arvados-cli 0.1.20150527201024 → 0.1.20150529160203

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA1:
-  metadata.gz: bafef169f074ca61d533298a5e462905d9764846
-  data.tar.gz: bd2e304a1cef255c8362650d82ed490d88d7a56d
+  metadata.gz: bf3f34e6509a6c3847bd333b357dc0289fe35ea3
+  data.tar.gz: 3b0d0d5db9a1db742ebcff2d09f03745e010d733
 SHA512:
-  metadata.gz: fb00c43e5d7d57f3b17f82efaf494426766b12d5af483195ce89252ec11d402c9dcc5149c3d5dc340f393a125d37c53d93e6ef90d7a60808f2124aa44e7c6688
-  data.tar.gz: d7879e42888f1bb9a0b4305f4d1b1d2e24acb2b0bbf64fde6be45b33f560aee935c1ab936d62e21ddd330467b05cb234032214e47deee6dc76432b775ea3ad9a
+  metadata.gz: c27558d79844645f535a5d406fe7a3db73a3b3e21e3453950b54065c392016c431a7d7bdf0b32fcf110d590b20d8372ba3b0bf7ac4e46ba73b6bdddc81f2de55
+  data.tar.gz: 273174a7e34f5675a0d4c6b62cfc46b20a45435776af812e7852345296f75ff76621cbc515ecd3830027781f71ab8f1e0c608f2500f590249bb4bf5f7618fafc

data/bin/crunch-job

@@ -125,12 +125,14 @@ my $jobspec;
 my $job_api_token;
 my $no_clear_tmp;
 my $resume_stash;
+my $docker_bin = "/usr/bin/docker.io";
 GetOptions('force-unlock' => \$force_unlock,
            'git-dir=s' => \$git_dir,
            'job=s' => \$jobspec,
            'job-api-token=s' => \$job_api_token,
            'no-clear-tmp' => \$no_clear_tmp,
            'resume-stash=s' => \$resume_stash,
+           'docker-bin=s' => \$docker_bin,
     );
 
 if (defined $job_api_token) {
@@ -138,7 +140,6 @@ if (defined $job_api_token) {
 }
 
 my $have_slurm = exists $ENV{SLURM_JOBID} && exists $ENV{SLURM_NODELIST};
-my $local_job = 0;
 
 
 $SIG{'USR1'} = sub
@@ -150,8 +151,6 @@ $SIG{'USR2'} = sub
   $main::ENV{CRUNCH_DEBUG} = 0;
 };
 
-
-
 my $arv = Arvados->new('apiVersion' => 'v1');
 
 my $Job;
@@ -160,12 +159,41 @@ my $dbh;
 my $sth;
 my @jobstep;
 
-my $User = api_call("users/current");
-
+my $local_job;
 if ($jobspec =~ /^[-a-z\d]+$/)
 {
   # $jobspec is an Arvados UUID, not a JSON job specification
   $Job = api_call("jobs/get", uuid => $jobspec);
+  $local_job = 0;
+}
+else
+{
+  $Job = JSON::decode_json($jobspec);
+  $local_job = 1;
+}
+
+
+# Make sure our workers (our slurm nodes, localhost, or whatever) are
+# at least able to run basic commands: they aren't down or severely
+# misconfigured.
+my $cmd = ['true'];
+if ($Job->{docker_image_locator}) {
+  $cmd = [$docker_bin, 'ps', '-q'];
+}
+Log(undef, "Sanity check is `@$cmd`");
+srun(["srun", "--nodes=\Q$ENV{SLURM_NNODES}\E", "--ntasks-per-node=1"],
+     $cmd,
+     {fork => 1});
+if ($? != 0) {
+  Log(undef, "Sanity check failed: ".exit_status_s($?));
+  exit EX_TEMPFAIL;
+}
+Log(undef, "Sanity check OK");
+
+
+my $User = api_call("users/current");
+
+if (!$local_job) {
   if (!$force_unlock) {
     # Claim this job, and make sure nobody else does
     eval { api_call("jobs/lock", uuid => $Job->{uuid}); };
@@ -177,8 +205,6 @@ if ($jobspec =~ /^[-a-z\d]+$/)
 }
 else
 {
-  $Job = JSON::decode_json($jobspec);
-
   if (!$resume_stash)
   {
     map { croak ("No $_ specified") unless $Job->{$_} }
@@ -376,7 +402,6 @@ if (!defined $no_clear_tmp) {
 }
 
 # If this job requires a Docker image, install that.
-my $docker_bin = "/usr/bin/docker.io";
 my ($docker_locator, $docker_stream, $docker_hash, $docker_limitmem);
 if ($docker_locator = $Job->{docker_image_locator}) {
   ($docker_stream, $docker_hash) = find_docker_image($docker_locator);
@@ -1139,7 +1164,7 @@ sub reapchildren
 
     Log ($jobstepid, sprintf('failure (#%d, %s) after %d seconds',
                              ++$Jobstep->{'failures'},
-                             $temporary_fail ? 'temporary ' : 'permanent',
+                             $temporary_fail ? 'temporary' : 'permanent',
                              $elapsed));
 
     if (!$temporary_fail || $Jobstep->{'failures'} >= 3) {
@@ -1708,7 +1733,13 @@ sub srun
   my $show_cmd = Dumper($args);
   $show_cmd =~ s/(TOKEN\\*=)[^\s\']+/${1}[...]/g;
   $show_cmd =~ s/\n/ /g;
-  warn "starting: $show_cmd\n";
+  if ($opts->{fork}) {
+    Log(undef, "starting: $show_cmd");
+  } else {
+    # This is a child process: parent is in charge of reading our
+    # stderr and copying it to Log() if needed.
+    warn "starting: $show_cmd\n";
+  }
 
   if (defined $stdin) {
     my $child = open STDIN, "-|";
@@ -2052,6 +2083,7 @@ if (readlink ("$destdir.archive_hash") eq $archive_hash && -d $destdir) {
 unlink "$destdir.archive_hash";
 mkdir $destdir;
 
+warn "Extracting archive: $archive_hash\n";
 if (!open(TARX, "|-", "tar", "-xC", $destdir)) {
   die "Error launching 'tar -xC $destdir': $!";
 }

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: arvados-cli
 version: !ruby/object:Gem::Version
-  version: 0.1.20150527201024
+  version: 0.1.20150529160203
 platform: ruby
 authors:
 - Arvados Authors
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2015-05-27 00:00:00.000000000 Z
+date: 2015-05-29 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: arvados
@@ -178,7 +178,7 @@ dependencies:
     - - "<"
       - !ruby/object:Gem::Version
         version: 1.0.0
-description: Arvados command line tools, git commit 7a53d874994a5a9af273cee1329d9635b7e03edb
+description: Arvados command line tools, git commit 1ec1d552c77e18e2912e400ae395ca00f4e51c3c
 email: gem-dev@curoverse.com
 executables:
 - arv