use strict;
use POSIX ':sys_wait_h';
+use POSIX qw(strftime);
use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK);
use Arvados;
use Digest::MD5 qw(md5_hex);
if ($treeish =~ m{^\d{1,4}$}) {
my $gitlog = `git --git-dir=\Q$repo\E log --pretty="format:%H" --grep="git-svn-id:.*\@"\Q$treeish\E" " master`;
chomp $gitlog;
- if ($gitlog =~ /^[a-f0-9]{40}$/) {
+ Log(undef, "git Subversion search exited $?");
+ if (($? == 0) && ($gitlog =~ /^[a-f0-9]{40}$/)) {
$commit = $gitlog;
- Log (undef, "Using commit $commit for script_version $treeish");
+ Log(undef, "Using commit $commit for Subversion revision $treeish");
}
}
if (!defined $commit) {
my $found = `git --git-dir=\Q$repo\E rev-list -1 ''\Q$treeish\E`;
chomp $found;
- if ($found =~ /^[0-9a-f]{40}$/s) {
+ Log(undef, "git rev-list exited $? with result '$found'");
+ if (($? == 0) && ($found =~ /^[0-9a-f]{40}$/s)) {
$commit = $found;
+ Log(undef, "Using commit $commit for tree-ish $treeish");
if ($commit ne $treeish) {
# Make sure we record the real commit id in the database,
# frozentokey, logs, etc. -- instead of an abbreviation or a
# branch name which can become ambiguous or point to a
# different commit in the future.
- $ENV{"CRUNCH_SRC_COMMIT"} = $commit;
- Log (undef, "Using commit $commit for tree-ish $treeish");
- if ($commit ne $treeish) {
- $Job->{'script_version'} = $commit;
- !$job_has_uuid or
- $Job->update_attributes('script_version' => $commit) or
- croak("Error while updating job");
- }
+ $Job->{'script_version'} = $commit;
+ !$job_has_uuid or
+ $Job->update_attributes('script_version' => $commit) or
+ croak("Error while updating job");
}
}
}
@execargs = ("sh", "-c",
"mkdir -p $ENV{CRUNCH_INSTALL} && cd $ENV{CRUNCH_TMP} && perl -");
$git_archive = `git --git-dir=\Q$repo\E archive ''\Q$commit\E`;
+ croak("git archive failed: exit " . ($? >> 8)) if ($? != 0);
}
else {
croak ("could not figure out commit id for $treeish");
}
+ # Note: this section is almost certainly unnecessary if we're
+ # running tasks in docker containers.
my $installpid = fork();
if ($installpid == 0)
{
$command .=
"&& perl -";
}
- $command .= "&& exec arv-mount --allow-other $ENV{TASK_KEEPMOUNT} --exec ";
+ $command .= "&& exec arv-mount --by-id --allow-other $ENV{TASK_KEEPMOUNT} --exec ";
if ($docker_hash)
{
$command .= "crunchstat -cgroup-root=/sys/fs/cgroup -cgroup-parent=docker -cgroup-cid=$ENV{TASK_WORK}/docker.cid -poll=10000 ";
while (my ($env_key, $env_val) = each %ENV)
{
if ($env_key =~ /^(ARVADOS|JOB|TASK)_/) {
- if ($env_key eq "TASK_KEEPMOUNT") {
+ if ($env_key eq "TASK_WORK") {
+ $command .= "--env=\QTASK_WORK=/tmp/crunch-job\E ";
+ }
+ elsif ($env_key eq "TASK_KEEPMOUNT") {
$command .= "--env=\QTASK_KEEPMOUNT=/keep\E ";
}
else {
delete $Jobstep->{stderr};
delete $Jobstep->{finishtime};
+ $Jobstep->{'arvados_task'}->{started_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{starttime});
+ $Jobstep->{'arvados_task'}->save;
+
splice @jobstep_todo, $todo_ptr, 1;
--$todo_ptr;
freeze();
my $collated_output = &collate_output();
-if ($job_has_uuid) {
- $Job->update_attributes('running' => 0,
- 'success' => $collated_output && $main::success,
- 'finished_at' => scalar gmtime)
-}
-
if (!$collated_output) {
Log(undef, "output undef");
}
eval {
open(my $orig_manifest, '-|', 'arv-get', $collated_output)
or die "failed to get collated manifest: $!";
- # Read the original manifest, and strip permission hints from it,
- # so we can put the result in a Collection.
