Merge branch '5524-crunch-temp-keep-errors' closes #5524

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

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index b74d7af427ce0bc874454d3375082007abd2dbd9..294696cdcf4702ea94569fc598253f1bcd6ad4c7 100755 (executable)
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -719,13 +719,15 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
     my $command =
        "if [ -e $ENV{TASK_WORK} ]; then rm -rf $ENV{TASK_WORK}; fi; "
         ."mkdir -p $ENV{CRUNCH_TMP} $ENV{JOB_WORK} $ENV{TASK_WORK} $ENV{TASK_KEEPMOUNT} "
-       ."&& cd $ENV{CRUNCH_TMP} ";
+       ."&& cd $ENV{CRUNCH_TMP} "
+        ."&& MEM=\$(cat /proc/meminfo | grep MemTotal | sed 's/\\s\\s*/ /g' |cut -d' ' -f2) "
+        ."&& MEMLIMIT=\$(( (\$MEM * 95) / ($ENV{CRUNCH_NODE_SLOTS} * 100) )) ";
     $command .= "&& exec arv-mount --by-id --allow-other $ENV{TASK_KEEPMOUNT} --exec ";
     if ($docker_hash)
     {
       my $cidfile = "$ENV{CRUNCH_TMP}/$Jobstep->{arvados_task}->{uuid}-$Jobstep->{failures}.cid";
       $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -cgroup-parent=docker -cgroup-cid=$cidfile -poll=10000 ";
-      $command .= "$docker_bin run --rm=true --attach=stdout --attach=stderr --attach=stdin -i --user=crunch --cidfile=$cidfile --sig-proxy ";
+      $command .= "$docker_bin run --rm=true --attach=stdout --attach=stderr --attach=stdin -i --user=crunch --cidfile=$cidfile --sig-proxy --memory=\${MEMLIMIT}k ";
 
       # Dynamically configure the container to use the host system as its
       # DNS server.  Get the host's global addresses from the ip command,
@@ -1233,7 +1235,7 @@ sub preprocess_stderr
       # whoa.
       $main::please_freeze = 1;
     }
-    elsif ($line =~ /srun: error: (Node failure on|Unable to create job step|.*: Communication connection failure)/) {
+    elsif ($line =~ /(srun: error: (Node failure on|Unable to create job step|.*: Communication connection failure))|arvados.errors.Keep/) {
       $jobstep[$job]->{node_fail} = 1;
       ban_node_by_slot($jobstep[$job]->{slotindex});
     }