2 # -*- mode: perl; perl-indent-level: 2; indent-tabs-mode: nil; -*-
6 crunch-job: Execute job steps, save snapshots as requested, collate output.
10 Obtain job details from Arvados, run tasks on compute nodes (typically
11 invoked by scheduler on controller):
13 crunch-job --job x-y-z --git-dir /path/to/repo/.git
15 Obtain job details from command line, run tasks on local machine
16 (typically invoked by application or developer on VM):
18 crunch-job --job '{"script_version":"/path/to/working/tree","script":"scriptname",...}'
20 crunch-job --job '{"repository":"https://github.com/curoverse/arvados.git","script_version":"master","script":"scriptname",...}'
28 If the job is already locked, steal the lock and run it anyway.
32 Path to a .git directory (or a git URL) where the commit given in the
33 job's C<script_version> attribute is to be found. If this is I<not>
34 given, the job's C<repository> attribute will be used.
38 Arvados API authorization token to use during the course of the job.
42 Do not clear per-job/task temporary directories during initial job
43 setup. This can speed up development and debugging when running jobs
48 UUID of the job to run, or a JSON-encoded job resource without a
49 UUID. If the latter is given, a new job object will be created.
53 =head1 RUNNING JOBS LOCALLY
55 crunch-job's log messages appear on stderr along with the job tasks'
56 stderr streams. The log is saved in Keep at each checkpoint and when
59 If the job succeeds, the job's output locator is printed on stdout.
61 While the job is running, the following signals are accepted:
65 =item control-C, SIGINT, SIGQUIT
67 Save a checkpoint, terminate any job tasks that are running, and stop.
71 Save a checkpoint and continue.
75 Refresh node allocation (i.e., check whether any nodes have been added
76 or unallocated) and attributes of the Job record that should affect
77 behavior (e.g., cancel job if cancelled_at becomes non-nil).
85 use POSIX ':sys_wait_h';
86 use POSIX qw(strftime);
87 use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK);
91 use Digest::MD5 qw(md5_hex);
97 use File::Path qw( make_path remove_tree );
99 use constant TASK_TEMPFAIL => 111;
100 use constant EX_TEMPFAIL => 75;
101 use constant EX_RETRY_UNLOCKED => 93;
103 $ENV{"TMPDIR"} ||= "/tmp";
104 unless (defined $ENV{"CRUNCH_TMP"}) {
105 $ENV{"CRUNCH_TMP"} = $ENV{"TMPDIR"} . "/crunch-job";
106 if ($ENV{"USER"} ne "crunch" && $< != 0) {
107 # use a tmp dir unique for my uid
108 $ENV{"CRUNCH_TMP"} .= "-$<";
112 # Create the tmp directory if it does not exist
113 if ( ! -d $ENV{"CRUNCH_TMP"} ) {
114 make_path $ENV{"CRUNCH_TMP"} or die "Failed to create temporary working directory: " . $ENV{"CRUNCH_TMP"};
117 $ENV{"JOB_WORK"} = $ENV{"CRUNCH_TMP"} . "/work";
118 $ENV{"CRUNCH_INSTALL"} = "$ENV{CRUNCH_TMP}/opt";
119 $ENV{"CRUNCH_WORK"} = $ENV{"JOB_WORK"}; # deprecated
120 mkdir ($ENV{"JOB_WORK"});
129 my $cgroup_root = "/sys/fs/cgroup";
130 my $docker_bin = "docker.io";
131 my $docker_run_args = "";
132 GetOptions('force-unlock' => \$force_unlock,
133 'git-dir=s' => \$git_dir,
134 'job=s' => \$jobspec,
135 'job-api-token=s' => \$job_api_token,
136 'no-clear-tmp' => \$no_clear_tmp,
137 'resume-stash=s' => \$resume_stash,
138 'cgroup-root=s' => \$cgroup_root,
139 'docker-bin=s' => \$docker_bin,
140 'docker-run-args=s' => \$docker_run_args,
143 if (defined $job_api_token) {
144 $ENV{ARVADOS_API_TOKEN} = $job_api_token;
147 my $have_slurm = exists $ENV{SLURM_JOB_ID} && exists $ENV{SLURM_NODELIST};
152 $main::ENV{CRUNCH_DEBUG} = 1;
156 $main::ENV{CRUNCH_DEBUG} = 0;
159 my $arv = Arvados->new('apiVersion' => 'v1');
168 if ($jobspec =~ /^[-a-z\d]+$/)
170 # $jobspec is an Arvados UUID, not a JSON job specification
171 $Job = api_call("jobs/get", uuid => $jobspec);
176 $local_job = JSON::decode_json($jobspec);
180 # Make sure our workers (our slurm nodes, localhost, or whatever) are
181 # at least able to run basic commands: they aren't down or severely
184 if (($Job || $local_job)->{docker_image_locator}) {
185 $cmd = [$docker_bin, 'ps', '-q'];
187 Log(undef, "Sanity check is `@$cmd`");
188 my ($exited, $stdout, $stderr) = srun_sync(
189 ["srun", "--nodes=\Q$ENV{SLURM_NNODES}\E", "--ntasks-per-node=1"],
191 {label => "sanity check"});
193 Log(undef, "Sanity check failed: ".exit_status_s($exited));
196 Log(undef, "Sanity check OK");
199 my $User = api_call("users/current");
202 if (!$force_unlock) {
203 # Claim this job, and make sure nobody else does
204 eval { api_call("jobs/lock", uuid => $Job->{uuid}); };
206 Log(undef, "Error while locking job, exiting ".EX_TEMPFAIL);
215 map { croak ("No $_ specified") unless $local_job->{$_} }
216 qw(script script_version script_parameters);
219 $local_job->{'is_locked_by_uuid'} = $User->{'uuid'};
220 $local_job->{'started_at'} = gmtime;
221 $local_job->{'state'} = 'Running';
223 $Job = api_call("jobs/create", job => $local_job);
225 $job_id = $Job->{'uuid'};
227 my $keep_logfile = $job_id . '.log.txt';
228 log_writer_start($keep_logfile);
230 $Job->{'runtime_constraints'} ||= {};
231 $Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0;
232 my $max_ncpus = $Job->{'runtime_constraints'}->{'max_tasks_per_node'};
234 my $gem_versions = `gem list --quiet arvados-cli 2>/dev/null`;
236 $gem_versions =~ s/^arvados-cli \(/ with arvados-cli Gem version(s) /;
237 chomp($gem_versions);
238 chop($gem_versions); # Closing parentheses
243 "running from " . ((-e $0) ? realpath($0) : "stdin") . $gem_versions);
245 Log (undef, "check slurm allocation");
248 # Should use $ENV{SLURM_TASKS_PER_NODE} instead of sinfo? (eg. "4(x3),2,4(x2)")
252 my $localcpus = 0 + `grep -cw ^processor /proc/cpuinfo` || 1;
253 push @sinfo, "$localcpus localhost";
255 if (exists $ENV{SLURM_NODELIST})
257 push @sinfo, `sinfo -h --format='%c %N' --nodes=\Q$ENV{SLURM_NODELIST}\E`;
261 my ($ncpus, $slurm_nodelist) = split;
262 $ncpus = $max_ncpus if $max_ncpus && $ncpus > $max_ncpus;
265 while ($slurm_nodelist =~ s/^([^\[,]+?(\[.*?\])?)(,|$)//)
268 if ($nodelist =~ /\[((\d+)(-(\d+))?(,(\d+)(-(\d+))?)*)\]/)
271 foreach (split (",", $ranges))
284 push @nodelist, map {
286 $n =~ s/\[[-,\d]+\]/$_/;
293 push @nodelist, $nodelist;
296 foreach my $nodename (@nodelist)
298 Log (undef, "node $nodename - $ncpus slots");
299 my $node = { name => $nodename,
301 # The number of consecutive times a task has been dispatched
302 # to this node and failed.
304 # The number of consecutive times that SLURM has reported
305 # a node failure since the last successful task.
307 # Don't dispatch work to this node until this time
308 # (in seconds since the epoch) has passed.
310 foreach my $cpu (1..$ncpus)
312 push @slot, { node => $node,
316 push @node, @nodelist;
321 # Ensure that we get one jobstep running on each allocated node before
322 # we start overloading nodes with concurrent steps
324 @slot = sort { $a->{cpu} <=> $b->{cpu} } @slot;
327 $Job->update_attributes(
328 'tasks_summary' => { 'failed' => 0,
333 Log (undef, "start");
334 $SIG{'INT'} = sub { $main::please_freeze = 1; };
335 $SIG{'QUIT'} = sub { $main::please_freeze = 1; };
336 $SIG{'TERM'} = \&croak;
337 $SIG{'TSTP'} = sub { $main::please_freeze = 1; };
338 $SIG{'ALRM'} = sub { $main::please_info = 1; };
339 $SIG{'CONT'} = sub { $main::please_continue = 1; };
340 $SIG{'HUP'} = sub { $main::please_refresh = 1; };
342 $main::please_freeze = 0;
343 $main::please_info = 0;
344 $main::please_continue = 0;
345 $main::please_refresh = 0;
346 my $jobsteps_must_output_keys = 0; # becomes 1 when any task outputs a key
348 grep { $ENV{$1} = $2 if /^(NOCACHE.*?)=(.*)/ } split ("\n", $$Job{knobs});
349 $ENV{"CRUNCH_JOB_UUID"} = $job_id;
350 $ENV{"JOB_UUID"} = $job_id;
353 my @jobstep_todo = ();
354 my @jobstep_done = ();
355 my @jobstep_tomerge = ();
356 my $jobstep_tomerge_level = 0;
357 my $squeue_checked = 0;
358 my $sinfo_checked = 0;
359 my $latest_refresh = scalar time;
363 if (defined $Job->{thawedfromkey})
365 thaw ($Job->{thawedfromkey});
369 my $first_task = api_call("job_tasks/create", job_task => {
370 'job_uuid' => $Job->{'uuid'},
375 push @jobstep, { 'level' => 0,
377 'arvados_task' => $first_task,
379 push @jobstep_todo, 0;
385 must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
388 my $build_script = handle_readall(\*DATA);
389 my $nodelist = join(",", @node);
390 my $git_tar_count = 0;
392 if (!defined $no_clear_tmp) {
393 # Find FUSE mounts under $CRUNCH_TMP and unmount them. Then clean
394 # up work directories crunch_tmp/work, crunch_tmp/opt,
397 # TODO: When #5036 is done and widely deployed, we can limit mount's
398 # -t option to simply fuse.keep.
