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 $docker_bin = "/usr/bin/docker.io";
130 GetOptions('force-unlock' => \$force_unlock,
131 'git-dir=s' => \$git_dir,
132 'job=s' => \$jobspec,
133 'job-api-token=s' => \$job_api_token,
134 'no-clear-tmp' => \$no_clear_tmp,
135 'resume-stash=s' => \$resume_stash,
136 'docker-bin=s' => \$docker_bin,
139 if (defined $job_api_token) {
140 $ENV{ARVADOS_API_TOKEN} = $job_api_token;
143 my $have_slurm = exists $ENV{SLURM_JOB_ID} && exists $ENV{SLURM_NODELIST};
148 $main::ENV{CRUNCH_DEBUG} = 1;
152 $main::ENV{CRUNCH_DEBUG} = 0;
155 my $arv = Arvados->new('apiVersion' => 'v1');
164 if ($jobspec =~ /^[-a-z\d]+$/)
166 # $jobspec is an Arvados UUID, not a JSON job specification
167 $Job = api_call("jobs/get", uuid => $jobspec);
172 $Job = JSON::decode_json($jobspec);
177 # Make sure our workers (our slurm nodes, localhost, or whatever) are
178 # at least able to run basic commands: they aren't down or severely
181 if ($Job->{docker_image_locator}) {
182 $cmd = [$docker_bin, 'ps', '-q'];
184 Log(undef, "Sanity check is `@$cmd`");
185 srun(["srun", "--nodes=\Q$ENV{SLURM_NNODES}\E", "--ntasks-per-node=1"],
189 Log(undef, "Sanity check failed: ".exit_status_s($?));
192 Log(undef, "Sanity check OK");
195 my $User = api_call("users/current");
198 if (!$force_unlock) {
199 # Claim this job, and make sure nobody else does
200 eval { api_call("jobs/lock", uuid => $Job->{uuid}); };
202 Log(undef, "Error while locking job, exiting ".EX_TEMPFAIL);
211 map { croak ("No $_ specified") unless $Job->{$_} }
212 qw(script script_version script_parameters);
215 $Job->{'is_locked_by_uuid'} = $User->{'uuid'};
216 $Job->{'started_at'} = gmtime;
217 $Job->{'state'} = 'Running';
219 $Job = api_call("jobs/create", job => $Job);
221 $job_id = $Job->{'uuid'};
223 my $keep_logfile = $job_id . '.log.txt';
224 log_writer_start($keep_logfile);
226 $Job->{'runtime_constraints'} ||= {};
227 $Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0;
228 my $max_ncpus = $Job->{'runtime_constraints'}->{'max_tasks_per_node'};
230 my $gem_versions = `gem list --quiet arvados-cli 2>/dev/null`;
232 $gem_versions =~ s/^arvados-cli \(/ with arvados-cli Gem version(s) /;
233 chomp($gem_versions);
234 chop($gem_versions); # Closing parentheses
239 "running from " . ((-e $0) ? realpath($0) : "stdin") . $gem_versions);
241 Log (undef, "check slurm allocation");
244 # Should use $ENV{SLURM_TASKS_PER_NODE} instead of sinfo? (eg. "4(x3),2,4(x2)")
248 my $localcpus = 0 + `grep -cw ^processor /proc/cpuinfo` || 1;
249 push @sinfo, "$localcpus localhost";
251 if (exists $ENV{SLURM_NODELIST})
253 push @sinfo, `sinfo -h --format='%c %N' --nodes=\Q$ENV{SLURM_NODELIST}\E`;
257 my ($ncpus, $slurm_nodelist) = split;
258 $ncpus = $max_ncpus if $max_ncpus && $ncpus > $max_ncpus;
261 while ($slurm_nodelist =~ s/^([^\[,]+?(\[.*?\])?)(,|$)//)
264 if ($nodelist =~ /\[((\d+)(-(\d+))?(,(\d+)(-(\d+))?)*)\]/)
267 foreach (split (",", $ranges))
280 push @nodelist, map {
282 $n =~ s/\[[-,\d]+\]/$_/;
289 push @nodelist, $nodelist;
292 foreach my $nodename (@nodelist)
294 Log (undef, "node $nodename - $ncpus slots");
295 my $node = { name => $nodename,
297 # The number of consecutive times a task has been dispatched
298 # to this node and failed.
