use POSIX ':sys_wait_h';
use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK);
use Arvados;
+use Digest::MD5 qw(md5_hex);
use Getopt::Long;
use IPC::Open2;
use IO::Select;
use File::Temp;
use Fcntl ':flock';
+use File::Path qw( make_path );
+
+use constant EX_TEMPFAIL => 75;
$ENV{"TMPDIR"} ||= "/tmp";
unless (defined $ENV{"CRUNCH_TMP"}) {
$ENV{"CRUNCH_TMP"} .= "-$<";
}
}
+
+# Create the tmp directory if it does not exist
+if ( ! -d $ENV{"CRUNCH_TMP"} ) {
+ make_path $ENV{"CRUNCH_TMP"} or die "Failed to create temporary working directory: " . $ENV{"CRUNCH_TMP"};
+}
+
$ENV{"JOB_WORK"} = $ENV{"CRUNCH_TMP"} . "/work";
$ENV{"CRUNCH_INSTALL"} = "$ENV{CRUNCH_TMP}/opt";
$ENV{"CRUNCH_WORK"} = $ENV{"JOB_WORK"}; # deprecated
my $arv = Arvados->new('apiVersion' => 'v1');
-my $metastream;
+my $local_logfile;
my $User = $arv->{'users'}->{'current'}->execute;
{
$Job = $arv->{'jobs'}->{'get'}->execute('uuid' => $jobspec);
if (!$force_unlock) {
+ # If some other crunch-job process has grabbed this job (or we see
+ # other evidence that the job is already underway) we exit
+ # EX_TEMPFAIL so crunch-dispatch (our parent process) doesn't
+ # mark the job as failed.
if ($Job->{'is_locked_by_uuid'}) {
- croak("Job is locked: " . $Job->{'is_locked_by_uuid'});
+ Log(undef, "Job is locked by " . $Job->{'is_locked_by_uuid'});
+ exit EX_TEMPFAIL;
}
if ($Job->{'success'} ne undef) {
- croak("Job 'success' flag (" . $Job->{'success'} . ") is not null");
+ Log(undef, "Job 'success' flag (" . $Job->{'success'} . ") is not null");
+ exit EX_TEMPFAIL;
}
if ($Job->{'running'}) {
- croak("Job 'running' flag is already set");
+ Log(undef, "Job 'running' flag is already set");
+ exit EX_TEMPFAIL;
}
if ($Job->{'started_at'}) {
- croak("Job 'started_at' time is already set (" . $Job->{'started_at'} . ")");
+ Log(undef, "Job 'started_at' time is already set (" . $Job->{'started_at'} . ")");
+ exit EX_TEMPFAIL;
}
}
}
$job_id = $Job->{'uuid'};
my $keep_logfile = $job_id . '.log.txt';
-my $local_logfile = File::Temp->new();
+$local_logfile = File::Temp->new();
$Job->{'runtime_constraints'} ||= {};
$Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0;
}
if (exists $ENV{SLURM_NODELIST})
{
- push @sinfo, `sinfo -h --format='%c %N' --nodes='$ENV{SLURM_NODELIST}'`;
+ push @sinfo, `sinfo -h --format='%c %N' --nodes=\Q$ENV{SLURM_NODELIST}\E`;
}
foreach (@sinfo)
{
# Claim this job, and make sure nobody else does
unless ($Job->update_attributes('is_locked_by_uuid' => $User->{'uuid'}) &&
$Job->{'is_locked_by_uuid'} == $User->{'uuid'}) {
- croak("Error while updating / locking job");
+ Log(undef, "Error while updating / locking job, exiting ".EX_TEMPFAIL);
+ exit EX_TEMPFAIL;
}
$Job->update_attributes('started_at' => scalar gmtime,
'running' => 1,
if ($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*']);
+ ['bash', '-c', 'if mount | grep -q $JOB_WORK/; then for i in $JOB_WORK/*keep; do /bin/fusermount -z -u $i; done; fi; sleep 1; rm -rf $JOB_WORK $CRUNCH_TMP/opt $CRUNCH_TMP/src*']);
exit (1);
}
while (1)
my $commit;
my $git_archive;
my $treeish = $Job->{'script_version'};
- my $repo = $git_dir || $ENV{'CRUNCH_DEFAULT_GIT_DIR'};
- # Todo: let script_version specify repository instead of expecting
- # parent process to figure it out.