399 my ($exited, $stdout, $stderr) = srun_sync(
400 ["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
401 ['bash', '-ec', '-o', 'pipefail', 'mount -t fuse,fuse.keep | awk "(index(\$3, \"$CRUNCH_TMP\") == 1){print \$3}" | xargs -r -n 1 fusermount -u -z; sleep 1; rm -rf $JOB_WORK $CRUNCH_INSTALL $CRUNCH_TMP/task $CRUNCH_TMP/src* $CRUNCH_TMP/*.cid'],
402 {label => "clean work dirs"});
404 exit(EX_RETRY_UNLOCKED);
408 # If this job requires a Docker image, install that.
409 my ($docker_locator, $docker_stream, $docker_hash, $docker_limitmem, $dockeruserarg);
410 if ($docker_locator = $Job->{docker_image_locator}) {
411 Log (undef, "Install docker image $docker_locator");
412 ($docker_stream, $docker_hash) = find_docker_image($docker_locator);
415 croak("No Docker image hash found from locator $docker_locator");
417 Log (undef, "docker image hash is $docker_hash");
418 $docker_stream =~ s/^\.//;
419 my $docker_install_script = qq{
421 [[ \$($docker_bin inspect --format="{{.ID}}" \Q$docker_hash\E) = \Q$docker_hash\E ]]
423 if loaded 2>/dev/null; then
424 echo >&2 "image is already present"
427 echo >&2 "docker image is not present; loading"
428 arv-get \Q$docker_locator$docker_stream/$docker_hash.tar\E | $docker_bin load
430 echo >&2 "`docker load` exited 0, but image is not found (!)"
433 echo >&2 "image loaded successfully"
436 my ($exited, $stdout, $stderr) = srun_sync(
437 ["srun", "--nodelist=" . join(',', @node)],
438 ["/bin/bash", "-o", "pipefail", "-ec", $docker_install_script],
439 {label => "load docker image"});
442 exit(EX_RETRY_UNLOCKED);
445 # Determine whether this version of Docker supports memory+swap limits.
446 ($exited, $stdout, $stderr) = srun_sync(
447 ["srun", "--nodes=1"],
448 [$docker_bin, 'run', '--help'],
449 {label => "check --memory-swap feature"});
450 $docker_limitmem = ($stdout =~ /--memory-swap/);
452 # Find a non-root Docker user to use.
453 # Tries the default user for the container, then 'crunch', then 'nobody',
454 # testing for whether the actual user id is non-zero. This defends against
455 # mistakes but not malice, but we intend to harden the security in the future
456 # so we don't want anyone getting used to their jobs running as root in their
458 my @tryusers = ("", "crunch", "nobody");
459 foreach my $try_user (@tryusers) {
462 if ($try_user eq "") {
463 $label = "check whether default user is UID 0";
466 $label = "check whether user '$try_user' is UID 0";
467 $try_user_arg = "--user=$try_user";
469 my ($exited, $stdout, $stderr) = srun_sync(
470 ["srun", "--nodes=1"],
472 "$docker_bin run $docker_run_args $try_user_arg $docker_hash id --user"],
475 if ($exited == 0 && $stdout =~ /^\d+$/ && $stdout > 0) {
476 $dockeruserarg = $try_user_arg;
477 if ($try_user eq "") {
478 Log(undef, "Container will run with default user");
480 Log(undef, "Container will run with $dockeruserarg");
486 if (!defined $dockeruserarg) {
487 croak("Could not find a user in container that is not UID 0 (tried default user, @tryusers) or there was a problem running 'id' in the container.");
490 if ($Job->{arvados_sdk_version}) {
491 # The job also specifies an Arvados SDK version. Add the SDKs to the
492 # tar file for the build script to install.
493 Log(undef, sprintf("Packing Arvados SDK version %s for installation",
494 $Job->{arvados_sdk_version}));
495 add_git_archive("git", "--git-dir=$git_dir", "archive",
496 "--prefix=.arvados.sdk/",
497 $Job->{arvados_sdk_version}, "sdk");
501 if (!defined $git_dir && $Job->{'script_version'} =~ m{^/}) {
502 # If script_version looks like an absolute path, *and* the --git-dir
503 # argument was not given -- which implies we were not invoked by
504 # crunch-dispatch -- we will use the given path as a working
505 # directory instead of resolving script_version to a git commit (or
506 # doing anything else with git).
507 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{'script_version'};
508 $ENV{"CRUNCH_SRC"} = $Job->{'script_version'};
511 # Resolve the given script_version to a git commit sha1. Also, if
512 # the repository is remote, clone it into our local filesystem: this
513 # ensures "git archive" will work, and is necessary to reliably
514 # resolve a symbolic script_version like "master^".
515 $ENV{"CRUNCH_SRC"} = "$ENV{CRUNCH_TMP}/src";
517 Log (undef, "Looking for version ".$Job->{script_version}." from repository ".$Job->{repository});
519 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{script_version};
521 # If we're running under crunch-dispatch, it will have already
522 # pulled the appropriate source tree into its own repository, and
523 # given us that repo's path as $git_dir.
525 # If we're running a "local" job, we might have to fetch content
526 # from a remote repository.
528 # (Currently crunch-dispatch gives a local path with --git-dir, but
529 # we might as well accept URLs there too in case it changes its
531 my $repo = $git_dir || $Job->{'repository'};
533 # Repository can be remote or local. If remote, we'll need to fetch it
534 # to a local dir before doing `git log` et al.
537 if ($repo =~ m{://|^[^/]*:}) {
538 # $repo is a git url we can clone, like git:// or https:// or
539 # file:/// or [user@]host:repo.git. Note "user/name@host:foo" is
540 # not recognized here because distinguishing that from a local
541 # path is too fragile. If you really need something strange here,
542 # use the ssh:// form.
543 $repo_location = 'remote';
544 } elsif ($repo =~ m{^\.*/}) {
545 # $repo is a local path to a git index. We'll also resolve ../foo
546 # to ../foo/.git if the latter is a directory. To help
547 # disambiguate local paths from named hosted repositories, this
548 # form must be given as ./ or ../ if it's a relative path.
549 if (-d "$repo/.git") {
550 $repo = "$repo/.git";
552 $repo_location = 'local';
554 # $repo is none of the above. It must be the name of a hosted
556 my $arv_repo_list = api_call("repositories/list",
557 'filters' => [['name','=',$repo]]);
558 my @repos_found = @{$arv_repo_list->{'items'}};
559 my $n_found = $arv_repo_list->{'serverResponse'}->{'items_available'};
561 Log(undef, "Repository '$repo' -> "
562 . join(", ", map { $_->{'uuid'} } @repos_found));
565 croak("Error: Found $n_found repositories with name '$repo'.");
567 $repo = $repos_found[0]->{'fetch_url'};
568 $repo_location = 'remote';
570 Log(undef, "Using $repo_location repository '$repo'");
571 $ENV{"CRUNCH_SRC_URL"} = $repo;
573 # Resolve given script_version (we'll call that $treeish here) to a
574 # commit sha1 ($commit).
575 my $treeish = $Job->{'script_version'};
577 if ($repo_location eq 'remote') {
578 # We minimize excess object-fetching by re-using the same bare
579 # repository in CRUNCH_TMP/.git for multiple crunch-jobs -- we
580 # just keep adding remotes to it as needed.
581 my $local_repo = $ENV{'CRUNCH_TMP'}."/.git";
582 my $gitcmd = "git --git-dir=\Q$local_repo\E";
584 # Set up our local repo for caching remote objects, making
586 if (!-d $local_repo) {
587 make_path($local_repo) or croak("Error: could not create $local_repo");
589 # This works (exits 0 and doesn't delete fetched objects) even
590 # if $local_repo is already initialized:
591 `$gitcmd init --bare`;
593 croak("Error: $gitcmd init --bare exited ".exit_status_s($?));
596 # If $treeish looks like a hash (or abbrev hash) we look it up in
597 # our local cache first, since that's cheaper. (We don't want to
598 # do that with tags/branches though -- those change over time, so
599 # they should always be resolved by the remote repo.)
600 if ($treeish =~ /^[0-9a-f]{7,40}$/s) {
601 # Hide stderr because it's normal for this to fail:
602 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E 2>/dev/null`;
604 # Careful not to resolve a branch named abcdeff to commit 1234567:
605 $sha1 =~ /^$treeish/ &&
606 $sha1 =~ /^([0-9a-f]{40})$/s) {
608 Log(undef, "Commit $commit already present in $local_repo");
612 if (!defined $commit) {
613 # If $treeish isn't just a hash or abbrev hash, or isn't here
614 # yet, we need to fetch the remote to resolve it correctly.
616 # First, remove all local heads. This prevents a name that does
617 # not exist on the remote from resolving to (or colliding with)
618 # a previously fetched branch or tag (possibly from a different
620 remove_tree("$local_repo/refs/heads", {keep_root => 1});
622 Log(undef, "Fetching objects from $repo to $local_repo");
623 `$gitcmd fetch --no-progress --tags ''\Q$repo\E \Q+refs/heads/*:refs/heads/*\E`;
625 croak("Error: `$gitcmd fetch` exited ".exit_status_s($?));
629 # Now that the data is all here, we will use our local repo for
630 # the rest of our git activities.
634 my $gitcmd = "git --git-dir=\Q$repo\E";
635 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E`;
636 unless ($? == 0 && $sha1 =~ /^([0-9a-f]{40})$/) {
637 croak("`$gitcmd rev-list` exited "
639 .", '$treeish' not found, giving up");
642 Log(undef, "Version $treeish is commit $commit");
644 if ($commit ne $Job->{'script_version'}) {
645 # Record the real commit id in the database, frozentokey, logs,
646 # etc. -- instead of an abbreviation or a branch name which can
647 # become ambiguous or point to a different commit in the future.
