[ARVADOS] created: d70f2d85958f988be3c227870d7adc578ee12c8a

[git at public.curoverse.com]

        at  d70f2d85958f988be3c227870d7adc578ee12c8a (commit)


commit d70f2d85958f988be3c227870d7adc578ee12c8a
Author: Brett Smith <brett at curoverse.com>
Date:   Wed Feb 25 09:50:53 2015 -0500

    5283: Improve reliability of crunch-job output collation.
    
    * Check the results of all pipe opens, exit statuses, and writes.
      Log any problems.
    * Have fetch_block return undef when it encounters trouble, rather
      than dying.  create_output_collection already checks for this, so it
      effectively bubbles up the error.
    * Retry all of the associated API calls.
    * Kill the manifest creation pipe if we give up on it, per the TODO.
    
    This probably won't resolve #5283, but hopefully these changes will
    give us additional information to help diagnose the problem.

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index d69aee6..0d49eae 100755
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -1252,16 +1252,19 @@ sub process_stderr
 sub fetch_block
 {
   my $hash = shift;
-  my ($keep, $child_out, $output_block);
-
-  my $cmd = "arv-get \Q$hash\E";
-  open($keep, '-|', $cmd) or die "fetch_block: $cmd: $!";
-  $output_block = '';
+  my $keep;
+  if (!open($keep, "-|", "arv-get", "--retries", retry_count(), $hash)) {
+    Log(undef, "fetch_block run error from arv-get $hash: $!");
+    return undef;
+  }
+  my $output_block = "";
   while (1) {
     my $buf;
     my $bytes = sysread($keep, $buf, 1024 * 1024);
     if (!defined $bytes) {
-      die "reading from arv-get: $!";
+      Log(undef, "fetch_block read error from arv-get: $!");
+      $output_block = undef;
+      last;
     } elsif ($bytes == 0) {
       # sysread returns 0 at the end of the pipe.
       last;
@@ -1271,6 +1274,10 @@ sub fetch_block
     }
   }
   close $keep;
+  if ($?) {
+    Log(undef, "fetch_block arv-get exited " . exit_status_s($?));
+    $output_block = undef;
+  }
   return $output_block;
 }
 
@@ -1283,14 +1290,13 @@ sub create_output_collection
   Log (undef, "collate");
 
   my ($child_out, $child_in);
-  my $pid = open2($child_out, $child_in, 'python', '-c',
-                  'import arvados; ' .
-                  'import sys; ' .
-                  'print arvados.api()' .
-                  '.collections()' .
-                  '.create(body={"manifest_text":sys.stdin.read()})' .
-                  '.execute()["portable_data_hash"]'
-      );
+  my $pid = open2($child_out, $child_in, 'python', '-c', q{
+import arvados
+import sys
+print (arvados.api("v1").collections().
+       create(body={"manifest_text": sys.stdin.read()}).
+       execute(num_retries=int(sys.argv[1]))["portable_data_hash"])
+}, retry_count());
 
   my $task_idx = -1;
   for (@jobstep)
@@ -1298,35 +1304,46 @@ sub create_output_collection
     ++$task_idx;
     next unless exists $_->{'arvados_task'}->{'output'};
     my $output = $_->{'arvados_task'}->{output};
-    if ($output !~ /^[0-9a-f]{32}(\+\S+)*$/)
-    {
-      print $child_in $output;
-    }
-    elsif (defined (my $outblock = fetch_block ($output)))
-    {
-      print $child_in $outblock;
+    my $next_write;
+    if ($output =~ /^[0-9a-f]{32}(\+\S+)*$/) {
+      $next_write = fetch_block($output);
+    } else {
+      $next_write = $output;
     }
-    else
-    {
+    if (defined($next_write)) {
+      if (!defined(syswrite($child_in, $next_write))) {
+        # There's been an error writing.  Stop the loop.
+        # We'll log details about the exit code later.
+        last;
+      }
+    } else {
       my $uuid = $_->{'arvados_task'}->{'uuid'};
       Log (undef, "Error retrieving '$output' output by task $task_idx ($uuid)");
       $main::success = 0;
     }
   }
-  $child_in->close;
+  close($child_in);
 
   my $joboutput;
   my $s = IO::Select->new($child_out);
   if ($s->can_read(120)) {
     sysread($child_out, $joboutput, 64 * 1024 * 1024);
-    chomp($joboutput);
-    # TODO: Ensure exit status == 0.
+    waitpid($pid, 0);
+    if ($?) {
+      Log(undef, "output collection creation exited " . exit_status_s($?));
+      $joboutput = undef;
+    } else {
+      chomp($joboutput);
+    }
   } else {
     Log (undef, "timed out while creating output collection");
+    foreach my $signal (2, 2, 2, 15, 15, 9) {
+      kill($signal, $pid);
+      last if waitpid($pid, WNOHANG) == -1;
+      sleep(1);
+    }
   }
-  # TODO: kill $pid instead of waiting, now that we've decided to
-  # ignore further output.
-  waitpid($pid, 0);
+  close($child_out);
 
   return $joboutput;
 }

-----------------------------------------------------------------------


hooks/post-receive
--