- $ENV{"CRUNCH_SRC_URL"} = $repo;
- # Create/update our clone of the remote git repo
+ # If we're running under crunch-dispatch, it will have pulled the
+ # appropriate source tree into its own repository, and given us that
+ # repo's path as $git_dir. If we're running a "local" job, and a
+ # script_version was specified, it's up to the user to provide the
+ # full path to a local repository in Job->{repository}.
+ #
+ # TODO: Accept URLs too, not just local paths. Use git-ls-remote and
+ # git-archive --remote where appropriate.
+ #
+ # TODO: Accept a locally-hosted Arvados repository by name or
+ # UUID. Use arvados.v1.repositories.list or .get to figure out the
+ # appropriate fetch-url.
+ my $repo = $git_dir || $ENV{'CRUNCH_DEFAULT_GIT_DIR'} || $Job->{'repository'};
+
+ $ENV{"CRUNCH_SRC_URL"} = $repo;
- if (!-d $ENV{"CRUNCH_SRC"}) {
- system(qw(git clone), $repo, $ENV{"CRUNCH_SRC"}) == 0
- or croak ("git clone $repo failed: exit ".($?>>8));
- system("cd $ENV{CRUNCH_SRC} && git config clean.requireForce false");
+ if (-d "$repo/.git") {
+ # We were given a working directory, but we are only interested in
+ # the index.
+ $repo = "$repo/.git";
}
- `cd $ENV{CRUNCH_SRC} && git remote set-url origin \"\$CRUNCH_SRC_URL\" && git fetch -q --tags origin`;
# If this looks like a subversion r#, look for it in git-svn commit messages
if ($treeish =~ m{^\d{1,4}$}) {
- my $gitlog = `cd $ENV{CRUNCH_SRC} && git log --pretty="format:%H" --grep="git-svn-id:.*\@$treeish " origin/master`;
+ 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}$/) {
$commit = $gitlog;
# If that didn't work, try asking git to look it up as a tree-ish.
if (!defined $commit) {
-
- my $cooked_treeish = $treeish;
- if ($treeish !~ m{^[0-9a-f]{5,}$}) {
- # Looks like a git branch name -- make sure git knows it's
- # relative to the remote repo
- $cooked_treeish = "origin/$treeish";
- }
-
- my $found = `cd $ENV{CRUNCH_SRC} && git rev-list -1 $cooked_treeish`;
+ my $found = `git --git-dir=\Q$repo\E rev-list -1 ''\Q$treeish\E`;
chomp $found;
if ($found =~ /^[0-9a-f]{40}$/s) {
$commit = $found;
$ENV{"CRUNCH_SRC_COMMIT"} = $commit;
@execargs = ("sh", "-c",
"mkdir -p $ENV{CRUNCH_INSTALL} && cd $ENV{CRUNCH_TMP} && perl -");
- $git_archive = `cd $ENV{CRUNCH_SRC} && git archive $commit`;
+ $git_archive = `git --git-dir=\Q$repo\E archive ''\Q$commit\E`;
}
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)
{
must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
}
-
+# If this job requires a Docker image, install that.