648 if (!$Job->update_attributes('script_version' => $commit)) {
649 croak("Error: failed to update job's script_version attribute");
653 $ENV{"CRUNCH_SRC_COMMIT"} = $commit;
654 add_git_archive("$gitcmd archive ''\Q$commit\E");
657 my $git_archive = combined_git_archive();
658 if (!defined $git_archive) {
659 Log(undef, "Skip install phase (no git archive)");
661 Log(undef, "Warning: This probably means workers have no source tree!");
666 my $install_script_tries_left = 3;
667 for (my $attempts = 0; $attempts < 3; $attempts++) {
668 my @srunargs = ("srun",
669 "--nodelist=$nodelist",
670 "-D", $ENV{'TMPDIR'}, "--job-name=$job_id");
671 my @execargs = ("sh", "-c",
672 "mkdir -p $ENV{CRUNCH_INSTALL} && cd $ENV{CRUNCH_TMP} && perl -");
674 $ENV{"CRUNCH_GIT_ARCHIVE_HASH"} = md5_hex($git_archive);
675 my ($stdout, $stderr);
676 ($exited, $stdout, $stderr) = srun_sync(
677 \@srunargs, \@execargs,
678 {label => "run install script on all workers"},
679 $build_script . $git_archive);
681 my $stderr_anything_from_script = 0;
682 for my $line (split(/\n/, $stderr)) {
683 if ($line !~ /^(srun: error: |starting: \[)/) {
684 $stderr_anything_from_script = 1;
688 last if $exited == 0 || $main::please_freeze;
690 # If the install script fails but doesn't print an error message,
691 # the next thing anyone is likely to do is just run it again in
692 # case it was a transient problem like "slurm communication fails
693 # because the network isn't reliable enough". So we'll just do
694 # that ourselves (up to 3 attempts in total). OTOH, if there is an
695 # error message, the problem is more likely to have a real fix and
696 # we should fail the job so the fixing process can start, instead
697 # of doing 2 more attempts.
698 last if $stderr_anything_from_script;
701 foreach my $tar_filename (map { tar_filename_n($_); } (1..$git_tar_count)) {
702 unlink($tar_filename);
710 foreach (qw (script script_version script_parameters runtime_constraints))
714 (ref($Job->{$_}) ? JSON::encode_json($Job->{$_}) : $Job->{$_}));
716 foreach (split (/\n/, $Job->{knobs}))
718 Log (undef, "knob " . $_);
722 'filters' => [['hostname', 'in', \@node]],
723 'order' => 'hostname',
724 'limit' => scalar(@node),
726 for my $n (@{$resp->{items}}) {
727 Log(undef, "$n->{hostname} $n->{uuid} ".JSON::encode_json($n->{properties}));
732 $main::success = undef;
738 my $thisround_succeeded = 0;
739 my $thisround_failed = 0;
740 my $thisround_failed_multiple = 0;
741 my $working_slot_count = scalar(@slot);
743 @jobstep_todo = sort { $jobstep[$a]->{level} <=> $jobstep[$b]->{level}
744 or $a <=> $b } @jobstep_todo;
745 my $level = $jobstep[$jobstep_todo[0]]->{level};
747 my $initial_tasks_this_level = 0;
748 foreach my $id (@jobstep_todo) {
749 $initial_tasks_this_level++ if ($jobstep[$id]->{level} == $level);
752 # If the number of tasks scheduled at this level #T is smaller than the number
753 # of slots available #S, only use the first #T slots, or the first slot on
754 # each node, whichever number is greater.
756 # When we dispatch tasks later, we'll allocate whole-node resources like RAM
757 # based on these numbers. Using fewer slots makes more resources available
758 # to each individual task, which should normally be a better strategy when
759 # there are fewer of them running with less parallelism.
761 # Note that this calculation is not redone if the initial tasks at
762 # this level queue more tasks at the same level. This may harm
763 # overall task throughput for that level.
765 if ($initial_tasks_this_level < @node) {
766 @freeslot = (0..$#node);
767 } elsif ($initial_tasks_this_level < @slot) {
768 @freeslot = (0..$initial_tasks_this_level - 1);
770 @freeslot = (0..$#slot);
772 my $round_num_freeslots = scalar(@freeslot);
773 print STDERR "crunch-job have ${round_num_freeslots} free slots for ${initial_tasks_this_level} initial tasks at this level, ".scalar(@node)." nodes, and ".scalar(@slot)." slots\n";
775 my %round_max_slots = ();
776 for (my $ii = $#freeslot; $ii >= 0; $ii--) {
777 my $this_slot = $slot[$freeslot[$ii]];
778 my $node_name = $this_slot->{node}->{name};
779 $round_max_slots{$node_name} ||= $this_slot->{cpu};
780 last if (scalar(keys(%round_max_slots)) >= @node);
783 Log(undef, "start level $level with $round_num_freeslots slots");
786 my $progress_is_dirty = 1;
787 my $progress_stats_updated = 0;
789 update_progress_stats();
793 for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
795 # Don't create new tasks if we already know the job's final result.
796 last if defined($main::success);
798 my $id = $jobstep_todo[$todo_ptr];
799 my $Jobstep = $jobstep[$id];
800 if ($Jobstep->{level} != $level)
805 pipe $reader{$id}, "writer" or croak("pipe() failed: $!");
806 set_nonblocking($reader{$id});
808 my $childslot = $freeslot[0];
809 my $childnode = $slot[$childslot]->{node};
810 my $childslotname = join (".",
811 $slot[$childslot]->{node}->{name},
812 $slot[$childslot]->{cpu});
814 my $childpid = fork();
817 $SIG{'INT'} = 'DEFAULT';
818 $SIG{'QUIT'} = 'DEFAULT';
819 $SIG{'TERM'} = 'DEFAULT';
821 foreach (values (%reader))
825 fcntl ("writer", F_SETFL, 0) or croak ($!); # no close-on-exec
826 open(STDOUT,">&writer");
827 open(STDERR,">&writer");
832 delete $ENV{"GNUPGHOME"};
833 $ENV{"TASK_UUID"} = $Jobstep->{'arvados_task'}->{'uuid'};
834 $ENV{"TASK_QSEQUENCE"} = $id;
835 $ENV{"TASK_SEQUENCE"} = $level;
836 $ENV{"JOB_SCRIPT"} = $Job->{script};
837 while (my ($param, $value) = each %{$Job->{script_parameters}}) {
838 $param =~ tr/a-z/A-Z/;
839 $ENV{"JOB_PARAMETER_$param"} = $value;
841 $ENV{"TASK_SLOT_NODE"} = $slot[$childslot]->{node}->{name};
842 $ENV{"TASK_SLOT_NUMBER"} = $slot[$childslot]->{cpu};
843 $ENV{"TASK_WORK"} = $ENV{"CRUNCH_TMP"}."/task/$childslotname";
844 $ENV{"HOME"} = $ENV{"TASK_WORK"};
845 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
846 $ENV{"CRUNCH_NODE_SLOTS"} = $round_max_slots{$ENV{TASK_SLOT_NODE}};
847 $ENV{"PATH"} = $ENV{"CRUNCH_INSTALL"} . "/bin:" . $ENV{"PATH"};
849 my $keep_mnt = $ENV{"TASK_WORK"}.".keep";
855 "--nodelist=".$childnode->{name},
856 qw(-n1 -c1 -N1 -D), $ENV{'TMPDIR'},
857 "--job-name=$job_id.$id.$$",
860 my $stdbuf = " stdbuf --output=0 --error=0 ";
862 my $arv_file_cache = "";
863 if (defined($Job->{'runtime_constraints'}->{'keep_cache_mb_per_task'})) {
864 $arv_file_cache = "--file-cache=" . ($Job->{'runtime_constraints'}->{'keep_cache_mb_per_task'} * 1024 * 1024);
868 "if [ -e \Q$ENV{TASK_WORK}\E ]; then rm -rf \Q$ENV{TASK_WORK}\E; fi; "
869 ."mkdir -p \Q$ENV{CRUNCH_TMP}\E \Q$ENV{JOB_WORK}\E \Q$ENV{TASK_WORK}\E \Q$keep_mnt\E "
870 ."&& cd \Q$ENV{CRUNCH_TMP}\E "
871 # These environment variables get used explicitly later in
872 # $command. No tool is expected to read these values directly.
873 .q{&& MEM=$(awk '($1 == "MemTotal:"){print $2}' </proc/meminfo) }
874 .q{&& SWAP=$(awk '($1 == "SwapTotal:"){print $2}' </proc/meminfo) }
875 ."&& MEMLIMIT=\$(( (\$MEM * 95) / ($ENV{CRUNCH_NODE_SLOTS} * 100) )) "
876 ."&& let SWAPLIMIT=\$MEMLIMIT+\$SWAP "
877 .q{&& declare -a VOLUMES=() }
878 .q{&& if which crunchrunner >/dev/null ; then VOLUMES+=("--volume=$(which crunchrunner):/usr/local/bin/crunchrunner:ro") ; fi }
879 .q{&& if test -f /etc/ssl/certs/ca-certificates.crt ; then VOLUMES+=("--volume=/etc/ssl/certs/ca-certificates.crt:/etc/arvados/ca-certificates.crt:ro") ; }
880 .q{elif test -f /etc/pki/tls/certs/ca-bundle.crt ; then VOLUMES+=("--volume=/etc/pki/tls/certs/ca-bundle.crt:/etc/arvados/ca-certificates.crt:ro") ; fi };
882 $command .= "&& exec arv-mount --read-write --mount-by-pdh=by_pdh --mount-tmp=tmp --crunchstat-interval=10 --allow-other $arv_file_cache \Q$keep_mnt\E --exec ";
883 $ENV{TASK_KEEPMOUNT} = "$keep_mnt/by_pdh";
884 $ENV{TASK_KEEPMOUNT_TMP} = "$keep_mnt/tmp";
888 my $containername = "$Jobstep->{arvados_task}->{uuid}-$Jobstep->{failures}";
889 my $cidfile = "$ENV{CRUNCH_TMP}/$containername.cid";
890 $command .= "crunchstat -cgroup-root=\Q$cgroup_root\E -cgroup-parent=docker -cgroup-cid=$cidfile -poll=10000 ";
891 $command .= "$docker_bin run $docker_run_args --name=$containername --attach=stdout --attach=stderr --attach=stdin -i \Q$dockeruserarg\E --cidfile=$cidfile --sig-proxy ";
892 # We only set memory limits if Docker lets us limit both memory and swap.
893 # Memory limits alone have been supported longer, but subprocesses tend
894 # to get SIGKILL if they exceed that without any swap limit set.
895 # See #5642 for additional background.
