Merge branch '6321-slurm-oserror' closes #6321

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

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index 33e775ff148a0266ad017182e3e002111d3ce196..5539012c49e19e2b9afbeba26850aabbec9405d1 100755 (executable)
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -390,12 +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', '-o', 'pipefail', '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)
@@ -980,14 +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();
     }
@@ -1094,8 +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;
@@ -2205,11 +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");
@@ -2218,7 +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);