2882: Added crunchstat service to collect cgroup stats, and added support to

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

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index 14b1be9a99dabc11e0c82bfc1be3755885dddddc..da005a3884a2b984c185d9134576fefe71f234aa 100755 (executable)
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -139,7 +139,7 @@ $SIG{'USR2'} = sub
 
 
 my $arv = Arvados->new('apiVersion' => 'v1');
-my $metastream;
+my $local_logfile;
 
 my $User = $arv->{'users'}->{'current'}->execute;
 
@@ -185,7 +185,7 @@ else
 $job_id = $Job->{'uuid'};
 
 my $keep_logfile = $job_id . '.log.txt';
-my $local_logfile = File::Temp->new();
+$local_logfile = File::Temp->new();
 
 $Job->{'runtime_constraints'} ||= {};
 $Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0;
@@ -517,7 +517,7 @@ if ($docker_image) {
   }
   # If the Docker image was specified as a hash, pull will fail.
   # Ignore that error.  We'll see what happens when we try to run later.
-  if (($? != 0) && ($docker_image !~ /^[0-9a-fA-F]{5,}$/))
+  if (($? != 0) && ($docker_image !~ /^[0-9a-fA-F]{5,64}$/))
   {
     croak("Installing Docker image $docker_image returned exit code $?");
   }
@@ -640,7 +640,8 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
     $command .= "&& exec arv-mount --allow-other $ENV{TASK_KEEPMOUNT} --exec ";
     if ($docker_image)
     {
-      $command .= "$docker_bin run -i -a stdin -a stdout -a stderr ";
+      $command .= "crunchstat -cgroup-parent=/sys/fs/cgroup/lxc -cgroup-cid=$ENV{TASK_WORK}/docker.cid ";
+      $command .= "$docker_bin run -i -a stdin -a stdout -a stderr -cidfile=$ENV{TASK_WORK}/docker.cid ";
       # Dynamically configure the container to use the host system as its
       # DNS server.  Get the host's global addresses from the ip command,
       # and turn them into docker --dns options using gawk.
@@ -653,9 +654,13 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
       }
       while (my ($env_key, $env_val) = each %ENV)
       {
-        $command .= "-e \Q$env_key=$env_val\E ";
+        if ($env_key =~ /^(JOB|TASK)_/) {
+          $command .= "-e \Q$env_key=$env_val\E ";
+        }
       }
-      $command .= "$docker_image ";
+      $command .= "\Q$docker_image\E ";
+    } else {
+      $command .= "crunchstat -cgroup-path=/sys/fs/cgroup "
     }
     $command .= "$ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
     my @execargs = ('bash', '-c', $command);
@@ -946,13 +951,19 @@ sub reapchildren
   delete $proc{$pid};
 
   # Load new tasks
-  my $newtask_list = $arv->{'job_tasks'}->{'list'}->execute(
-    'where' => {
-      'created_by_job_task_uuid' => $Jobstep->{'arvados_task'}->{uuid}
-    },
-    'order' => 'qsequence'
-  );
-  foreach my $arvados_task (@{$newtask_list->{'items'}}) {
+  my $newtask_list = [];
+  my $newtask_results;
+  do {
+    $newtask_results = $arv->{'job_tasks'}->{'list'}->execute(
+      'where' => {
+        'created_by_job_task_uuid' => $Jobstep->{'arvados_task'}->{uuid}
+      },
+      'order' => 'qsequence',
+      'offset' => scalar(@$newtask_list),
+    );
+    push(@$newtask_list, @{$newtask_results->{items}});
+  } while (@{$newtask_results->{items}});
+  foreach my $arvados_task (@$newtask_list) {
     my $jobstep = {
       'level' => $arvados_task->{'sequence'},
       'failures' => 0,
@@ -1245,15 +1256,15 @@ sub Log                         # ($jobstep_id, $logmessage)
   $message =~ s{([^ -\176])}{"\\" . sprintf ("%03o", ord($1))}ge;
   $message .= "\n";
   my $datetime;
-  if ($metastream || -t STDERR) {
+  if ($local_logfile || -t STDERR) {
     my @gmtime = gmtime;
     $datetime = sprintf ("%04d-%02d-%02d_%02d:%02d:%02d",
                         $gmtime[5]+1900, $gmtime[4]+1, @gmtime[3,2,1,0]);
   }
   print STDERR ((-t STDERR) ? ($datetime." ".$message) : $message);
 
-  if ($metastream) {
-    print $metastream $datetime . " " . $message;
+  if ($local_logfile) {
+    print $local_logfile $datetime . " " . $message;
   }
 }
 
@@ -1266,7 +1277,7 @@ sub croak
   freeze() if @jobstep_todo;
   collate_output() if @jobstep_todo;
   cleanup();
-  save_meta() if $metastream;
+  save_meta() if $local_logfile;
   die;
 }
 
@@ -1290,6 +1301,7 @@ sub save_meta
       . quotemeta($local_logfile->filename);
   my $loglocator = `$cmd`;
   die "system $cmd failed: $?" if $?;
+  chomp($loglocator);
 
   $local_logfile = undef;   # the temp file is automatically deleted
   Log (undef, "log manifest is $loglocator");