896 if ($docker_limitmem) {
897 $command .= "--memory=\${MEMLIMIT}k --memory-swap=\${SWAPLIMIT}k ";
900 # The source tree and $destdir directory (which we have
901 # installed on the worker host) are available in the container,
902 # under the same path.
903 $command .= "--volume=\Q$ENV{CRUNCH_SRC}:$ENV{CRUNCH_SRC}:ro\E ";
904 $command .= "--volume=\Q$ENV{CRUNCH_INSTALL}:$ENV{CRUNCH_INSTALL}:ro\E ";
906 # Currently, we make the "by_pdh" directory in arv-mount's mount
907 # point appear at /keep inside the container (instead of using
908 # the same path as the host like we do with CRUNCH_SRC and
909 # CRUNCH_INSTALL). However, crunch scripts and utilities must
910 # not rely on this. They must use $TASK_KEEPMOUNT.
911 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT}:/keep:ro\E ";
912 $ENV{TASK_KEEPMOUNT} = "/keep";
914 # Ditto TASK_KEEPMOUNT_TMP, as /keep_tmp.
915 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT_TMP}:/keep_tmp\E ";
916 $ENV{TASK_KEEPMOUNT_TMP} = "/keep_tmp";
918 # TASK_WORK is almost exactly like a docker data volume: it
919 # starts out empty, is writable, and persists until no
920 # containers use it any more. We don't use --volumes-from to
921 # share it with other containers: it is only accessible to this
922 # task, and it goes away when this task stops.
924 # However, a docker data volume is writable only by root unless
925 # the mount point already happens to exist in the container with
926 # different permissions. Therefore, we [1] assume /tmp already
927 # exists in the image and is writable by the crunch user; [2]
928 # avoid putting TASK_WORK inside CRUNCH_TMP (which won't be
929 # writable if they are created by docker while setting up the
930 # other --volumes); and [3] create $TASK_WORK inside the
931 # container using $build_script.
932 $command .= "--volume=/tmp ";
933 $ENV{"TASK_WORK"} = "/tmp/crunch-job-task-work/$childslotname";
934 $ENV{"HOME"} = $ENV{"TASK_WORK"};
935 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
937 # TODO: Share a single JOB_WORK volume across all task
938 # containers on a given worker node, and delete it when the job
939 # ends (and, in case that doesn't work, when the next job
942 # For now, use the same approach as TASK_WORK above.
943 $ENV{"JOB_WORK"} = "/tmp/crunch-job-work";
945 # Bind mount the crunchrunner binary and host TLS certificates file into
947 $command .= '"${VOLUMES[@]}" ';
949 while (my ($env_key, $env_val) = each %ENV)
951 if ($env_key =~ /^(ARVADOS|CRUNCH|JOB|TASK)_/) {
952 $command .= "--env=\Q$env_key=$env_val\E ";
955 $command .= "--env=\QHOME=$ENV{HOME}\E ";
956 $command .= "\Q$docker_hash\E ";
958 if ($Job->{arvados_sdk_version}) {
960 $command .= "perl - \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E";
962 $command .= "/bin/sh -c \'python -c " .
963 '"from pkg_resources import get_distribution as get; print \"Using Arvados SDK version\", get(\"arvados-python-client\").version"' .
964 ">&2 2>/dev/null; " .
965 "mkdir -p \"$ENV{JOB_WORK}\" \"$ENV{TASK_WORK}\" && " .
966 "if which stdbuf >/dev/null ; then " .
967 " exec $stdbuf \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E ;" .
969 " exec \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E ;" .
974 $command .= "crunchstat -cgroup-root=\Q$cgroup_root\E -poll=10000 ";
976 $command .= "perl - $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
979 my @execargs = ('bash', '-c', $command);
980 srun (\@srunargs, \@execargs, undef, $build_script);
981 # exec() failed, we assume nothing happened.
982 die "srun() failed on build script\n";
985 if (!defined $childpid)
996 jobstepname => "$job_id.$id.$childpid",
998 croak ("assert failed: \$slot[$childslot]->{'pid'} exists") if exists $slot[$childslot]->{pid};
999 $slot[$childslot]->{pid} = $childpid;
1001 Log ($id, "job_task ".$Jobstep->{'arvados_task'}->{'uuid'});
1002 Log ($id, "child $childpid started on $childslotname");
1003 $Jobstep->{starttime} = time;
1004 $Jobstep->{node} = $childnode->{name};
1005 $Jobstep->{slotindex} = $childslot;
1006 delete $Jobstep->{stderr};
1007 delete $Jobstep->{finishtime};
1008 delete $Jobstep->{tempfail};
1010 $Jobstep->{'arvados_task'}->{started_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{starttime});
1011 $Jobstep->{'arvados_task'}->save;
1013 splice @jobstep_todo, $todo_ptr, 1;
1016 $progress_is_dirty = 1;
1020 ($round_num_freeslots > @freeslot && $todo_ptr+1 > $#jobstep_todo))
1022 last THISROUND if $main::please_freeze;
1023 if ($main::please_info)
1025 $main::please_info = 0;
1027 create_output_collection();
1029 update_progress_stats();
1034 if (!$gotsome || ($latest_refresh + 2 < scalar time))
1036 check_refresh_wanted();
1038 update_progress_stats();
1040 elsif (time - $progress_stats_updated >= 30 || $progress_is_dirty)
1042 update_progress_stats();
1045 select (undef, undef, undef, 0.1);
1047 $working_slot_count = scalar(grep { $_->{node}->{fail_count} == 0 &&
1048 $_->{node}->{hold_count} < 4 } @slot);
1049 if (($thisround_failed_multiple >= 8 && $thisround_succeeded == 0) ||
1050 ($thisround_failed_multiple >= 16 && $thisround_failed_multiple > $thisround_succeeded))
1052 my $message = "Repeated failure rate too high ($thisround_failed_multiple/"
1053 .($thisround_failed+$thisround_succeeded)
1054 .") -- giving up on this round";
1055 Log (undef, $message);
1059 # move slots from freeslot to holdslot (or back to freeslot) if necessary
1060 for (my $i=$#freeslot; $i>=0; $i--) {
1061 if ($slot[$freeslot[$i]]->{node}->{hold_until} > scalar time) {
1062 push @holdslot, (splice @freeslot, $i, 1);
1065 for (my $i=$#holdslot; $i>=0; $i--) {
1066 if ($slot[$holdslot[$i]]->{node}->{hold_until} <= scalar time) {
1067 push @freeslot, (splice @holdslot, $i, 1);
1071 # give up if no nodes are succeeding
1072 if ($working_slot_count < 1) {
1073 Log(undef, "Every node has failed -- giving up");
1080 push @freeslot, splice @holdslot;
1081 map { $slot[$freeslot[$_]]->{node}->{losing_streak} = 0 } (0..$#freeslot);
1084 Log (undef, "wait for last ".(scalar keys %proc)." children to finish");
1087 if ($main::please_continue) {
1088 $main::please_continue = 0;
1091 $main::please_info = 0, freeze(), create_output_collection(), save_meta(1) if $main::please_info;
1093 if (!reapchildren())
1095 check_refresh_wanted();
1097 update_progress_stats();
1098 select (undef, undef, undef, 0.1);
1099 killem (keys %proc) if $main::please_freeze;
1103 update_progress_stats();
1104 freeze_if_want_freeze();
1107 if (!defined $main::success)
1109 if (!@jobstep_todo) {
1111 } elsif ($working_slot_count < 1) {
1112 save_output_collection();
1114 exit(EX_RETRY_UNLOCKED);
1115 } elsif ($thisround_succeeded == 0 &&
1116 ($thisround_failed == 0 || $thisround_failed > 4)) {
1117 my $message = "stop because $thisround_failed tasks failed and none succeeded";
1118 Log (undef, $message);
1123 goto ONELEVEL if !defined $main::success;
1126 release_allocation();
1128 my $collated_output = save_output_collection();
1129 Log (undef, "finish");
1134 if ($collated_output && $main::success) {
1135 $final_state = 'Complete';
1137 $final_state = 'Failed';
1139 $Job->update_attributes('state' => $final_state);
1141 exit (($final_state eq 'Complete') ? 0 : 1);
1145 sub update_progress_stats
1147 $progress_stats_updated = time;
1148 return if !$progress_is_dirty;
1149 my ($todo, $done, $running) = (scalar @jobstep_todo,
1150 scalar @jobstep_done,
1151 scalar keys(%proc));
1152 $Job->{'tasks_summary'} ||= {};
1153 $Job->{'tasks_summary'}->{'todo'} = $todo;
1154 $Job->{'tasks_summary'}->{'done'} = $done;
1155 $Job->{'tasks_summary'}->{'running'} = $running;
1156 $Job->update_attributes('tasks_summary' => $Job->{'tasks_summary'});
1157 Log (undef, "status: $done done, $running running, $todo todo");
1158 $progress_is_dirty = 0;
1165 my $children_reaped = 0;
1166 my @successful_task_uuids = ();
1168 while((my $pid = waitpid (-1, WNOHANG)) > 0)
1170 my $childstatus = $?;
1172 my $whatslot = ($slot[$proc{$pid}->{slot}]->{node}->{name}
1174 . $slot[$proc{$pid}->{slot}]->{cpu});
1175 my $jobstepidx = $proc{$pid}->{jobstepidx};
1178 my $elapsed = time - $proc{$pid}->{time};
1179 my $Jobstep = $jobstep[$jobstepidx];
1181 my $exitvalue = $childstatus >> 8;
1182 my $exitinfo = "exit ".exit_status_s($childstatus);
1183 $Jobstep->{'arvados_task'}->reload;
1184 my $task_success = $Jobstep->{'arvados_task'}->{success};
1186 Log ($jobstepidx, "child $pid on $whatslot $exitinfo success=$task_success");
1188 if (!defined $task_success) {
1189 # task did not indicate one way or the other --> fail
1190 Log($jobstepidx, sprintf(
1191 "ERROR: Task process exited %s, but never updated its task record to indicate success and record its output.",
1192 exit_status_s($childstatus)));
1193 $Jobstep->{'arvados_task'}->{success} = 0;
1194 $Jobstep->{'arvados_task'}->save;
1201 $temporary_fail ||= $Jobstep->{tempfail};
1202 $temporary_fail ||= ($exitvalue == TASK_TEMPFAIL);
1204 ++$thisround_failed;
1205 ++$thisround_failed_multiple if $Jobstep->{'failures'} >= 1;