300 # The number of consecutive times that SLURM has reported
301 # a node failure since the last successful task.
303 # Don't dispatch work to this node until this time
304 # (in seconds since the epoch) has passed.
306 foreach my $cpu (1..$ncpus)
308 push @slot, { node => $node,
312 push @node, @nodelist;
317 # Ensure that we get one jobstep running on each allocated node before
318 # we start overloading nodes with concurrent steps
320 @slot = sort { $a->{cpu} <=> $b->{cpu} } @slot;
323 $Job->update_attributes(
324 'tasks_summary' => { 'failed' => 0,
329 Log (undef, "start");
330 $SIG{'INT'} = sub { $main::please_freeze = 1; };
331 $SIG{'QUIT'} = sub { $main::please_freeze = 1; };
332 $SIG{'TERM'} = \&croak;
333 $SIG{'TSTP'} = sub { $main::please_freeze = 1; };
334 $SIG{'ALRM'} = sub { $main::please_info = 1; };
335 $SIG{'CONT'} = sub { $main::please_continue = 1; };
336 $SIG{'HUP'} = sub { $main::please_refresh = 1; };
338 $main::please_freeze = 0;
339 $main::please_info = 0;
340 $main::please_continue = 0;
341 $main::please_refresh = 0;
342 my $jobsteps_must_output_keys = 0; # becomes 1 when any task outputs a key
344 grep { $ENV{$1} = $2 if /^(NOCACHE.*?)=(.*)/ } split ("\n", $$Job{knobs});
345 $ENV{"CRUNCH_JOB_UUID"} = $job_id;
346 $ENV{"JOB_UUID"} = $job_id;
349 my @jobstep_todo = ();
350 my @jobstep_done = ();
351 my @jobstep_tomerge = ();
352 my $jobstep_tomerge_level = 0;
353 my $squeue_checked = 0;
354 my $latest_refresh = scalar time;
358 if (defined $Job->{thawedfromkey})
360 thaw ($Job->{thawedfromkey});
364 my $first_task = api_call("job_tasks/create", job_task => {
365 'job_uuid' => $Job->{'uuid'},
370 push @jobstep, { 'level' => 0,
372 'arvados_task' => $first_task,
374 push @jobstep_todo, 0;
380 must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
383 my $build_script = handle_readall(\*DATA);
384 my $nodelist = join(",", @node);
385 my $git_tar_count = 0;
387 if (!defined $no_clear_tmp) {
388 # Clean out crunch_tmp/work, crunch_tmp/opt, crunch_tmp/src*
389 Log (undef, "Clean work dirs");
391 my $cleanpid = fork();
394 # Find FUSE mounts that look like Keep mounts (the mount path has the
395 # word "keep") and unmount them. Then clean up work directories.
396 # TODO: When #5036 is done and widely deployed, we can get rid of the
397 # regular expression and just unmount everything with type fuse.keep.
398 srun (["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
399 ['bash', '-ec', 'mount -t fuse,fuse.keep | awk \'($3 ~ /\ykeep\y/){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']);
404 last if $cleanpid == waitpid (-1, WNOHANG);
405 freeze_if_want_freeze ($cleanpid);
406 select (undef, undef, undef, 0.1);
408 Log (undef, "Cleanup command exited ".exit_status_s($?));
411 # If this job requires a Docker image, install that.
412 my ($docker_locator, $docker_stream, $docker_hash, $docker_limitmem);
413 if ($docker_locator = $Job->{docker_image_locator}) {
414 ($docker_stream, $docker_hash) = find_docker_image($docker_locator);
417 croak("No Docker image hash found from locator $docker_locator");
419 $docker_stream =~ s/^\.//;
420 my $docker_install_script = qq{
421 if ! $docker_bin images -q --no-trunc --all | grep -qxF \Q$docker_hash\E; then
422 arv-get \Q$docker_locator$docker_stream/$docker_hash.tar\E | $docker_bin load
425 my $docker_pid = fork();
426 if ($docker_pid == 0)
428 srun (["srun", "--nodelist=" . join(',', @node)],
429 ["/bin/sh", "-ec", $docker_install_script]);
434 last if $docker_pid == waitpid (-1, WNOHANG);
435 freeze_if_want_freeze ($docker_pid);
436 select (undef, undef, undef, 0.1);
440 croak("Installing Docker image from $docker_locator exited "
444 # Determine whether this version of Docker supports memory+swap limits.