+my $docker_bin = "/usr/bin/docker.io";
+my ($docker_locator, $docker_stream, $docker_hash);
+if ($docker_locator = $Job->{docker_image_locator}) {
+ ($docker_stream, $docker_hash) = find_docker_image($docker_locator);
+ if (!$docker_hash)
+ {
+ croak("No Docker image hash found from locator $docker_locator");
+ }
+ $docker_stream =~ s/^\.//;
+ my $docker_install_script = qq{
+if ! $docker_bin images -q --no-trunc | grep -qxF \Q$docker_hash\E; then
+ arv-get \Q$docker_locator$docker_stream/$docker_hash.tar\E | $docker_bin load
+fi
+};
+ my $docker_pid = fork();
+ if ($docker_pid == 0)
+ {
+ srun (["srun", "--nodelist=" . join(',', @node)],
+ ["/bin/sh", "-ec", $docker_install_script]);
+ exit ($?);
+ }
+ while (1)
+ {
+ last if $docker_pid == waitpid (-1, WNOHANG);
+ freeze_if_want_freeze ($docker_pid);
+ select (undef, undef, undef, 0.1);
+ }
+ if ($? != 0)
+ {
+ croak("Installing Docker image from $docker_locator returned exit code $?");
+ }
+}
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"} = $ENV{"JOB_WORK"}."/$id.$$";
+ $ENV{"HOME"} = $ENV{"TASK_WORK"};
$ENV{"TASK_KEEPMOUNT"} = $ENV{"TASK_WORK"}.".keep";
$ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
$ENV{"CRUNCH_NODE_SLOTS"} = $slot[$childslot]->{node}->{ncpus};
qw(-n1 -c1 -N1 -D), $ENV{'TMPDIR'},
"--job-name=$job_id.$id.$$",
);
- my @execargs = qw(sh);
my $build_script_to_send = "";
my $command =
"if [ -e $ENV{TASK_WORK} ]; then rm -rf $ENV{TASK_WORK}; fi; "
- ."mkdir -p $ENV{JOB_WORK} $ENV{CRUNCH_TMP} $ENV{TASK_WORK} $ENV{TASK_KEEPMOUNT} "
+ ."mkdir -p $ENV{CRUNCH_TMP} $ENV{JOB_WORK} $ENV{TASK_WORK} $ENV{TASK_KEEPMOUNT} "
."&& cd $ENV{CRUNCH_TMP} ";
if ($build_script)
{
$build_script_to_send = $build_script;
+ }
+ $command .= "&& exec arv-mount --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 ";
+ $command .= "$docker_bin run --rm=true --attach=stdout --attach=stderr --user=crunch --cidfile=$ENV{TASK_WORK}/docker.cid ";
+ # Dynamically configure the container to use the host system as its
+ # DNS server. Get the host's global addresses from the ip command,
+ # and turn them into docker --dns options using gawk.
$command .=
- "&& perl -";
+ q{$(ip -o address show scope global |
+ gawk 'match($4, /^([0-9\.:]+)\//, x){print "--dns", x[1]}') };
+ $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT}:/keep:ro\E ";
+ $command .= "--env=\QHOME=/home/crunch\E ";
+ while (my ($env_key, $env_val) = each %ENV)
+ {
+ if ($env_key =~ /^(ARVADOS|JOB|TASK)_/) {
+ if ($env_key eq "TASK_KEEPMOUNT") {
+ $command .= "--env=\QTASK_KEEPMOUNT=/keep\E ";
+ }
+ else {
+ $command .= "--env=\Q$env_key=$env_val\E ";
+ }
+ }
+ }
+ $command .= "--env=\QCRUNCH_NODE_SLOTS=$ENV{CRUNCH_NODE_SLOTS}\E ";
+ $command .= "--env=\QCRUNCH_SRC=/tmp/crunch-src\E ";
+ $command .= "\Q$docker_hash\E ";
+ $command .= "stdbuf --output=0 --error=0 ";
+ $command .= "perl - ";
+ $command .= "/tmp/crunch-src/crunch_scripts/" . $Job->{"script"};
+ } else {
+ # Non-docker run
+ $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -poll=10000 ";
+ $command .= "stdbuf --output=0 --error=0 ";
+ $command .= "perl - ";
+ $command .= "$ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
}
- $command .=
- "&& exec arv-mount $ENV{TASK_KEEPMOUNT} --exec $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
+
my @execargs = ('bash', '-c', $command);
srun (\@srunargs, \@execargs, undef, $build_script_to_send);
- exit (111);
+ # exec() failed, we assume nothing happened.
