Merge branch 'master' into 3036-collection-uuids

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

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index d5edf0beed78240749d2712c8a9b00642e2e4945..638353d76fae023f4cf2c874c60cb7ea8c40dc2b 100755 (executable)
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -84,6 +84,8 @@ use File::Temp;
 use Fcntl ':flock';
 use File::Path qw( make_path );
 
+use constant EX_TEMPFAIL => 75;
+
 $ENV{"TMPDIR"} ||= "/tmp";
 unless (defined $ENV{"CRUNCH_TMP"}) {
   $ENV{"CRUNCH_TMP"} = $ENV{"TMPDIR"} . "/crunch-job";
@@ -151,24 +153,24 @@ if ($job_has_uuid)
   $Job = $arv->{'jobs'}->{'get'}->execute('uuid' => $jobspec);
   if (!$force_unlock) {
     # If some other crunch-job process has grabbed this job (or we see
-    # other evidence that the job is already underway) we exit 111 so
-    # crunch-dispatch (our parent process) doesn't mark the job as
-    # failed.
+    # other evidence that the job is already underway) we exit
+    # EX_TEMPFAIL so crunch-dispatch (our parent process) doesn't
+    # mark the job as failed.
     if ($Job->{'is_locked_by_uuid'}) {
-      Log(undef, "Job is locked by " . $Job->{'is_locked_by_uuid'} . ", exiting 111");
-      exit(111);
+      Log(undef, "Job is locked by " . $Job->{'is_locked_by_uuid'});
+      exit EX_TEMPFAIL;
     }
     if ($Job->{'success'} ne undef) {
       Log(undef, "Job 'success' flag (" . $Job->{'success'} . ") is not null");
-      exit(111);
+      exit EX_TEMPFAIL;
     }
     if ($Job->{'running'}) {
       Log(undef, "Job 'running' flag is already set");
-      exit(111);
+      exit EX_TEMPFAIL;
     }
     if ($Job->{'started_at'}) {
       Log(undef, "Job 'started_at' time is already set (" . $Job->{'started_at'} . ")");
-      exit(111);
+      exit EX_TEMPFAIL;
     }
   }
 }
@@ -281,8 +283,8 @@ if ($job_has_uuid)
   # Claim this job, and make sure nobody else does
   unless ($Job->update_attributes('is_locked_by_uuid' => $User->{'uuid'}) &&
           $Job->{'is_locked_by_uuid'} == $User->{'uuid'}) {
-    Log(undef, "Error while updating / locking job, exiting 111");
-    exit(111);
+    Log(undef, "Error while updating / locking job, exiting ".EX_TEMPFAIL);
+    exit EX_TEMPFAIL;
   }
   $Job->update_attributes('started_at' => scalar gmtime,
                           'running' => 1,
@@ -392,7 +394,7 @@ else
     if ($cleanpid == 0)
     {
       srun (["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
-           ['bash', '-c', 'if mount | grep -q $JOB_WORK/; then sudo /bin/umount $JOB_WORK/* 2>/dev/null; fi; sleep 1; rm -rf $JOB_WORK $CRUNCH_TMP/opt $CRUNCH_TMP/src*']);
+           ['bash', '-c', 'if mount | grep -q $JOB_WORK/; then for i in $JOB_WORK/*keep; do /bin/fusermount -z -u $i; done; fi; sleep 1; rm -rf $JOB_WORK $CRUNCH_TMP/opt $CRUNCH_TMP/src*']);
       exit (1);
     }
     while (1)
@@ -647,7 +649,7 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
     my $build_script_to_send = "";
     my $command =
        "if [ -e $ENV{TASK_WORK} ]; then rm -rf $ENV{TASK_WORK}; fi; "
-        ."mkdir -p $ENV{JOB_WORK} $ENV{CRUNCH_TMP} $ENV{TASK_WORK} $ENV{TASK_KEEPMOUNT}"
+        ."mkdir -p $ENV{CRUNCH_TMP} $ENV{JOB_WORK} $ENV{TASK_WORK} $ENV{TASK_KEEPMOUNT} "
        ."&& cd $ENV{CRUNCH_TMP} ";
     if ($build_script)
     {
@@ -672,10 +674,7 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
       while (my ($env_key, $env_val) = each %ENV)
       {
         if ($env_key =~ /^(ARVADOS|JOB|TASK)_/) {
-          if ($env_key eq "TASK_WORK") {
-            $command .= "--env=\QTASK_WORK=/tmp/crunch-job\E ";
-          }
-          elsif ($env_key eq "TASK_KEEPMOUNT") {
+          if ($env_key eq "TASK_KEEPMOUNT") {
             $command .= "--env=\QTASK_KEEPMOUNT=/keep\E ";
           }
           else {
@@ -893,7 +892,7 @@ else {
 Log (undef, "finish");
 
 save_meta();
-exit 0;
+exit ($Job->{'success'} ? 1 : 0);
 
 
 
@@ -1221,9 +1220,8 @@ sub collate_output
   my $joboutput;
   for (@jobstep)
   {
-    next if (!exists $_->{'arvados_task'}->{output} ||
-             !$_->{'arvados_task'}->{'success'} ||
-             $_->{'exitcode'} != 0);
+    next if (!exists $_->{'arvados_task'}->{'output'} ||
+             !$_->{'arvados_task'}->{'success'});
     my $output = $_->{'arvados_task'}->{output};
     if ($output !~ /^[0-9a-f]{32}(\+\S+)*$/)
     {
@@ -1499,10 +1497,17 @@ __DATA__
 # checkout-and-build
 
 use Fcntl ':flock';
+use File::Path qw( make_path );
 
 my $destdir = $ENV{"CRUNCH_SRC"};
 my $commit = $ENV{"CRUNCH_SRC_COMMIT"};
 my $repo = $ENV{"CRUNCH_SRC_URL"};
+my $task_work = $ENV{"TASK_WORK"};
+
+if ($task_work) {
+    make_path $task_work;
+    -e $task_work or die "Failed to create temporary working directory ($task_work): $!";
+}
 
 open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
 flock L, LOCK_EX;