Merge branch 'master' into 2221-complete-docker

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

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index 2b31531a52fab57543ff5432a5285f8070c7aa40..40b0033757c8aeb03960ff397d9f311bca21314b 100755 (executable)
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -33,6 +33,12 @@ Path to .git directory where the specified commit is found.
 
 Arvados API authorization token to use during the course of the job.
 
+=item --no-clear-tmp
+
+Do not clear per-job/task temporary directories during initial job
+setup. This can speed up development and debugging when running jobs
+locally.
+
 =back
 
 =head1 RUNNING JOBS LOCALLY
@@ -74,6 +80,7 @@ use Getopt::Long;
 use IPC::Open2;
 use IO::Select;
 use File::Temp;
+use Fcntl ':flock';
 
 $ENV{"TMPDIR"} ||= "/tmp";
 unless (defined $ENV{"CRUNCH_TMP"}) {
@@ -92,11 +99,13 @@ my $force_unlock;
 my $git_dir;
 my $jobspec;
 my $job_api_token;
+my $no_clear_tmp;
 my $resume_stash;
 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,
     );
 
@@ -320,6 +329,12 @@ else
 }
 
 
+if (!$have_slurm)
+{
+  must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
+}
+
+
 my $build_script;
 
 
@@ -328,6 +343,11 @@ $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{script_version};
 my $skip_install = ($local_job && $Job->{script_version} =~ m{^/});
 if ($skip_install)
 {
+  if (!defined $no_clear_tmp) {
+    my $clear_tmp_cmd = 'rm -rf $JOB_WORK $CRUNCH_TMP/opt $CRUNCH_TMP/src*';
+    system($clear_tmp_cmd) == 0
+       or croak ("`$clear_tmp_cmd` failed: ".($?>>8));
+  }
   $ENV{"CRUNCH_SRC"} = $Job->{script_version};
   for my $src_path ("$ENV{CRUNCH_SRC}/arvados/sdk/python") {
     if (-d $src_path) {
@@ -348,22 +368,24 @@ else
   Log (undef, "Install revision ".$Job->{script_version});
   my $nodelist = join(",", @node);
 
-  # Clean out crunch_tmp/work, crunch_tmp/opt, crunch_tmp/src*
+  if (!defined $no_clear_tmp) {
+    # Clean out crunch_tmp/work, crunch_tmp/opt, crunch_tmp/src*
 
-  my $cleanpid = fork();
-  if ($cleanpid == 0)
-  {
-    srun (["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
-         ['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 $CRUNCH_TMP/src*']);
-    exit (1);
-  }
-  while (1)
-  {
-    last if $cleanpid == waitpid (-1, WNOHANG);
-    freeze_if_want_freeze ($cleanpid);
-    select (undef, undef, undef, 0.1);
+    my $cleanpid = fork();
+    if ($cleanpid == 0)
+    {
+      srun (["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
+           ['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 $CRUNCH_TMP/src*']);
+      exit (1);
+    }
+    while (1)
+    {
+      last if $cleanpid == waitpid (-1, WNOHANG);
+      freeze_if_want_freeze ($cleanpid);
+      select (undef, undef, undef, 0.1);
+    }
+    Log (undef, "Clean-work-dir exited $?");
   }
-  Log (undef, "Clean-work-dir exited $?");
 
   # Install requested code version
 
@@ -460,6 +482,12 @@ else
   Log (undef, "Install exited $?");
 }
 
+if (!$have_slurm)
+{
+  # Grab our lock again (we might have deleted and re-created CRUNCH_TMP above)
+  must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
+}
+
 
 
 foreach (qw (script script_version script_parameters runtime_constraints))
@@ -551,7 +579,7 @@ 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_WORK"} = $ENV{"JOB_WORK"}."/".$slot[$childslot]->{cpu};
+    $ENV{"TASK_WORK"} = $ENV{"JOB_WORK"}."/$id.$$";
     $ENV{"TASK_KEEPMOUNT"} = $ENV{"TASK_WORK"}."/keep";
     $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
     $ENV{"CRUNCH_NODE_SLOTS"} = $slot[$childslot]->{node}->{ncpus};
@@ -1304,6 +1332,15 @@ sub ban_node_by_slot {
   Log (undef, "backing off node " . $slot[$slotid]->{node}->{name} . " for 60 seconds");
 }
 
+sub must_lock_now
+{
+  my ($lockfile, $error_message) = @_;
+  open L, ">", $lockfile or croak("$lockfile: $!");
+  if (!flock L, LOCK_EX|LOCK_NB) {
+    croak("Can't lock $lockfile: $error_message\n");
+  }
+}
+
 __DATA__
 #!/usr/bin/perl