+ Log(undef, "srun() failed on build script");
+ die;
}
close("writer");
if (!defined $childpid)
release_allocation();
freeze();
+my $collated_output = &collate_output();
+
if ($job_has_uuid) {
- $Job->update_attributes('output' => &collate_output(),
- 'running' => 0,
- 'success' => $Job->{'output'} && $main::success,
+ $Job->update_attributes('running' => 0,
+ 'success' => $collated_output && $main::success,
'finished_at' => scalar gmtime)
}
-if ($Job->{'output'})
-{
+if (!$collated_output) {
+ Log(undef, "output undef");
+}
+else {
eval {
- my $manifest_text = `arv keep get \Q$Job->{'output'}\E`;
- $arv->{'collections'}->{'create'}->execute('collection' => {
- 'uuid' => $Job->{'output'},
- 'manifest_text' => $manifest_text,
- });
- if ($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',
- });
+ open(my $orig_manifest, '-|', 'arv-get', $collated_output)
+ or die "failed to get collated manifest: $!";
+ my $orig_manifest_text = '';
+ while (my $manifest_line = <$orig_manifest>) {
+ $orig_manifest_text .= $manifest_line;
}
+ my $output = $arv->{'collections'}->{'create'}->execute('collection' => {
+ 'manifest_text' => $orig_manifest_text,
+ });
+ Log(undef, "output " . $output->{portable_data_hash});
+ $Job->update_attributes('output' => $output->{portable_data_hash}) if $job_has_uuid;
};
if ($@) {
Log (undef, "Failed to register output manifest: $@");
Log (undef, "finish");
save_meta();
-exit 0;
+exit ($Job->{'success'} ? 1 : 0);
push @freeslot, $proc{$pid}->{slot};
delete $proc{$pid};
- # Load new tasks
- my $newtask_list = $arv->{'job_tasks'}->{'list'}->execute(
- 'where' => {
- 'created_by_job_task_uuid' => $Jobstep->{'arvados_task'}->{uuid}
- },
- 'order' => 'qsequence'
- );
- foreach my $arvados_task (@{$newtask_list->{'items'}}) {
- my $jobstep = {
- 'level' => $arvados_task->{'sequence'},
- 'failures' => 0,
- 'arvados_task' => $arvados_task
- };
- push @jobstep, $jobstep;
- push @jobstep_todo, $#jobstep;
+ if ($task_success) {
+ # Load new tasks
+ my $newtask_list = [];
+ my $newtask_results;
+ do {
+ $newtask_results = $arv->{'job_tasks'}->{'list'}->execute(
+ 'where' => {
+ 'created_by_job_task_uuid' => $Jobstep->{'arvados_task'}->{uuid}
+ },
+ 'order' => 'qsequence',
+ 'offset' => scalar(@$newtask_list),
+ );
+ push(@$newtask_list, @{$newtask_results->{items}});
+ } while (@{$newtask_results->{items}});
+ foreach my $arvados_task (@$newtask_list) {
+ my $jobstep = {
+ 'level' => $arvados_task->{'sequence'},
+ 'failures' => 0,
+ 'arvados_task' => $arvados_task
+ };
+ push @jobstep, $jobstep;
+ push @jobstep_todo, $#jobstep;
+ }
}
$progress_is_dirty = 1;
my $hash = shift;
my ($keep, $child_out, $output_block);
- my $cmd = "arv keep get \Q$hash\E";
+ my $cmd = "arv-get \Q$hash\E";
open($keep, '-|', $cmd) or die "fetch_block: $cmd: $!";
- sysread($keep, $output_block, 64 * 1024 * 1024);
+ $output_block = '';
+ while (1) {
+ my $buf;
+ my $bytes = sysread($keep, $buf, 1024 * 1024);
+ if (!defined $bytes) {
+ die "reading from arv-get: $!";
+ } elsif ($bytes == 0) {
+ # sysread returns 0 at the end of the pipe.