1207 # Check for signs of a failed or misconfigured node
1208 if (++$slot[$proc{$pid}->{slot}]->{node}->{losing_streak} >=
1209 2+$slot[$proc{$pid}->{slot}]->{node}->{ncpus}) {
1210 # Don't count this against jobstep failure thresholds if this
1211 # node is already suspected faulty and srun exited quickly
1212 if ($slot[$proc{$pid}->{slot}]->{node}->{hold_until} &&
1214 Log ($jobstepidx, "blaming failure on suspect node " .
1215 $slot[$proc{$pid}->{slot}]->{node}->{name});
1216 $temporary_fail ||= 1;
1218 ban_node_by_slot($proc{$pid}->{slot});
1221 Log ($jobstepidx, sprintf('failure (#%d, %s) after %d seconds',
1222 ++$Jobstep->{'failures'},
1223 $temporary_fail ? 'temporary' : 'permanent',
1226 if (!$temporary_fail || $Jobstep->{'failures'} >= 3) {
1227 # Give up on this task, and the whole job
1230 # Put this task back on the todo queue
1231 push @jobstep_todo, $jobstepidx;
1232 $Job->{'tasks_summary'}->{'failed'}++;
1236 push @successful_task_uuids, $Jobstep->{'arvados_task'}->{uuid};
1237 ++$thisround_succeeded;
1238 $slot[$proc{$pid}->{slot}]->{node}->{losing_streak} = 0;
1239 $slot[$proc{$pid}->{slot}]->{node}->{hold_until} = 0;
1240 $slot[$proc{$pid}->{slot}]->{node}->{fail_count} = 0;
1241 push @jobstep_done, $jobstepidx;
1242 Log ($jobstepidx, "success in $elapsed seconds");
1244 $Jobstep->{exitcode} = $childstatus;
1245 $Jobstep->{finishtime} = time;
1246 $Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
1247 $Jobstep->{'arvados_task'}->save;
1248 process_stderr_final ($jobstepidx);
1249 Log ($jobstepidx, sprintf("task output (%d bytes): %s",
1250 length($Jobstep->{'arvados_task'}->{output}),
1251 $Jobstep->{'arvados_task'}->{output}));
1253 close $reader{$jobstepidx};
1254 delete $reader{$jobstepidx};
1255 delete $slot[$proc{$pid}->{slot}]->{pid};
1256 push @freeslot, $proc{$pid}->{slot};
1259 $progress_is_dirty = 1;
1262 if (scalar(@successful_task_uuids) > 0)
1264 Log (undef, sprintf("%d tasks exited (%d succeeded), checking for new tasks from API server.", $children_reaped, scalar(@successful_task_uuids)));
1266 my $newtask_list = [];
1267 my $newtask_results;
1269 $newtask_results = api_call(
1271 'filters' => [["created_by_job_task_uuid","in",\@successful_task_uuids]],
1272 'order' => 'qsequence',
1273 'offset' => scalar(@$newtask_list),
1275 push(@$newtask_list, @{$newtask_results->{items}});
1276 } while (@{$newtask_results->{items}});
1277 Log (undef, sprintf("Got %d new tasks from API server.", scalar(@$newtask_list)));
1278 foreach my $arvados_task (@$newtask_list) {
1280 'level' => $arvados_task->{'sequence'},
1282 'arvados_task' => $arvados_task
1284 push @jobstep, $jobstep;
1285 push @jobstep_todo, $#jobstep;
1289 return $children_reaped;
1292 sub check_refresh_wanted
1294 my @stat = stat $ENV{"CRUNCH_REFRESH_TRIGGER"};
1296 $stat[9] > $latest_refresh &&
1297 # ...and we have actually locked the job record...
1298 $job_id eq $Job->{'uuid'}) {
1299 $latest_refresh = scalar time;
1300 my $Job2 = api_call("jobs/get", uuid => $jobspec);
1301 for my $attr ('cancelled_at',
1302 'cancelled_by_user_uuid',
1303 'cancelled_by_client_uuid',
1305 $Job->{$attr} = $Job2->{$attr};
1307 if ($Job->{'state'} ne "Running") {
1308 if ($Job->{'state'} eq "Cancelled") {
1309 Log (undef, "Job cancelled at " . $Job->{'cancelled_at'} . " by user " . $Job->{'cancelled_by_user_uuid'});
1311 Log (undef, "Job state unexpectedly changed to " . $Job->{'state'});
1314 $main::please_freeze = 1;
1321 my $last_squeue_check = $squeue_checked;
1323 # Do not call `squeue` or check the kill list more than once every
1325 return if $last_squeue_check > time - 15;
1326 $squeue_checked = time;
1328 # Look for children from which we haven't received stderr data since
1329 # the last squeue check. If no such children exist, all procs are
1330 # alive and there's no need to even look at squeue.
1332 # As long as the crunchstat poll interval (10s) is shorter than the
1333 # squeue check interval (15s) this should make the squeue check an
1335 my $silent_procs = 0;
1336 for my $js (map {$jobstep[$_->{jobstepidx}]} values %proc)
1338 if (!exists($js->{stderr_at}))
1340 $js->{stderr_at} = 0;
1342 if ($js->{stderr_at} < $last_squeue_check)
1347 return if $silent_procs == 0;
1349 # use killem() on procs whose killtime is reached
1350 while (my ($pid, $procinfo) = each %proc)
1352 my $js = $jobstep[$procinfo->{jobstepidx}];
1353 if (exists $procinfo->{killtime}
1354 && $procinfo->{killtime} <= time
1355 && $js->{stderr_at} < $last_squeue_check)
1358 if ($js->{stderr_at}) {
1359 $sincewhen = " in last " . (time - $js->{stderr_at}) . "s";
1361 Log($procinfo->{jobstepidx}, "killing orphaned srun process $pid (task not in slurm queue, no stderr received$sincewhen)");
1368 # here is an opportunity to check for mysterious problems with local procs
1372 # Get a list of steps still running. Note: squeue(1) says --steps
1373 # selects a format (which we override anyway) and allows us to
1374 # specify which steps we're interested in (which we don't).
1375 # Importantly, it also changes the meaning of %j from "job name" to
1376 # "step name" and (although this isn't mentioned explicitly in the
1377 # docs) switches from "one line per job" mode to "one line per step"
1378 # mode. Without it, we'd just get a list of one job, instead of a
1380 my @squeue = `squeue --jobs=\Q$ENV{SLURM_JOB_ID}\E --steps --format='%j' --noheader`;
1383 Log(undef, "warning: squeue exit status $? ($!)");
1388 # which of my jobsteps are running, according to squeue?
1390 for my $jobstepname (@squeue)
1392 $ok{$jobstepname} = 1;
1395 # Check for child procs >60s old and not mentioned by squeue.
1396 while (my ($pid, $procinfo) = each %proc)
1398 if ($procinfo->{time} < time - 60
1399 && $procinfo->{jobstepname}
1400 && !exists $ok{$procinfo->{jobstepname}}
1401 && !exists $procinfo->{killtime})
1403 # According to slurm, this task has ended (successfully or not)
1404 # -- but our srun child hasn't exited. First we must wait (30
1405 # seconds) in case this is just a race between communication
1406 # channels. Then, if our srun child process still hasn't
1407 # terminated, we'll conclude some slurm communication
1408 # error/delay has caused the task to die without notifying srun,
1409 # and we'll kill srun ourselves.
1410 $procinfo->{killtime} = time + 30;
1411 Log($procinfo->{jobstepidx}, "notice: task is not in slurm queue but srun process $pid has not exited");
1418 # If a node fails in a multi-node "srun" call during job setup, the call
1419 # may hang instead of exiting with a nonzero code. This function checks
1420 # "sinfo" for the health of the nodes that were allocated and ensures that
1421 # they are all still in the "alloc" state. If a node that is allocated to
1422 # this job is not in "alloc" state, then set please_freeze.
1424 # This is only called from srun_sync() for node configuration. If a
1425 # node fails doing actual work, there are other recovery mechanisms.
1427 # Do not call `sinfo` more than once every 15 seconds.
1428 return if $sinfo_checked > time - 15;
1429 $sinfo_checked = time;
1431 # The output format "%t" means output node states.
1432 my @sinfo = `sinfo --nodes=\Q$ENV{SLURM_NODELIST}\E --noheader -o "%t"`;
1435 Log(undef, "warning: sinfo exit status $? ($!)");
1442 if ($_ != "alloc" && $_ != "alloc*") {
1443 $main::please_freeze = 1;
1448 sub release_allocation
1452 Log (undef, "release job allocation");
1453 system "scancel $ENV{SLURM_JOB_ID}";
1462 my $sel = IO::Select->new();
1463 foreach my $jobstepidx (keys %reader)
1465 my $fd = $reader{$jobstepidx};
1467 $fd_job{$fd} = $jobstepidx;
1469 if (my $stdout_fd = $jobstep[$jobstepidx]->{stdout_r}) {
1470 $sel->add($stdout_fd);
1471 $fd_job{$stdout_fd} = $jobstepidx;
1474 # select on all reader fds with 0.1s timeout
1475 my @ready_fds = $sel->can_read(0.1);
1476 foreach my $fd (@ready_fds)
1479 if (0 < sysread ($fd, $buf, 65536))
1482 print STDERR $buf if $ENV{CRUNCH_DEBUG};
1484 my $jobstepidx = $fd_job{$fd};
1485 if ($jobstep[$jobstepidx]->{stdout_r} == $fd) {
1486 $jobstep[$jobstepidx]->{stdout_captured} .= $buf;
1490 $jobstep[$jobstepidx]->{stderr_at} = time;
1491 $jobstep[$jobstepidx]->{stderr} .= $buf;
1493 # Consume everything up to the last \n
1494 preprocess_stderr ($jobstepidx);
1496 if (length ($jobstep[$jobstepidx]->{stderr}) > 16384)
1498 # If we get a lot of stderr without a newline, chop off the
1499 # front to avoid letting our buffer grow indefinitely.
1500 substr ($jobstep[$jobstepidx]->{stderr},
1501 0, length($jobstep[$jobstepidx]->{stderr}) - 8192) = "";
1509 # Consume all full lines of stderr for a jobstep. Everything after the
1510 # last newline will remain in $jobstep[$jobstepidx]->{stderr} after
1512 sub preprocess_stderr
1514 my $jobstepidx = shift;
1515 # slotindex is only defined for children running Arvados job tasks.
