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
use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK);
use Arvados;
use Getopt::Long;
-use Warehouse;
-use Warehouse::Stream;
-use IPC::System::Simple qw(capturex);
+use IPC::Open2;
+use IO::Select;
+use File::Temp;
+use Fcntl ':flock';
$ENV{"TMPDIR"} ||= "/tmp";
unless (defined $ENV{"CRUNCH_TMP"}) {
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,
);
}
$job_id = $Job->{'uuid'};
-$metastream = Warehouse::Stream->new(whc => new Warehouse);
-$metastream->clear;
-$metastream->name('.');
-$metastream->write_start($job_id . '.log.txt');
-
+my $keep_logfile = $job_id . '.log.txt';
+my $local_logfile = File::Temp->new();
$Job->{'runtime_constraints'} ||= {};
$Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0;
}
+if (!$have_slurm)
+{
+ must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
+}
+
+
my $build_script;
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 $src_path) {
Log (undef, "Install revision ".$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
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))
}
$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};
if ($Job->{'output'})
{
eval {
- my $manifest_text = capturex("whget", $Job->{'output'});
+ my $manifest_text = `arv keep get \Q$Job->{'output'}\E`;
$arv->{'collections'}->{'create'}->execute('collection' => {
'uuid' => $Job->{'output'},
'manifest_text' => $manifest_text,
} split ("\n", $jobstep[$job]->{stderr});
}
+sub fetch_block
+{
+ my $hash = shift;
+ my ($child_out, $output_block);
+
+ my $cmd = "arv keep get \Q$hash\E";
+ open($keep, '-|', $cmd) or die "fetch_block: $cmd: $!"
+ sysread($keep, $output_block, 64 * 1024 * 1024);
+ close $keep;
+ return $output_block;
+}
sub collate_output
{
- my $whc = Warehouse->new;
Log (undef, "collate");
- $whc->write_start (1);
+
+ my ($child_out, $child_in);
+ my $pid = open2($child_out, $child_in, 'arv', 'keep', 'put', '--raw');
my $joboutput;
for (@jobstep)
{
if ($output !~ /^[0-9a-f]{32}(\+\S+)*$/)
{
$output_in_keep ||= $output =~ / [0-9a-f]{32}\S*\+K/;
- $whc->write_data ($output);
+ print $child_in $output;
}
elsif (@jobstep == 1)
{
$joboutput = $output;
- $whc->write_finish;
+ last;
}
- elsif (defined (my $outblock = $whc->fetch_block ($output)))
+ elsif (defined (my $outblock = fetch_block ($output)))
{
$output_in_keep ||= $outblock =~ / [0-9a-f]{32}\S*\+K/;
- $whc->write_data ($outblock);
+ print $child_in $outblock;
}
else
{
- my $errstr = $whc->errstr;
- $whc->write_data ("XXX fetch_block($output) failed: $errstr XXX\n");
+ Log (undef, "XXX fetch_block($output) failed XXX");
$main::success = 0;
}
}
- $joboutput = $whc->write_finish if !defined $joboutput;
+ $child_in->close;
+
+ if (!defined $joboutput) {
+ my $s = IO::Select->new($child_out);
+ if ($s->can_read(120)) {
+ sysread($child_out, $joboutput, 64 * 1024 * 1024);
+ } else {
+ Log (undef, "timed out reading from 'arv keep put'");
+ }
+ }
+ waitpid($pid, 0);
+
if ($joboutput)
{
Log (undef, "output $joboutput");
}
print STDERR ((-t STDERR) ? ($datetime." ".$message) : $message);
- return if !$metastream;
- $metastream->write_data ($datetime . " " . $message);
+ if ($metastream) {
+ print $metastream $datetime . " " . $message;
+ }
}
sub save_meta
{
my $justcheckpoint = shift; # false if this will be the last meta saved
- my $m = $metastream;
- $m = $m->copy if $justcheckpoint;
- $m->write_finish;
- my $whc = Warehouse->new;
- my $loglocator = $whc->store_block ($m->as_string);
- $arv->{'collections'}->{'create'}->execute('collection' => {
- 'uuid' => $loglocator,
- 'manifest_text' => $m->as_string,
- });
- undef $metastream if !$justcheckpoint; # otherwise Log() will try to use it
+ return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm
+
+ $local_logfile->flush;
+ my $cmd = "arv keep put --filename \Q$keep_logfile\E "
+ . quotemeta($local_logfile->filename);
+ my $loglocator = `$cmd`;
+ die "system $cmd failed: $?" if $?;
+
+ $local_logfile = undef; # the temp file is automatically deleted
Log (undef, "log manifest is $loglocator");
$Job->{'log'} = $loglocator;
$Job->update_attributes('log', $loglocator) if $job_has_uuid;
sub thaw
{
croak ("Thaw not implemented");
-
- my $whc;
- my $key = shift;
- Log (undef, "thaw from $key");
-
- @jobstep = ();
- @jobstep_done = ();
- @jobstep_todo = ();
- @jobstep_tomerge = ();
- $jobstep_tomerge_level = 0;
- my $frozenjob = {};
-
- my $stream = new Warehouse::Stream ( whc => $whc,
- hash => [split (",", $key)] );
- $stream->rewind;
- while (my $dataref = $stream->read_until (undef, "\n\n"))
- {
- if ($$dataref =~ /^job /)
- {
- foreach (split ("\n", $$dataref))
- {
- my ($k, $v) = split ("=", $_, 2);
- $frozenjob->{$k} = freezeunquote ($v);
- }
- next;
- }
-
- if ($$dataref =~ /^merge (\d+) (.*)/)
- {
- $jobstep_tomerge_level = $1;
- @jobstep_tomerge
- = map { freezeunquote ($_) } split ("\n", freezeunquote($2));
- next;
- }
-
- my $Jobstep = { };
- foreach (split ("\n", $$dataref))
- {
- my ($k, $v) = split ("=", $_, 2);
- $Jobstep->{$k} = freezeunquote ($v) if $k;
- }
- $Jobstep->{'failures'} = 0;
- push @jobstep, $Jobstep;
-
- if ($Jobstep->{exitcode} eq "0")
- {
- push @jobstep_done, $#jobstep;
- }
- else
- {
- push @jobstep_todo, $#jobstep;
- }
- }
-
- foreach (qw (script script_version script_parameters))
- {
- $Job->{$_} = $frozenjob->{$_};
- }
- $Job->save if $job_has_uuid;
}
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
projects / arvados.git / blobdiff