- my @stripped_manifest_lines = ();
my $orig_manifest_text = '';
while (my $manifest_line = <$orig_manifest>) {
$orig_manifest_text .= $manifest_line;
- my @words = split(/ /, $manifest_line, -1);
- foreach my $ii (0..$#words) {
- if ($words[$ii] =~ /^[0-9a-f]{32}\+/) {
- $words[$ii] =~ s/\+A[0-9a-f]{40}@[0-9a-f]{8}\b//;
- }
- }
- push(@stripped_manifest_lines, join(" ", @words));
}
- my $stripped_manifest_text = join("", @stripped_manifest_lines);
my $output = $arv->{'collections'}->{'create'}->execute('collection' => {
- 'uuid' => md5_hex($stripped_manifest_text),
'manifest_text' => $orig_manifest_text,
});
- Log(undef, "output " . $output->{uuid});
- $Job->update_attributes('output' => $output->{uuid}) if $job_has_uuid;
- if ($Job->{'output_is_persistent'}) {
- $arv->{'links'}->{'create'}->execute('link' => {
- 'tail_kind' => 'arvados#user',
- 'tail_uuid' => $User->{'uuid'},
- 'head_kind' => 'arvados#collection',
- 'head_uuid' => $Job->{'output'},
- 'link_class' => 'resources',
- 'name' => 'wants',
- });
- }
+ Log(undef, "output uuid " . $output->{uuid});
+ Log(undef, "output hash " . $output->{portable_data_hash});
+ $Job->update_attributes('output' => $output->{portable_data_hash}) if $job_has_uuid;
};
if ($@) {
Log (undef, "Failed to register output manifest: $@");
Log (undef, "finish");
save_meta();
+
+if ($job_has_uuid) {
+ $Job->update_attributes('running' => 0,
+ 'success' => $collated_output && $main::success,
+ 'finished_at' => scalar gmtime)
+}
+
exit ($Job->{'success'} ? 1 : 0);
}
$Jobstep->{exitcode} = $childstatus;
$Jobstep->{finishtime} = time;
+ $Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
+ $Jobstep->{'arvados_task'}->save;
process_stderr ($jobstepid, $task_success);
Log ($jobstepid, "output " . $Jobstep->{'arvados_task'}->{output});
Log (undef, "collate");
my ($child_out, $child_in);
- my $pid = open2($child_out, $child_in, 'arv-put', '--raw');
+ my $pid = open2($child_out, $child_in, 'arv-put', '--raw',
+ '--retries', put_retry_count());
my $joboutput;
for (@jobstep)
{
return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm
$local_logfile->flush;
- my $cmd = "arv-put --filename ''\Q$keep_logfile\E "
- . quotemeta($local_logfile->filename);
+ my $retry_count = put_retry_count();
+ my $cmd = "arv-put --portable-data-hash --retries $retry_count " .
+ "--filename ''\Q$keep_logfile\E " . quotemeta($local_logfile->filename);
my $loglocator = `$cmd`;
die "system $cmd failed: $?" if $?;
chomp($loglocator);
# If so, return its stream name and Docker hash.
# If not, return undef for both values.
my $locator = shift;
+ my ($streamname, $filename);
if (my $image = $arv->{collections}->{get}->execute(uuid => $locator)) {
- my @file_list = @{$image->{files}};
- if ((scalar(@file_list) == 1) &&
- ($file_list[0][1] =~ /^([0-9A-Fa-f]{64})\.tar$/)) {
- return ($file_list[0][0], $1);
+ foreach my $line (split(/\n/, $image->{manifest_text})) {
+ my @tokens = split(/\s+/, $line);
+ next if (!@tokens);
+ $streamname = shift(@tokens);
+ foreach my $filedata (grep(/^\d+:\d+:/, @tokens)) {
+ if (defined($filename)) {
+ return (undef, undef); # More than one file in the Collection.
+ } else {
+ $filename = (split(/:/, $filedata, 3))[2];
+ }
+ }
}
}
- return (undef, undef);
+ if (defined($filename) and ($filename =~ /^([0-9A-Fa-f]{64})\.tar$/)) {
+ return ($streamname, $1);
+ } else {
+ return (undef, undef);
+ }
+}
+
+sub put_retry_count {
+ # Calculate a --retries argument for arv-put that will have it try
+ # approximately as long as this Job has been running.
+ my $stoptime = shift || time;
+ my $starttime = $jobstep[0]->{starttime};
+ my $timediff = defined($starttime) ? ($stoptime - $starttime) : 1;
+ my $retries = 0;
+ while ($timediff >= 2) {
+ $retries++;
+ $timediff /= 2;
+ }
+ return ($retries > 3) ? $retries : 3;
}
__DATA__
my $repo = $ENV{"CRUNCH_SRC_URL"};
my $task_work = $ENV{"TASK_WORK"};
-make_path $task_work or die "Failed to create temporary working directory ($task_work): $!";
+for my $dir ($destdir, $task_work) {
+ if ($dir) {
+ make_path $dir;
+ -e $dir or die "Failed to create temporary directory ($dir): $!";
+ }
+}
open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
flock L, LOCK_EX;
if (readlink ("$destdir.commit") eq $commit && -d $destdir) {
- exit 0;
+ if (@ARGV) {
+ exec(@ARGV);
+ die "Cannot exec `@ARGV`: $!";
+ } else {
+ exit 0;
+ }
}
unlink "$destdir.commit";
close L;
-exit 0;
+if (@ARGV) {
+ exec(@ARGV);
+ die "Cannot exec `@ARGV`: $!";
+} else {
+ exit 0;
+}
sub shell_or_die
{