatch script.  crunch-job running on the API
 # server invokes this script on individual compute nodes, or localhost if we're
@@ -2210,11 +2205,12 @@ if (-d $sdk_root) {
 my $python_dir = "$install_dir/python";
 if ((-d $python_dir) and can_run("python2.7")) {
   open(my $egg_info_pipe, "-|",
-       "python2.7 \Q$python_dir/setup.py\E --quiet egg_info 2>&1 >/dev/null");
+       "python2.7 \Q$python_dir/setup.py\E egg_info 2>&1 >/dev/null");
   my @egg_info_errors = <$egg_info_pipe>;
   close($egg_info_pipe);
+
   if ($?) {
-    if (@egg_info_errors and ($egg_info_errors[-1] =~ /\bgit\b/)) {
+    if (@egg_info_errors and (($egg_info_errors[-1] =~ /\bgit\b/) or ($egg_info_errors[-1] =~ /\[Errno 2\]/))) {
       # egg_info apparently failed because it couldn't ask git for a build tag.
       # Specify no build tag.
       open(my $pysdk_cfg, ">>", "$python_dir/setup.cfg");
@@ -2223,7 +2219,7 @@ if ((-d $python_dir) and can_run("python2.7")) {
     } else {
       my $egg_info_exit = $? >> 8;
       foreach my $errline (@egg_info_errors) {
-        print STDERR_ORIG $errline;
+        warn $errline;
       }
       warn "python setup.py egg_info failed: exit $egg_info_exit";
       exit ($egg_info_exit || 1);