add util functions, fix up tmp dirs

[arvados.git] / services / crunch / crunch-job

diff --git a/services/crunch/crunch-job b/services/crunch/crunch-job
index 30bc668280905d0bbab89e011f40c37a0096b376..6c0031c6171bb8135bd06a3c110af3c1ae5667ad 100755 (executable)
--- a/services/crunch/crunch-job
+++ b/services/crunch/crunch-job
@@ -76,8 +76,13 @@ use IPC::System::Simple qw(capturex);
 
 $ENV{"TMPDIR"} ||= "/tmp";
 $ENV{"CRUNCH_TMP"} = $ENV{"TMPDIR"} . "/crunch-job";
-$ENV{"CRUNCH_WORK"} = $ENV{"CRUNCH_TMP"} . "/work";
-mkdir ($ENV{"CRUNCH_TMP"});
+if ($ENV{"USER"} ne "crunch" && $< != 0) {
+  # use a tmp dir unique for my uid
+  $ENV{"CRUNCH_TMP"} .= "-$<";
+}
+$ENV{"JOB_WORK"} = $ENV{"CRUNCH_TMP"} . "/work";
+$ENV{"CRUNCH_TMP"} = $ENV{"JOB_WORK"}; # deprecated
+mkdir ($ENV{"JOB_WORK"});
 
 my $force_unlock;
 my $git_dir;
@@ -97,6 +102,7 @@ if (defined $job_api_token) {
 
 my $have_slurm = exists $ENV{SLURM_JOBID} && exists $ENV{SLURM_NODELIST};
 my $job_has_uuid = $jobspec =~ /^[-a-z\d]+$/;
+my $local_job = !$job_has_uuid;
 
 
 $SIG{'HUP'} = sub
@@ -119,7 +125,8 @@ my $metastream = Warehouse::Stream->new(whc => new Warehouse);
 $metastream->clear;
 $metastream->write_start('log.txt');
 
-my $User = {};
+my $User = $arv->{'users'}->{'current'}->execute;
+
 my $Job = {};
 my $job_id;
 my $dbh;
@@ -127,10 +134,9 @@ my $sth;
 if ($job_has_uuid)
 {
   $Job = $arv->{'jobs'}->{'get'}->execute('uuid' => $jobspec);
-  $User = $arv->{'users'}->{'current'}->execute;
   if (!$force_unlock) {
-    if ($Job->{'is_locked_by'}) {
-      croak("Job is locked: " . $Job->{'is_locked_by'});
+    if ($Job->{'is_locked_by_uuid'}) {
+      croak("Job is locked: " . $Job->{'is_locked_by_uuid'});
     }
     if ($Job->{'success'} ne undef) {
       croak("Job 'success' flag (" . $Job->{'success'} . ") is not null");
@@ -153,11 +159,12 @@ else
     qw(script script_version script_parameters);
   }
 
-  if (!defined $Job->{'uuid'}) {
-    my $hostname = `hostname -s`;
-    chomp $hostname;
-    $Job->{'uuid'} = sprintf ("%s-t%d-p%d", $hostname, time, $$);
-  }
+  $Job->{'is_locked_by_uuid'} = $User->{'uuid'};
+  $Job->{'started_at'} = gmtime;
+
+  $Job = $arv->{'jobs'}->{'create'}->execute('job' => $Job);
+
+  $job_has_uuid = 1;
 }
 $job_id = $Job->{'uuid'};
 
@@ -249,7 +256,7 @@ if ($job_has_uuid)
 {
   # Claim this job, and make sure nobody else does
 
-  $Job->{'is_locked_by'} = $User->{'uuid'};
+  $Job->{'is_locked_by_uuid'} = $User->{'uuid'};
   $Job->{'started_at'} = gmtime;
   $Job->{'running'} = 1;
   $Job->{'success'} = undef;
@@ -258,7 +265,7 @@ if ($job_has_uuid)
                               'running' => 0,
                               'done' => 0 };
   if ($job_has_uuid) {
-    unless ($Job->save() && $Job->{'is_locked_by'} == $User->{'uuid'}) {
+    unless ($Job->save() && $Job->{'is_locked_by_uuid'} == $User->{'uuid'}) {
       croak("Error while updating / locking job");
     }
   }
@@ -314,21 +321,21 @@ else
 
 
 my $build_script;
-do {
-  local $/ = undef;
-  $build_script = <DATA>;
-};
 