1516 # Be prepared to handle the undef case (for setup srun calls, etc.).
1517 my $job_slot_index = $jobstep[$jobstepidx]->{slotindex};
1519 while ($jobstep[$jobstepidx]->{stderr} =~ /^(.*?)\n/) {
1521 substr $jobstep[$jobstepidx]->{stderr}, 0, 1+length($line), "";
1522 Log ($jobstepidx, "stderr $line");
1523 if ($line =~ /srun: error: (SLURM job $ENV{SLURM_JOB_ID} has expired|Unable to confirm allocation for job $ENV{SLURM_JOB_ID})/i) {
1524 # If the allocation is revoked, we can't possibly continue, so mark all
1525 # nodes as failed. This will cause the overall exit code to be
1526 # EX_RETRY_UNLOCKED instead of failure so that crunch_dispatch can re-run
1528 $main::please_freeze = 1;
1529 foreach my $st (@slot) {
1530 $st->{node}->{fail_count}++;
1533 elsif ($line =~ /srun: error: .*?\b(Node failure on|Aborting, .*?\bio error\b)/i) {
1534 $jobstep[$jobstepidx]->{tempfail} = 1;
1535 if (defined($job_slot_index)) {
1536 $slot[$job_slot_index]->{node}->{fail_count}++;
1537 ban_node_by_slot($job_slot_index);
1540 elsif ($line =~ /srun: error: (Unable to create job step|.*?: Communication connection failure)/i) {
1541 $jobstep[$jobstepidx]->{tempfail} = 1;
1542 ban_node_by_slot($job_slot_index) if (defined($job_slot_index));
1544 elsif ($line =~ /\bKeep(Read|Write|Request)Error:/) {
1545 $jobstep[$jobstepidx]->{tempfail} = 1;
1551 sub process_stderr_final
1553 my $jobstepidx = shift;
1554 preprocess_stderr ($jobstepidx);
1557 Log ($jobstepidx, "stderr $_");
1558 } split ("\n", $jobstep[$jobstepidx]->{stderr});
1559 $jobstep[$jobstepidx]->{stderr} = '';
1566 if (!open($keep, "-|", "arv-get", "--retries", retry_count(), $hash)) {
1567 Log(undef, "fetch_block run error from arv-get $hash: $!");
1570 my $output_block = "";
1573 my $bytes = sysread($keep, $buf, 1024 * 1024);
1574 if (!defined $bytes) {
1575 Log(undef, "fetch_block read error from arv-get: $!");
1576 $output_block = undef;
1578 } elsif ($bytes == 0) {
1579 # sysread returns 0 at the end of the pipe.
1582 # some bytes were read into buf.
1583 $output_block .= $buf;
1588 Log(undef, "fetch_block arv-get exited " . exit_status_s($?));
1589 $output_block = undef;
1591 return $output_block;
1594 # Create a collection by concatenating the output of all tasks (each
1595 # task's output is either a manifest fragment, a locator for a
1596 # manifest fragment stored in Keep, or nothing at all). Return the
1597 # portable_data_hash of the new collection.
1598 sub create_output_collection
1600 Log (undef, "collate");
1602 my ($child_out, $child_in);
1603 my $pid = open2($child_out, $child_in, 'python', '-c', q{
1606 print (arvados.api("v1").collections().
1607 create(body={"manifest_text": sys.stdin.read(),
1608 "owner_uuid": sys.argv[2]}).
1609 execute(num_retries=int(sys.argv[1]))["portable_data_hash"])
1610 }, retry_count(), $Job->{owner_uuid});
1613 my $manifest_size = 0;
1617 my $output = $_->{'arvados_task'}->{output};
1618 next if (!defined($output));
1620 if ($output =~ /^[0-9a-f]{32}(\+\S+)*$/) {
1621 $next_write = fetch_block($output);
1623 $next_write = $output;
1625 if (defined($next_write)) {
1626 if (!defined(syswrite($child_in, $next_write))) {
1627 # There's been an error writing. Stop the loop.
1628 # We'll log details about the exit code later.
1631 $manifest_size += length($next_write);
1634 my $uuid = $_->{'arvados_task'}->{'uuid'};
1635 Log (undef, "Error retrieving '$output' output by task $task_idx ($uuid)");
1640 Log(undef, "collated output manifest text to send to API server is $manifest_size bytes with access tokens");
1643 my $s = IO::Select->new($child_out);
1644 if ($s->can_read(120)) {
1645 sysread($child_out, $joboutput, 1024 * 1024);
1648 Log(undef, "output collection creation exited " . exit_status_s($?));
1654 Log (undef, "timed out while creating output collection");
1655 foreach my $signal (2, 2, 2, 15, 15, 9) {
1656 kill($signal, $pid);
1657 last if waitpid($pid, WNOHANG) == -1;
1666 # Calls create_output_collection, logs the result, and returns it.
1667 # If that was successful, save that as the output in the job record.
1668 sub save_output_collection {
1669 my $collated_output = create_output_collection();
1671 if (!$collated_output) {
1672 Log(undef, "Failed to write output collection");
1675 Log(undef, "job output $collated_output");
1676 $Job->update_attributes('output' => $collated_output);
1678 return $collated_output;
1685 my $sig = 2; # SIGINT first
1686 if (exists $proc{$_}->{"sent_$sig"} &&
1687 time - $proc{$_}->{"sent_$sig"} > 4)
1689 $sig = 15; # SIGTERM if SIGINT doesn't work
1691 if (exists $proc{$_}->{"sent_$sig"} &&
1692 time - $proc{$_}->{"sent_$sig"} > 4)
1694 $sig = 9; # SIGKILL if SIGTERM doesn't work
1696 if (!exists $proc{$_}->{"sent_$sig"})
1698 Log ($proc{$_}->{jobstepidx}, "sending 2x signal $sig to pid $_");
1700 select (undef, undef, undef, 0.1);
1703 kill $sig, $_; # srun wants two SIGINT to really interrupt
1705 $proc{$_}->{"sent_$sig"} = time;
1706 $proc{$_}->{"killedafter"} = time - $proc{$_}->{"time"};
1716 vec($bits,fileno($_),1) = 1;
1722 # Send log output to Keep via arv-put.
1724 # $log_pipe_in and $log_pipe_out are the input and output filehandles to the arv-put pipe.
1725 # $log_pipe_out_buf is a string containing all output read from arv-put so far.
1726 # $log_pipe_out_select is an IO::Select object around $log_pipe_out.
1727 # $log_pipe_pid is the pid of the arv-put subprocess.
1729 # The only functions that should access these variables directly are:
1731 # log_writer_start($logfilename)
1732 # Starts an arv-put pipe, reading data on stdin and writing it to
1733 # a $logfilename file in an output collection.
1735 # log_writer_read_output([$timeout])
1736 # Read output from $log_pipe_out and append it to $log_pipe_out_buf.
1737 # Passes $timeout to the select() call, with a default of 0.01.
1738 # Returns the result of the last read() call on $log_pipe_out, or
1739 # -1 if read() wasn't called because select() timed out.
1740 # Only other log_writer_* functions should need to call this.
1742 # log_writer_send($txt)
1743 # Writes $txt to the output log collection.
1745 # log_writer_finish()
1746 # Closes the arv-put pipe and returns the output that it produces.
1748 # log_writer_is_active()
1749 # Returns a true value if there is currently a live arv-put
1750 # process, false otherwise.
1752 my ($log_pipe_in, $log_pipe_out, $log_pipe_out_buf, $log_pipe_out_select,
1755 sub log_writer_start($)
1757 my $logfilename = shift;
1758 $log_pipe_pid = open2($log_pipe_out, $log_pipe_in,
1762 '--filename', $logfilename,
1764 $log_pipe_out_buf = "";
1765 $log_pipe_out_select = IO::Select->new($log_pipe_out);
1768 sub log_writer_read_output {
1769 my $timeout = shift || 0.01;
1771 while ($read && $log_pipe_out_select->can_read($timeout)) {
1772 $read = read($log_pipe_out, $log_pipe_out_buf, 65536,
1773 length($log_pipe_out_buf));
1775 if (!defined($read)) {
1776 Log(undef, "error reading log manifest from arv-put: $!");
1781 sub log_writer_send($)
1784 print $log_pipe_in $txt;
1785 log_writer_read_output();
1788 sub log_writer_finish()
1790 return unless $log_pipe_pid;
1792 close($log_pipe_in);
1794 my $logger_failed = 0;
1795 my $read_result = log_writer_read_output(600);
1796 if ($read_result == -1) {
1797 $logger_failed = -1;
1798 Log (undef, "timed out reading from 'arv-put'");
1799 } elsif ($read_result != 0) {
1800 $logger_failed = -2;
1801 Log(undef, "failed to read arv-put log manifest to EOF");
1804 waitpid($log_pipe_pid, 0);
1806 $logger_failed ||= $?;