445 srun(["srun", "--nodelist=" . $node[0]],
446 ["/bin/sh", "-ec", "$docker_bin run --help | grep -qe --memory-swap="],
448 $docker_limitmem = ($? == 0);
450 if ($Job->{arvados_sdk_version}) {
451 # The job also specifies an Arvados SDK version. Add the SDKs to the
452 # tar file for the build script to install.
453 Log(undef, sprintf("Packing Arvados SDK version %s for installation",
454 $Job->{arvados_sdk_version}));
455 add_git_archive("git", "--git-dir=$git_dir", "archive",
456 "--prefix=.arvados.sdk/",
457 $Job->{arvados_sdk_version}, "sdk");
461 if (!defined $git_dir && $Job->{'script_version'} =~ m{^/}) {
462 # If script_version looks like an absolute path, *and* the --git-dir
463 # argument was not given -- which implies we were not invoked by
464 # crunch-dispatch -- we will use the given path as a working
465 # directory instead of resolving script_version to a git commit (or
466 # doing anything else with git).
467 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{'script_version'};
468 $ENV{"CRUNCH_SRC"} = $Job->{'script_version'};
471 # Resolve the given script_version to a git commit sha1. Also, if
472 # the repository is remote, clone it into our local filesystem: this
473 # ensures "git archive" will work, and is necessary to reliably
474 # resolve a symbolic script_version like "master^".
475 $ENV{"CRUNCH_SRC"} = "$ENV{CRUNCH_TMP}/src";
477 Log (undef, "Looking for version ".$Job->{script_version}." from repository ".$Job->{repository});
479 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{script_version};
481 # If we're running under crunch-dispatch, it will have already
482 # pulled the appropriate source tree into its own repository, and
483 # given us that repo's path as $git_dir.
485 # If we're running a "local" job, we might have to fetch content
486 # from a remote repository.
488 # (Currently crunch-dispatch gives a local path with --git-dir, but
489 # we might as well accept URLs there too in case it changes its
491 my $repo = $git_dir || $Job->{'repository'};
493 # Repository can be remote or local. If remote, we'll need to fetch it
494 # to a local dir before doing `git log` et al.
497 if ($repo =~ m{://|^[^/]*:}) {
498 # $repo is a git url we can clone, like git:// or https:// or
499 # file:/// or [user@]host:repo.git. Note "user/name@host:foo" is
500 # not recognized here because distinguishing that from a local
501 # path is too fragile. If you really need something strange here,
502 # use the ssh:// form.
503 $repo_location = 'remote';
504 } elsif ($repo =~ m{^\.*/}) {
505 # $repo is a local path to a git index. We'll also resolve ../foo
506 # to ../foo/.git if the latter is a directory. To help
507 # disambiguate local paths from named hosted repositories, this
508 # form must be given as ./ or ../ if it's a relative path.
509 if (-d "$repo/.git") {
510 $repo = "$repo/.git";
512 $repo_location = 'local';
514 # $repo is none of the above. It must be the name of a hosted
516 my $arv_repo_list = api_call("repositories/list",
517 'filters' => [['name','=',$repo]]);
518 my @repos_found = @{$arv_repo_list->{'items'}};
519 my $n_found = $arv_repo_list->{'serverResponse'}->{'items_available'};
521 Log(undef, "Repository '$repo' -> "
522 . join(", ", map { $_->{'uuid'} } @repos_found));
525 croak("Error: Found $n_found repositories with name '$repo'.");
527 $repo = $repos_found[0]->{'fetch_url'};
528 $repo_location = 'remote';
530 Log(undef, "Using $repo_location repository '$repo'");
531 $ENV{"CRUNCH_SRC_URL"} = $repo;
533 # Resolve given script_version (we'll call that $treeish here) to a
534 # commit sha1 ($commit).
535 my $treeish = $Job->{'script_version'};
537 if ($repo_location eq 'remote') {
538 # We minimize excess object-fetching by re-using the same bare
539 # repository in CRUNCH_TMP/.git for multiple crunch-jobs -- we
540 # just keep adding remotes to it as needed.
