#!/usr/bin/perl
-# -*- mode: perl; perl-indent-level: 2; -*-
+# -*- mode: perl; perl-indent-level: 2; indent-tabs-mode: nil; -*-
=head1 NAME
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 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";
$ENV{"CRUNCH_WORK"} = $ENV{"JOB_WORK"}; # deprecated
mkdir ($ENV{"JOB_WORK"});
+my $arv_cli;
+
+if (defined $ENV{"ARV_CLI"}) {
+ $arv_cli = $ENV{"ARV_CLI"};
+}
+else {
+ $arv_cli = 'arv';
+}
+
my $force_unlock;
my $git_dir;
my $jobspec;
my $arv = Arvados->new('apiVersion' => 'v1');
-my $metastream;
+my $local_logfile;
my $User = $arv->{'users'}->{'current'}->execute;
}
$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 = $job_id . '.log.txt';
+$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)
{
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 $ENV{CRUNCH_SRC} && git remote set-url origin \"\$CRUNCH_SRC_URL\" && git fetch -q origin`;
# If this looks like a subversion r#, look for it in git-svn commit messages
if ($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 $treeish");
must_lock_now("$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_image = $Job->{runtime_constraints}->{docker_image} || "";
+if ($docker_image) {
+ my $docker_pid = fork();
+ if ($docker_pid == 0)
+ {
+ srun (["srun", "--nodelist=" . join(' ', @node)],
+ [$docker_bin, 'pull', $docker_image]);
+ exit ($?);
+ }
+ while (1)
+ {
+ last if $docker_pid == waitpid (-1, WNOHANG);
+ freeze_if_want_freeze ($docker_pid);
+ select (undef, undef, undef, 0.1);
+ }
+ # If the Docker image was specified as a hash, pull will fail.
+ # Ignore that error. We'll see what happens when we try to run later.
+ if (($? != 0) && ($docker_image !~ /^[0-9a-fA-F]{5,64}$/))
+ {
+ croak("Installing Docker image $docker_image 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.$$";
- $ENV{"TASK_KEEPMOUNT"} = $ENV{"TASK_WORK"}."/keep";
+ $ENV{"TASK_KEEPMOUNT"} = $ENV{"TASK_WORK"}.".keep";
$ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
$ENV{"CRUNCH_NODE_SLOTS"} = $slot[$childslot]->{node}->{ncpus};
$ENV{"PATH"} = $ENV{"CRUNCH_INSTALL"} . "/bin:" . $ENV{"PATH"};
qw(-n1 -c1 -N1 -D), $ENV{'TMPDIR'},
"--job-name=$job_id.$id.$$",
);
- my @execargs = qw(sh);
my $build_script_to_send = "";
my $command =
"if [ -e $ENV{TASK_WORK} ]; then rm -rf $ENV{TASK_WORK}; fi; "
$command .=
"&& perl -";
}
- $command .=
- "&& exec arv-mount $ENV{TASK_KEEPMOUNT} --exec $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
+ $command .= "&& exec arv-mount --allow-other $ENV{TASK_KEEPMOUNT} --exec ";
+ if ($docker_image)
+ {
+ $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -cgroup-parent=docker -cgroup-cid=$ENV{TASK_WORK}/docker.cid -poll=10000 ";
+ $command .= "$docker_bin run -i -a stdin -a stdout -a stderr --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.
+ $command .=
+ q{$(ip -o address show scope global |
+ gawk 'match($4, /^([0-9\.:]+)\//, x){print "--dns", x[1]}') };
+ foreach my $env_key (qw(CRUNCH_SRC CRUNCH_TMP TASK_KEEPMOUNT))
+ {
+ $command .= "-v \Q$ENV{$env_key}:$ENV{$env_key}:rw\E ";
+ }
+ while (my ($env_key, $env_val) = each %ENV)
+ {
+ if ($env_key =~ /^(JOB|TASK)_/) {
+ $command .= "-e \Q$env_key=$env_val\E ";
+ }
+ }
+ $command .= "\Q$docker_image\E ";
+ } else {
+ $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -poll=10000 "
+ }
+ $command .= "stdbuf -o0 -e0 ";
+ $command .= "$ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
my @execargs = ('bash', '-c', $command);
srun (\@srunargs, \@execargs, undef, $build_script_to_send);
exit (111);
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)
{
eval {
- my $manifest_text = capturex("whget", $Job->{'output'});
- $arv->{'collections'}->{'create'}->execute('collection' => {
- 'uuid' => $Job->{'output'},
+ open(my $orig_manifest, '-|', 'arv', 'keep', '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 @manifest_lines = ();
+ while (my $manifest_line = <$orig_manifest>) {
+ 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(@manifest_lines, join(" ", @words));
+ }
+ my $manifest_text = join("", @manifest_lines);
+ my $output = $arv->{'collections'}->{'create'}->execute('collection' => {
+ 'uuid' => md5_hex($manifest_text),
'manifest_text' => $manifest_text,
});
+ $Job->update_attributes('output' => $output->{uuid});
+ if ($Job->{'output_is_persistent'}) {
+ $arv->{'links'}->{'create'}->execute('link' => {
+ 'tail_kind' => 'arvados#user',
+ 'tail_uuid' => $User->{'uuid'},
+ 'head_kind' => 'arvados#collection',
+ 'head_uuid' => $Job->{'output'},
+ 'link_class' => 'resources',
+ 'name' => 'wants',
+ });
+ }
};
if ($@) {
Log (undef, "Failed to register output manifest: $@");
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 $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,
} split ("\n", $jobstep[$job]->{stderr});
}
+sub fetch_block
+{
+ my $hash = shift;
+ my ($keep, $child_out, $output_block);
+
+ my $cmd = "$arv_cli 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_cli, '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, $joboutput, 64 * 1024 * 1024);
+ chomp($joboutput);
+ } else {
+ Log (undef, "timed out reading from 'arv keep put'");
+ }
+ }
+ waitpid($pid, 0);
+
if ($joboutput)
{
Log (undef, "output $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) : $message);
- return if !$metastream;
- $metastream->write_data ($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;
}
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_cli keep 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");
$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;
}
projects / arvados.git / blobdiff