+ last;
+ } else {
+ # some bytes were read into buf.
+ $output_block .= $buf;
+ }
+ }
close $keep;
return $output_block;
}
Log (undef, "collate");
my ($child_out, $child_in);
- my $pid = open2($child_out, $child_in, 'arv', 'keep', 'put', '--raw');
+ my $pid = open2($child_out, $child_in, 'arv-put', '--raw');
my $joboutput;
for (@jobstep)
{
- next if (!exists $_->{'arvados_task'}->{output} ||
- !$_->{'arvados_task'}->{'success'} ||
- $_->{'exitcode'} != 0);
+ next if (!exists $_->{'arvados_task'}->{'output'} ||
+ !$_->{'arvados_task'}->{'success'});
my $output = $_->{'arvados_task'}->{output};
if ($output !~ /^[0-9a-f]{32}(\+\S+)*$/)
{
sysread($child_out, $joboutput, 64 * 1024 * 1024);
chomp($joboutput);
} else {
- Log (undef, "timed out reading from 'arv keep put'");
+ Log (undef, "timed out reading from 'arv-put'");
}
}
waitpid($pid, 0);
- if ($joboutput)
- {
- Log (undef, "output $joboutput");
- $Job->update_attributes('output' => $joboutput) if $job_has_uuid;
- }
- else
- {
- Log (undef, "output undef");
- }
return $joboutput;
}
$message =~ s{([^ -\176])}{"\\" . sprintf ("%03o", ord($1))}ge;
$message .= "\n";
my $datetime;
- if ($metastream || -t STDERR) {
+ if ($local_logfile || -t STDERR) {
my @gmtime = gmtime;
$datetime = sprintf ("%04d-%02d-%02d_%02d:%02d:%02d",
$gmtime[5]+1900, $gmtime[4]+1, @gmtime[3,2,1,0]);
}
print STDERR ((-t STDERR) ? ($datetime." ".$message) : $message);
- if ($metastream) {
- print $metastream $datetime . " " . $message;
+ if ($local_logfile) {
+ print $local_logfile $datetime . " " . $message;
}
}
freeze() if @jobstep_todo;
collate_output() if @jobstep_todo;
cleanup();
- save_meta() if $metastream;
+ save_meta() if $local_logfile;
die;
}
return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm
$local_logfile->flush;
- my $cmd = "arv keep put --filename \Q$keep_logfile\E "
+ my $cmd = "arv-put --filename ''\Q$keep_logfile\E "
. quotemeta($local_logfile->filename);
my $loglocator = `$cmd`;
die "system $cmd failed: $?" if $?;
+ chomp($loglocator);
$local_logfile = undef; # the temp file is automatically deleted
Log (undef, "log manifest is $loglocator");
}
}
+sub find_docker_image {
+ # Given a Keep locator, check to see if it contains a Docker image.
+ # If so, return its stream name and Docker hash.
+ # If not, return undef for both values.
+ my $locator = shift;
+ 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);
+ }
+ }
+ return (undef, undef);
+}
+
__DATA__
#!/usr/bin/perl
# checkout-and-build
use Fcntl ':flock';
+use File::Path qw( make_path );
my $destdir = $ENV{"CRUNCH_SRC"};
my $commit = $ENV{"CRUNCH_SRC_COMMIT"};
my $repo = $ENV{"CRUNCH_SRC_URL"};
+my $task_work = $ENV{"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
{
projects / arvados.git / blobdiff