1807 Log(undef, "log_writer_finish: arv-put exited " . exit_status_s($?))
1810 close($log_pipe_out);
1811 my $arv_put_output = $logger_failed ? undef : $log_pipe_out_buf;
1812 $log_pipe_pid = $log_pipe_in = $log_pipe_out = $log_pipe_out_buf =
1813 $log_pipe_out_select = undef;
1815 return $arv_put_output;
1818 sub log_writer_is_active() {
1819 return $log_pipe_pid;
1822 sub Log # ($jobstepidx, $logmessage)
1824 my ($jobstepidx, $logmessage) = @_;
1825 if ($logmessage =~ /\n/) {
1826 for my $line (split (/\n/, $_[1])) {
1827 Log ($jobstepidx, $line);
1831 my $fh = select STDERR; $|=1; select $fh;
1833 if (defined($jobstepidx) && exists($jobstep[$jobstepidx]->{arvados_task})) {
1834 $task_qseq = $jobstepidx;
1836 my $message = sprintf ("%s %d %s %s", $job_id, $$, $task_qseq, $logmessage);
1837 $message =~ s{([^ -\176])}{"\\" . sprintf ("%03o", ord($1))}ge;
1840 if (log_writer_is_active() || -t STDERR) {
1841 my @gmtime = gmtime;
1842 $datetime = sprintf ("%04d-%02d-%02d_%02d:%02d:%02d",
1843 $gmtime[5]+1900, $gmtime[4]+1, @gmtime[3,2,1,0]);
1845 print STDERR ((-t STDERR) ? ($datetime." ".$message) : $message);
1847 if (log_writer_is_active()) {
1848 log_writer_send($datetime . " " . $message);
1855 my ($package, $file, $line) = caller;
1856 my $message = "@_ at $file line $line\n";
1857 Log (undef, $message);
1858 release_allocation();
1859 freeze() if @jobstep_todo;
1860 create_output_collection() if @jobstep_todo;
1870 if ($Job->{'state'} eq 'Cancelled') {
1871 $Job->update_attributes('finished_at' => scalar gmtime);
1873 $Job->update_attributes('state' => 'Failed');
1880 my $justcheckpoint = shift; # false if this will be the last meta saved
1881 return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm
1882 return unless log_writer_is_active();
1883 my $log_manifest = log_writer_finish();
1884 return unless defined($log_manifest);
1887 my $prev_log_coll = api_call("collections/get", uuid => $Job->{log});
1888 $log_manifest = $prev_log_coll->{manifest_text} . $log_manifest;
1891 my $log_coll = api_call(
1892 "collections/create", ensure_unique_name => 1, collection => {
1893 manifest_text => $log_manifest,
1894 owner_uuid => $Job->{owner_uuid},
1895 name => sprintf("Log from %s job %s", $Job->{script}, $Job->{uuid}),
1897 Log(undef, "log collection is " . $log_coll->{portable_data_hash});
1898 $Job->update_attributes('log' => $log_coll->{portable_data_hash});
1902 sub freeze_if_want_freeze
1904 if ($main::please_freeze)
1906 release_allocation();
1909 # kill some srun procs before freeze+stop
1910 map { $proc{$_} = {} } @_;
1913 killem (keys %proc);
1914 select (undef, undef, undef, 0.1);
1916 while (($died = waitpid (-1, WNOHANG)) > 0)
1918 delete $proc{$died};
1923 create_output_collection();
1933 Log (undef, "Freeze not implemented");
1940 croak ("Thaw not implemented");
1956 $s =~ s{\\(.)}{$1 eq "n" ? "\n" : $1}ge;
1962 my $srunargs = shift;
1963 my $execargs = shift;
1964 my $opts = shift || {};
1967 my $label = exists $opts->{label} ? $opts->{label} : "@$execargs";
1968 Log (undef, "$label: start");
1970 my ($stderr_r, $stderr_w);
1971 pipe $stderr_r, $stderr_w or croak("pipe() failed: $!");
1973 my ($stdout_r, $stdout_w);
1974 pipe $stdout_r, $stdout_w or croak("pipe() failed: $!");
1976 my $srunpid = fork();
1981 fcntl($stderr_w, F_SETFL, 0) or croak($!); # no close-on-exec
1982 fcntl($stdout_w, F_SETFL, 0) or croak($!);
1983 open(STDERR, ">&", $stderr_w);
1984 open(STDOUT, ">&", $stdout_w);
1985 srun ($srunargs, $execargs, $opts, $stdin);
1991 set_nonblocking($stderr_r);
1992 set_nonblocking($stdout_r);
1994 # Add entries to @jobstep and %proc so check_squeue() and
1995 # freeze_if_want_freeze() can treat it like a job task process.
1999 stderr_captured => '',
2000 stdout_r => $stdout_r,
2001 stdout_captured => '',
2003 my $jobstepidx = $#jobstep;
2005 jobstepidx => $jobstepidx,
2007 $reader{$jobstepidx} = $stderr_r;
2009 while ($srunpid != waitpid ($srunpid, WNOHANG)) {
2010 my $busy = readfrompipes();
2011 if (!$busy || ($latest_refresh + 2 < scalar time)) {
2012 check_refresh_wanted();
2017 select(undef, undef, undef, 0.1);
2019 killem(keys %proc) if $main::please_freeze;
2023 1 while readfrompipes();
2024 process_stderr_final ($jobstepidx);
2026 Log (undef, "$label: exit ".exit_status_s($exited));
2030 delete $proc{$srunpid};
2031 delete $reader{$jobstepidx};
2033 my $j = pop @jobstep;
2034 # If the srun showed signs of tempfail, ensure the caller treats that as a
2036 if ($main::please_freeze || $j->{tempfail}) {
2039 return ($exited, $j->{stdout_captured}, $j->{stderr_captured});
2045 my $srunargs = shift;
2046 my $execargs = shift;
2047 my $opts = shift || {};
2049 my $args = $have_slurm ? [@$srunargs, @$execargs] : $execargs;
2051 $Data::Dumper::Terse = 1;
2052 $Data::Dumper::Indent = 0;
2053 my $show_cmd = Dumper($args);
2054 $show_cmd =~ s/(TOKEN\\*=)[^\s\']+/${1}[...]/g;
2055 $show_cmd =~ s/\n/ /g;
2056 if ($opts->{fork}) {
2057 Log(undef, "starting: $show_cmd");
2059 # This is a child process: parent is in charge of reading our
2060 # stderr and copying it to Log() if needed.
2061 warn "starting: $show_cmd\n";
2064 if (defined $stdin) {
2065 my $child = open STDIN, "-|";
2066 defined $child or die "no fork: $!";
2068 print $stdin or die $!;
2069 close STDOUT or die $!;
2074 return system (@$args) if $opts->{fork};
2077 warn "ENV size is ".length(join(" ",%ENV));
2078 die "exec failed: $!: @$args";
2082 sub ban_node_by_slot {
2083 # Don't start any new jobsteps on this node for 60 seconds
2085 $slot[$slotid]->{node}->{hold_until} = 60 + scalar time;
2086 $slot[$slotid]->{node}->{hold_count}++;
2087 Log (undef, "backing off node " . $slot[$slotid]->{node}->{name} . " for 60 seconds");
2092 my ($lockfile, $error_message) = @_;
2093 open L, ">", $lockfile or croak("$lockfile: $!");
2094 if (!flock L, LOCK_EX|LOCK_NB) {
2095 croak("Can't lock $lockfile: $error_message\n");
2099 sub find_docker_image {
2100 # Given a Keep locator, check to see if it contains a Docker image.
2101 # If so, return its stream name and Docker hash.
2102 # If not, return undef for both values.
2103 my $locator = shift;
2104 my ($streamname, $filename);
2105 my $image = api_call("collections/get", uuid => $locator);
2107 foreach my $line (split(/\n/, $image->{manifest_text})) {
2108 my @tokens = split(/\s+/, $line);
2110 $streamname = shift(@tokens);
2111 foreach my $filedata (grep(/^\d+:\d+:/, @tokens)) {
2112 if (defined($filename)) {
2113 return (undef, undef); # More than one file in the Collection.
2115 $filename = (split(/:/, $filedata, 3))[2];
2120 if (defined($filename) and ($filename =~ /^((?:sha256:)?[0-9A-Fa-f]{64})\.tar$/)) {
2121 return ($streamname, $1);
2123 return (undef, undef);
2128 # Calculate the number of times an operation should be retried,
2129 # assuming exponential backoff, and that we're willing to retry as
2130 # long as tasks have been running. Enforce a minimum of 3 retries.
2131 my ($starttime, $endtime, $timediff, $retries);
2133 $starttime = $jobstep[0]->{starttime};
2134 $endtime = $jobstep[-1]->{finishtime};
2136 if (!defined($starttime)) {
2138 } elsif (!defined($endtime)) {
2139 $timediff = time - $starttime;
2141 $timediff = ($endtime - $starttime) - (time - $endtime);
2143 if ($timediff > 0) {
2144 $retries = int(log($timediff) / log(2));
2146 $retries = 1; # Use the minimum.
2148 return ($retries > 3) ? $retries : 3;
2152 # Pass in two function references.
2153 # This method will be called with the remaining arguments.
2154 # If it dies, retry it with exponential backoff until it succeeds,
2155 # or until the current retry_count is exhausted. After each failure
2156 # that can be retried, the second function will be called with
2157 # the current try count (0-based), next try time, and error message.
2158 my $operation = shift;
2159 my $retry_callback = shift;
2160 my $retries = retry_count();
2161 foreach my $try_count (0..$retries) {
2162 my $next_try = time + (2 ** $try_count);
2163 my $result = eval { $operation->(@_); };
2166 } elsif ($try_count < $retries) {
2167 $retry_callback->($try_count, $next_try, $@);
2168 my $sleep_time = $next_try - time;
2169 sleep($sleep_time) if ($sleep_time > 0);
2172 # Ensure the error message ends in a newline, so Perl doesn't add
2173 # retry_op's line number to it.
2179 # Pass in a /-separated API method name, and arguments for it.
2180 # This function will call that method, retrying as needed until
2181 # the current retry_count is exhausted, with a log on the first failure.
2182 my $method_name = shift;
2183 my $log_api_retry = sub {
2184 my ($try_count, $next_try_at, $errmsg) = @_;
2185 $errmsg =~ s/\s*\bat \Q$0\E line \d+\.?\s*//;
2186 $errmsg =~ s/\s/ /g;
2187 $errmsg =~ s/\s+$//;
2189 if ($next_try_at < time) {
2190 $retry_msg = "Retrying.";
2192 my $next_try_fmt = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($next_try_at);
2193 $retry_msg = "Retrying at $next_try_fmt.";
2195 Log(undef, "API method $method_name failed: $errmsg. $retry_msg");
2198 foreach my $key (split(/\//, $method_name)) {
2199 $method = $method->{$key};
2201 return retry_op(sub { $method->execute(@_); }, $log_api_retry, @_);
2205 # Given a $?, return a human-readable exit code string like "0" or
2206 # "1" or "0 with signal 1" or "1 with signal 11".
2207 my $exitcode = shift;
2208 my $s = $exitcode >> 8;
2209 if ($exitcode & 0x7f) {
2210 $s .= " with signal " . ($exitcode & 0x7f);
2212 if ($exitcode & 0x80) {
2213 $s .= " with core dump";
2218 sub handle_readall {
2219 # Pass in a glob reference to a file handle.
2220 # Read all its contents and return them as a string.
2221 my $fh_glob_ref = shift;
2223 return <$fh_glob_ref>;
2226 sub tar_filename_n {
2228 return sprintf("%s/git.%s.%d.tar", $ENV{CRUNCH_TMP}, $job_id, $n);
2231 sub add_git_archive {
2232 # Pass in a git archive command as a string or list, a la system().
