X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/4727a3b6417bd254b3e8bca1bd19a77773493572..91b379c4a76b8278354903158a65e5d47babd363:/sdk/cli/bin/crunch-job

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index 88f9d561b4..6423c1cf79 100755
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -109,9 +109,6 @@ unless (defined $ENV{"CRUNCH_TMP"}) {
   }
 }
 
-$ENV{"HOST_CRUNCHRUNNER_BIN"} ||= `which crunchrunner`;
-$ENV{"HOST_CERTS"} ||= "/etc/ssl/certs/ca-certificates.crt";
-
 # Create the tmp directory if it does not exist
 if ( ! -d $ENV{"CRUNCH_TMP"} ) {
   make_path $ENV{"CRUNCH_TMP"} or die "Failed to create temporary working directory: " . $ENV{"CRUNCH_TMP"};
@@ -855,7 +852,13 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
         .q{&& MEM=$(awk '($1 == "MemTotal:"){print $2}' </proc/meminfo) }
         .q{&& SWAP=$(awk '($1 == "SwapTotal:"){print $2}' </proc/meminfo) }
         ."&& MEMLIMIT=\$(( (\$MEM * 95) / ($ENV{CRUNCH_NODE_SLOTS} * 100) )) "
-        ."&& let SWAPLIMIT=\$MEMLIMIT+\$SWAP ";
+        ."&& let SWAPLIMIT=\$MEMLIMIT+\$SWAP "
+        # $VOLUME_CRUNCHRUNNER and $VOLUME_CERTS will be passed unquoted as
+        # arguments to `docker run`.  They must contain their own quoting.
+        .q{&& VOLUME_CRUNCHRUNNER="" VOLUME_CERTS="" }
+        .q{&& if which crunchrunner >/dev/null ; then VOLUME_CRUNCHRUNNER=\\"--volume=$(which crunchrunner):/usr/local/bin/crunchrunner\\" ; fi }
+        .q{&& if test -f /etc/ssl/certs/ca-certificates.crt ; then VOLUME_CERTS=\\"--volume=/etc/ssl/certs/ca-certificates.crt:/etc/arvados/ca-certificates.crt\\" ; }
+        .q{elif test -f /etc/pki/tls/certs/ca-bundle.crt ; then VOLUME_CERTS=\\"--volume=/etc/pki/tls/certs/ca-bundle.crt:/etc/arvados/ca-certificates.crt\\" ; fi };
 
     $command .= "&& exec arv-mount --read-write --mount-by-pdh=by_pdh --mount-tmp=tmp --crunchstat-interval=10 --allow-other $arv_file_cache \Q$keep_mnt\E --exec ";
     $ENV{TASK_KEEPMOUNT} = "$keep_mnt/by_pdh";
@@ -922,8 +925,7 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
 
       # Bind mount the crunchrunner binary and host TLS certificates file into
       # the container.
-      $command .= "--volume=\Q$ENV{HOST_CRUNCHRUNNER_BIN}:/usr/lib/crunchrunner/crunchrunner\E ";
-      $command .= "--volume=\Q$ENV{HOST_CERTS}:/usr/lib/crunchrunner/ca-certificates.crt\E ";
+      $command .= "\$VOLUME_CRUNCHRUNNER \$VOLUME_CERTS ";
 
       while (my ($env_key, $env_val) = each %ENV)
       {
@@ -1153,13 +1155,6 @@ sub reapchildren
                     . $slot[$proc{$pid}->{slot}]->{cpu});
     my $jobstepidx = $proc{$pid}->{jobstepidx};
 
-    if (!WIFEXITED($childstatus))
-    {
-      # child did not exit (may be temporarily stopped)
-      Log ($jobstepidx, "child $pid did not actually exit in reapchildren, ignoring for now.");
-      next;
-    }
-
     $children_reaped++;
     my $elapsed = time - $proc{$pid}->{time};
     my $Jobstep = $jobstep[$jobstepidx];
@@ -1467,6 +1462,9 @@ sub readfrompipes
 sub preprocess_stderr
 {
   my $jobstepidx = shift;
+  # slotindex is only defined for children running Arvados job tasks.
+  # Be prepared to handle the undef case (for setup srun calls, etc.).
+  my $job_slot_index = $jobstep[$jobstepidx]->{slotindex};
 
   while ($jobstep[$jobstepidx]->{stderr} =~ /^(.*?)\n/) {
     my $line = $1;
@@ -1476,19 +1474,16 @@ sub preprocess_stderr
       # whoa.
       $main::please_freeze = 1;
     }
-    elsif (!exists $jobstep[$jobstepidx]->{slotindex}) {
-      # Skip the following tempfail checks if this srun proc isn't
-      # attached to a particular worker slot.
-    }
     elsif ($line =~ /srun: error: (Node failure on|Aborting, .*\bio error\b)/) {
-      my $job_slot_index = $jobstep[$jobstepidx]->{slotindex};
-      $slot[$job_slot_index]->{node}->{fail_count}++;
       $jobstep[$jobstepidx]->{tempfail} = 1;
-      ban_node_by_slot($job_slot_index);
+      if (defined($job_slot_index)) {
+        $slot[$job_slot_index]->{node}->{fail_count}++;
+        ban_node_by_slot($job_slot_index);
+      }
     }
     elsif ($line =~ /srun: error: (Unable to create job step|.*: Communication connection failure)/) {
       $jobstep[$jobstepidx]->{tempfail} = 1;
-      ban_node_by_slot($jobstep[$jobstepidx]->{slotindex});
+      ban_node_by_slot($job_slot_index) if (defined($job_slot_index));
     }
     elsif ($line =~ /\bKeep(Read|Write|Request)Error:/) {
       $jobstep[$jobstepidx]->{tempfail} = 1;
@@ -1978,6 +1973,11 @@ sub srun_sync
   delete $reader{$jobstepidx};
 
   my $j = pop @jobstep;
+  # If the srun showed signs of tempfail, ensure the caller treats that as a
+  # failure case.
+  if ($main::please_freeze || $j->{tempfail}) {
+    $exited ||= 255;
+  }
   return ($exited, $j->{stdout_captured}, $j->{stderr_captured});
 }