 
 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{script_version};
 
-my $skip_install = (!$job_has_uuid && $Job->{script_version} =~ m{^/});
+my $skip_install = ($local_job && $Job->{script_version} =~ m{^/});
 if ($skip_install)
 {
   $ENV{"CRUNCH_SRC"} = $Job->{script_version};
 }
 else
 {
+  do {
+    local $/ = undef;
+    $build_script = <DATA>;
+  };
   Log (undef, "Install revision ".$Job->{script_version});
   my $nodelist = join(",", @node);
 
@@ -338,7 +345,7 @@ else
   if ($cleanpid == 0)
   {
     srun (["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
-         ['bash', '-c', 'if mount | grep -q $CRUNCH_WORK/; then sudo /bin/umount $CRUNCH_WORK/* 2>/dev/null; fi; sleep 1; rm -rf $CRUNCH_WORK $CRUNCH_TMP/opt']);
+         ['bash', '-c', 'if mount | grep -q $JOB_WORK/; then sudo /bin/umount $JOB_WORK/* 2>/dev/null; fi; sleep 1; rm -rf $JOB_WORK $CRUNCH_TMP/opt']);
     exit (1);
   }
   while (1)
@@ -542,7 +549,8 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
     }
     $ENV{"TASK_SLOT_NODE"} = $slot[$childslot]->{node}->{name};
     $ENV{"TASK_SLOT_NUMBER"} = $slot[$childslot]->{cpu};
-    $ENV{"TASK_TMPDIR"} = $ENV{"CRUNCH_WORK"}.$slot[$childslot]->{cpu};
+    $ENV{"TASK_WORK"} = $ENV{"JOB_WORK"}."/".$slot[$childslot]->{cpu};
+    $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
     $ENV{"CRUNCH_NODE_SLOTS"} = $slot[$childslot]->{node}->{ncpus};
 
     $ENV{"GZIP"} = "-n";
@@ -556,7 +564,7 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
     my @execargs = qw(sh);
     my $build_script_to_send = "";
     my $command =
-       "mkdir -p $ENV{CRUNCH_TMP}/revision "
+       "mkdir -p $ENV{JOB_WORK} $ENV{CRUNCH_TMP} $ENV{TASK_WORK} "
        ."&& cd $ENV{CRUNCH_TMP} ";
     if ($build_script)
     {
@@ -704,8 +712,11 @@ goto ONELEVEL if !defined $success;
 
 release_allocation();
 freeze();
+$Job->reload;
 $Job->{'output'} = &collate_output();
+$Job->{'running'} = 0;
 $Job->{'success'} = $Job->{'output'} && $success;
+$Job->{'finished_at'} = gmtime;
 $Job->save if $job_has_uuid;
 
 if ($Job->{'output'})
@@ -824,7 +835,7 @@ sub reapchildren
   # Load new tasks
   my $newtask_list = $arv->{'job_tasks'}->{'list'}->execute(
     'where' => {
-      'created_by_job_task' => $Jobstep->{'arvados_task'}->{uuid}
+      'created_by_job_task_uuid' => $Jobstep->{'arvados_task'}->{uuid}
     },
     'order' => 'qsequence'
   );
@@ -1106,11 +1117,11 @@ sub croak
 sub cleanup
 {
   return if !$job_has_uuid;
-  $Job->reload if $job_has_uuid;
+  $Job->reload;
   $Job->{'running'} = 0;
   $Job->{'success'} = 0;
   $Job->{'finished_at'} = gmtime;
-  $Job->save if $job_has_uuid;
+  $Job->save;
 }