2233 # This method will save its output to be included in the archive sent to the
2237 if (!open(GIT_ARCHIVE, ">", tar_filename_n($git_tar_count))) {
2238 croak("Failed to save git archive: $!");
2240 my $git_pid = open2(">&GIT_ARCHIVE", $git_input, @_);
2242 waitpid($git_pid, 0);
2245 croak("Failed to save git archive: git exited " . exit_status_s($?));
2249 sub combined_git_archive {
2250 # Combine all saved tar archives into a single archive, then return its
2251 # contents in a string. Return undef if no archives have been saved.
2252 if ($git_tar_count < 1) {
2255 my $base_tar_name = tar_filename_n(1);
2256 foreach my $tar_to_append (map { tar_filename_n($_); } (2..$git_tar_count)) {
2257 my $tar_exit = system("tar", "-Af", $base_tar_name, $tar_to_append);
2258 if ($tar_exit != 0) {
2259 croak("Error preparing build archive: tar -A exited " .
2260 exit_status_s($tar_exit));
2263 if (!open(GIT_TAR, "<", $base_tar_name)) {
2264 croak("Could not open build archive: $!");
2266 my $tar_contents = handle_readall(\*GIT_TAR);
2268 return $tar_contents;
2271 sub set_nonblocking {
2273 my $flags = fcntl ($fh, F_GETFL, 0) or croak ($!);
2274 fcntl ($fh, F_SETFL, $flags | O_NONBLOCK) or croak ($!);
2280 # This is crunch-job's internal dispatch script. crunch-job running on the API
2281 # server invokes this script on individual compute nodes, or localhost if we're
2282 # running a job locally. It gets called in two modes:
2284 # * No arguments: Installation mode. Read a tar archive from the DATA
2285 # file handle; it includes the Crunch script's source code, and
2286 # maybe SDKs as well. Those should be installed in the proper
2287 # locations. This runs outside of any Docker container, so don't try to
2288 # introspect Crunch's runtime environment.
2290 # * With arguments: Crunch script run mode. This script should set up the
2291 # environment, then run the command specified in the arguments. This runs
2292 # inside any Docker container.
2295 use File::Path qw( make_path remove_tree );
2296 use POSIX qw(getcwd);
2298 use constant TASK_TEMPFAIL => 111;
2300 # Map SDK subdirectories to the path environments they belong to.
2301 my %SDK_ENVVARS = ("perl/lib" => "PERLLIB", "ruby/lib" => "RUBYLIB");
2303 my $destdir = $ENV{"CRUNCH_SRC"};
2304 my $archive_hash = $ENV{"CRUNCH_GIT_ARCHIVE_HASH"};
2305 my $repo = $ENV{"CRUNCH_SRC_URL"};
2306 my $install_dir = $ENV{"CRUNCH_INSTALL"} || (getcwd() . "/opt");
2307 my $job_work = $ENV{"JOB_WORK"};
2308 my $task_work = $ENV{"TASK_WORK"};
2310 open(STDOUT_ORIG, ">&", STDOUT);
2311 open(STDERR_ORIG, ">&", STDERR);
2313 for my $dir ($destdir, $job_work, $task_work) {
2316 -e $dir or die "Failed to create temporary directory ($dir): $!";
2321 remove_tree($task_work, {keep_root => 1});
2324 ### Crunch script run mode
2326 # We want to do routine logging during task 0 only. This gives the user
2327 # the information they need, but avoids repeating the information for every
2330 if ($ENV{TASK_SEQUENCE} eq "0") {
2333 printf STDERR_ORIG "[Crunch] $msg\n", @_;
2339 my $python_src = "$install_dir/python";
2340 my $venv_dir = "$job_work/.arvados.venv";
2341 my $venv_built = -e "$venv_dir/bin/activate";
2342 if ((!$venv_built) and (-d $python_src) and can_run("virtualenv")) {
2343 shell_or_die(undef, "virtualenv", "--quiet", "--system-site-packages",
2344 "--python=python2.7", $venv_dir);
2345 shell_or_die(TASK_TEMPFAIL, "$venv_dir/bin/pip", "--quiet", "install", "-I", $python_src);
2347 $Log->("Built Python SDK virtualenv");
2350 my @pysdk_version_cmd = ("python", "-c",
2351 "from pkg_resources import get_distribution as get; print get('arvados-python-client').version");
2353 $Log->("Running in Python SDK virtualenv");
2354 @pysdk_version_cmd = ();
2355 my $orig_argv = join(" ", map { quotemeta($_); } @ARGV);
2356 @ARGV = ("/bin/sh", "-ec",
2357 ". \Q$venv_dir/bin/activate\E; exec $orig_argv");
2358 } elsif (-d $python_src) {
2359 $Log->("Warning: virtualenv not found inside Docker container default " .
2360 "\$PATH. Can't install Python SDK.");
2363 if (@pysdk_version_cmd) {
2364 open(my $pysdk_version_pipe, "-|", @pysdk_version_cmd);
2365 my $pysdk_version = <$pysdk_version_pipe>;
2366 close($pysdk_version_pipe);
2368 chomp($pysdk_version);
2369 $Log->("Using Arvados SDK version $pysdk_version");
2371 # A lot could've gone wrong here, but pretty much all of it means that
2372 # Python won't be able to load the Arvados SDK.
2373 $Log->("Warning: Arvados SDK not found");
2377 while (my ($sdk_dir, $sdk_envkey) = each(%SDK_ENVVARS)) {
2378 my $sdk_path = "$install_dir/$sdk_dir";
2380 if ($ENV{$sdk_envkey}) {
2381 $ENV{$sdk_envkey} = "$sdk_path:" . $ENV{$sdk_envkey};
2383 $ENV{$sdk_envkey} = $sdk_path;
2385 $Log->("Arvados SDK added to %s", $sdk_envkey);
2390 die "Cannot exec `@ARGV`: $!";
2393 ### Installation mode
2394 open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
2396 if (readlink ("$destdir.archive_hash") eq $archive_hash && -d $destdir) {
2397 # This exact git archive (source + arvados sdk) is already installed
2398 # here, so there's no need to reinstall it.
2400 # We must consume our DATA section, though: otherwise the process
2401 # feeding it to us will get SIGPIPE.
2403 while (read(DATA, $buf, 65536)) { }
2408 unlink "$destdir.archive_hash";
2412 # Ignore SIGPIPE: we check retval of close() instead. See perlipc(1).
2413 local $SIG{PIPE} = "IGNORE";
2414 warn "Extracting archive: $archive_hash\n";
2415 # --ignore-zeros is necessary sometimes: depending on how much NUL
2416 # padding tar -A put on our combined archive (which in turn depends
2417 # on the length of the component archives) tar without
2418 # --ignore-zeros will exit before consuming stdin and cause close()
2419 # to fail on the resulting SIGPIPE.
2420 if (!open(TARX, "|-", "tar", "--ignore-zeros", "-xC", $destdir)) {
2421 die "Error launching 'tar -xC $destdir': $!";
2423 # If we send too much data to tar in one write (> 4-5 MiB), it stops, and we
2424 # get SIGPIPE. We must feed it data incrementally.
2426 while (read(DATA, $tar_input, 65536)) {
2427 print TARX $tar_input;
2430 die "'tar -xC $destdir' exited $?: $!";
2436 my $sdk_root = "$destdir/.arvados.sdk/sdk";
2438 foreach my $sdk_lang (("python",
2439 map { (split /\//, $_, 2)[0]; } keys(%SDK_ENVVARS))) {
2440 if (-d "$sdk_root/$sdk_lang") {
2441 if (!rename("$sdk_root/$sdk_lang", "$install_dir/$sdk_lang")) {
2442 die "Failed to install $sdk_lang SDK: $!";
2448 my $python_dir = "$install_dir/python";
2449 if ((-d $python_dir) and can_run("python2.7")) {
2450 open(my $egg_info_pipe, "-|",
2451 "python2.7 \Q$python_dir/setup.py\E egg_info 2>&1 >/dev/null");
2452 my @egg_info_errors = <$egg_info_pipe>;
2453 close($egg_info_pipe);
2456 if (@egg_info_errors and (($egg_info_errors[-1] =~ /\bgit\b/) or ($egg_info_errors[-1] =~ /\[Errno 2\]/))) {
2457 # egg_info apparently failed because it couldn't ask git for a build tag.
2458 # Specify no build tag.
2459 open(my $pysdk_cfg, ">>", "$python_dir/setup.cfg");
2460 print $pysdk_cfg "\n[egg_info]\ntag_build =\n";
2463 my $egg_info_exit = $? >> 8;
2464 foreach my $errline (@egg_info_errors) {
2467 warn "python setup.py egg_info failed: exit $egg_info_exit";
2468 exit ($egg_info_exit || 1);
2473 # Hide messages from the install script (unless it fails: shell_or_die
2474 # will show $destdir.log in that case).
2475 open(STDOUT, ">>", "$destdir.log");
2476 open(STDERR, ">&", STDOUT);
2478 if (-e "$destdir/crunch_scripts/install") {
2479 shell_or_die (undef, "$destdir/crunch_scripts/install", $install_dir);
2480 } elsif (!-e "./install.sh" && -e "./tests/autotests.sh") {
2482 shell_or_die (undef, "./tests/autotests.sh", $install_dir);
2483 } elsif (-e "./install.sh") {
2484 shell_or_die (undef, "./install.sh", $install_dir);
2487 if ($archive_hash) {
2488 unlink "$destdir.archive_hash.new";
2489 symlink ($archive_hash, "$destdir.archive_hash.new") or die "$destdir.archive_hash.new: $!";
2490 rename ("$destdir.archive_hash.new", "$destdir.archive_hash") or die "$destdir.archive_hash: $!";
2496 my $command_name = shift;
2497 open(my $which, "-|", "which", $command_name);
2498 while (<$which>) { }
2505 my $exitcode = shift;
2507 if ($ENV{"DEBUG"}) {
2508 print STDERR "@_\n";
2510 if (system (@_) != 0) {
2513 my $exitstatus = sprintf("exit %d signal %d", $code >> 8, $code & 0x7f);
2514 open STDERR, ">&STDERR_ORIG";
2515 system ("cat $destdir.log >&2");
2516 warn "@_ failed ($err): $exitstatus";
2517 if (defined($exitcode)) {
2521 exit (($code >> 8) || 1);