X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/eef298a7cf58771583b255fb0e3e1cfcbcdb6e48..050a8d39b9ca831073520316dbc197270b11720d:/sdk/cli/bin/crunch-job

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index 06b3da99a9..402c2de60e 100755
--- 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"} = $ENV{"TMPDIR"} . "/crunch-job";
@@ -150,17 +152,25 @@ if ($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
+    # EX_TEMPFAIL so crunch-dispatch (our parent process) doesn't
+    # mark the job as failed.
     if ($Job->{'is_locked_by_uuid'}) {
-      croak("Job is locked: " . $Job->{'is_locked_by_uuid'});
+      Log(undef, "Job is locked by " . $Job->{'is_locked_by_uuid'});
+      exit EX_TEMPFAIL;
     }
     if ($Job->{'success'} ne undef) {
-      croak("Job 'success' flag (" . $Job->{'success'} . ") is not null");
+      Log(undef, "Job 'success' flag (" . $Job->{'success'} . ") is not null");
+      exit EX_TEMPFAIL;
     }
     if ($Job->{'running'}) {
-      croak("Job 'running' flag is already set");
+      Log(undef, "Job 'running' flag is already set");
+      exit EX_TEMPFAIL;
     }
     if ($Job->{'started_at'}) {
-      croak("Job 'started_at' time is already set (" . $Job->{'started_at'} . ")");
+      Log(undef, "Job 'started_at' time is already set (" . $Job->{'started_at'} . ")");
+      exit EX_TEMPFAIL;
     }
   }
 }
@@ -273,7 +283,8 @@ if ($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'}) {
-    croak("Error while updating / locking job");
+    Log(undef, "Error while updating / locking job, exiting ".EX_TEMPFAIL);
+    exit EX_TEMPFAIL;
   }
   $Job->update_attributes('started_at' => scalar gmtime,
                           'running' => 1,
@@ -383,7 +394,7 @@ else
     if ($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)
@@ -476,6 +487,8 @@ else
     croak ("could not figure out commit id for $treeish");
   }
 
+  # Note: this section is almost certainly unnecessary if we're
+  # running tasks in docker containers.
   my $installpid = fork();
   if ($installpid == 0)
   {
@@ -638,13 +651,11 @@ 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)
     {
       $build_script_to_send = $build_script;
-      $command .=
-	  "&& perl -";
     }
     $command .= "&& exec arv-mount --allow-other $ENV{TASK_KEEPMOUNT} --exec ";
     if ($docker_hash)
@@ -657,16 +668,12 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
       $command .=
           q{$(ip -o address show scope global |
               gawk 'match($4, /^([0-9\.:]+)\//, x){print "--dns", x[1]}') };
-      $command .= "--volume=\Q$ENV{CRUNCH_SRC}:/tmp/crunch-src:ro\E ";
       $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT}:/keep:ro\E ";
       $command .= "--env=\QHOME=/home/crunch\E ";
       while (my ($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 {
@@ -678,17 +685,21 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
       $command .= "--env=\QCRUNCH_SRC=/tmp/crunch-src\E ";
       $command .= "\Q$docker_hash\E ";
       $command .= "stdbuf --output=0 --error=0 ";
+      $command .= "perl - ";
       $command .= "/tmp/crunch-src/crunch_scripts/" . $Job->{"script"};
     } else {
       # Non-docker run
       $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -poll=10000 ";
       $command .= "stdbuf --output=0 --error=0 ";
+      $command .= "perl - ";
       $command .= "$ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
     }
 
     my @execargs = ('bash', '-c', $command);
     srun (\@srunargs, \@execargs, undef, $build_script_to_send);
-    exit (111);
+    # exec() failed, we assume nothing happened.
+    Log(undef, "srun() failed on build script");
+    die;
   }
   close("writer");
   if (!defined $childpid)
@@ -842,37 +853,15 @@ else {
   eval {
     open(my $orig_manifest, '-|', 'arv-get', $collated_output)
         or die "failed to get collated manifest: $!";
-    # Read the original manifest, and strip permission hints from it,
-    # so we can put the result in a Collection.
-    my @stripped_manifest_lines = ();
     my $orig_manifest_text = '';
     while (my $manifest_line = <$orig_manifest>) {
       $orig_manifest_text .= $manifest_line;
-      my @words = split(/ /, $manifest_line, -1);
-      foreach my $ii (0..$#words) {
-        if ($words[$ii] =~ /^[0-9a-f]{32}\+/) {
-          $words[$ii] =~ s/\+A[0-9a-f]{40}@[0-9a-f]{8}\b//;
-        }
-      }
-      push(@stripped_manifest_lines, join(" ", @words));
     }
-    my $stripped_manifest_text = join("", @stripped_manifest_lines);
     my $output = $arv->{'collections'}->{'create'}->execute('collection' => {
-      'uuid' => md5_hex($stripped_manifest_text),
       'manifest_text' => $orig_manifest_text,
     });
-    Log(undef, "output " . $output->{uuid});
-    $Job->update_attributes('output' => $output->{uuid}) if $job_has_uuid;
-    if ($Job->{'output_is_persistent'}) {
-      $arv->{'links'}->{'create'}->execute('link' => {
-        'tail_kind' => 'arvados#user',
-        'tail_uuid' => $User->{'uuid'},
-        'head_kind' => 'arvados#collection',
-        'head_uuid' => $Job->{'output'},
-        'link_class' => 'resources',
-        'name' => 'wants',
-      });
-    }
+    Log(undef, "output " . $output->{portable_data_hash});
+    $Job->update_attributes('output' => $output->{portable_data_hash}) if $job_has_uuid;
   };
   if ($@) {
     Log (undef, "Failed to register output manifest: $@");
@@ -882,7 +871,7 @@ else {
 Log (undef, "finish");
 
 save_meta();
-exit 0;
+exit ($Job->{'success'} ? 1 : 0);
 
 
 
@@ -1210,9 +1199,8 @@ 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+)*$/)
     {
@@ -1488,15 +1476,29 @@ __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"};
+
+for my $dir ($destdir, $task_work) {
+    if ($dir) {
+        make_path $dir;
+        -e $dir or die "Failed to create temporary directory ($dir): $!";
+    }
+}
 
 open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
 flock L, LOCK_EX;
 if (readlink ("$destdir.commit") eq $commit && -d $destdir) {
-    exit 0;
+    if (@ARGV) {
+        exec(@ARGV);
+        die "Cannot exec `@ARGV`: $!";
+    } else {
+        exit 0;
+    }
 }
 
 unlink "$destdir.commit";
@@ -1542,7 +1544,12 @@ if ($commit) {
 
 close L;
 
-exit 0;
+if (@ARGV) {
+    exec(@ARGV);
+    die "Cannot exec `@ARGV`: $!";
+} else {
+    exit 0;
+}
 
 sub shell_or_die
 {