541 my $local_repo = $ENV{'CRUNCH_TMP'}."/.git";
542 my $gitcmd = "git --git-dir=\Q$local_repo\E";
544 # Set up our local repo for caching remote objects, making
546 if (!-d $local_repo) {
547 make_path($local_repo) or croak("Error: could not create $local_repo");
549 # This works (exits 0 and doesn't delete fetched objects) even
550 # if $local_repo is already initialized:
551 `$gitcmd init --bare`;
553 croak("Error: $gitcmd init --bare exited ".exit_status_s($?));
556 # If $treeish looks like a hash (or abbrev hash) we look it up in
557 # our local cache first, since that's cheaper. (We don't want to
558 # do that with tags/branches though -- those change over time, so
559 # they should always be resolved by the remote repo.)
560 if ($treeish =~ /^[0-9a-f]{7,40}$/s) {
561 # Hide stderr because it's normal for this to fail:
562 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E 2>/dev/null`;
564 # Careful not to resolve a branch named abcdeff to commit 1234567:
565 $sha1 =~ /^$treeish/ &&
566 $sha1 =~ /^([0-9a-f]{40})$/s) {
568 Log(undef, "Commit $commit already present in $local_repo");
572 if (!defined $commit) {
573 # If $treeish isn't just a hash or abbrev hash, or isn't here
574 # yet, we need to fetch the remote to resolve it correctly.
576 # First, remove all local heads. This prevents a name that does
577 # not exist on the remote from resolving to (or colliding with)
578 # a previously fetched branch or tag (possibly from a different
580 remove_tree("$local_repo/refs/heads", {keep_root => 1});
582 Log(undef, "Fetching objects from $repo to $local_repo");
583 `$gitcmd fetch --no-progress --tags ''\Q$repo\E \Q+refs/heads/*:refs/heads/*\E`;
585 croak("Error: `$gitcmd fetch` exited ".exit_status_s($?));
589 # Now that the data is all here, we will use our local repo for
590 # the rest of our git activities.
594 my $gitcmd = "git --git-dir=\Q$repo\E";
595 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E`;
596 unless ($? == 0 && $sha1 =~ /^([0-9a-f]{40})$/) {
597 croak("`$gitcmd rev-list` exited "
599 .", '$treeish' not found. Giving up.");
602 Log(undef, "Version $treeish is commit $commit");
604 if ($commit ne $Job->{'script_version'}) {
605 # Record the real commit id in the database, frozentokey, logs,
606 # etc. -- instead of an abbreviation or a branch name which can
607 # become ambiguous or point to a different commit in the future.
608 if (!$Job->update_attributes('script_version' => $commit)) {
609 croak("Error: failed to update job's script_version attribute");
613 $ENV{"CRUNCH_SRC_COMMIT"} = $commit;
614 add_git_archive("$gitcmd archive ''\Q$commit\E");
617 my $git_archive = combined_git_archive();
618 if (!defined $git_archive) {
619 Log(undef, "Skip install phase (no git archive)");
621 Log(undef, "Warning: This probably means workers have no source tree!");
626 my $install_script_tries_left = 3;
627 for (my $attempts = 0; $attempts < 3; $attempts++) {
628 Log(undef, "Run install script on all workers");
630 my @srunargs = ("srun",
631 "--nodelist=$nodelist",
632 "-D", $ENV{'TMPDIR'}, "--job-name=$job_id");
633 my @execargs = ("sh", "-c",
634 "mkdir -p $ENV{CRUNCH_INSTALL} && cd $ENV{CRUNCH_TMP} && perl -");
636 $ENV{"CRUNCH_GIT_ARCHIVE_HASH"} = md5_hex($git_archive);
637 my ($install_stderr_r, $install_stderr_w);
638 pipe $install_stderr_r, $install_stderr_w or croak("pipe() failed: $!");
639 set_nonblocking($install_stderr_r);
640 my $installpid = fork();
641 if ($installpid == 0)
643 close($install_stderr_r);
644 fcntl($install_stderr_w, F_SETFL, 0) or croak($!); # no close-on-exec
645 open(STDOUT, ">&", $install_stderr_w);
646 open(STDERR, ">&", $install_stderr_w);
647 srun (\@srunargs, \@execargs, {}, $build_script . $git_archive);
650 close($install_stderr_w);
651 # Tell freeze_if_want_freeze how to kill the child, otherwise the
652 # "waitpid(installpid)" loop won't get interrupted by a freeze:
653 $proc{$installpid} = {};
655 # Track whether anything appears on stderr other than slurm errors
656 # ("srun: ...") and the "starting: ..." message printed by the
657 # srun subroutine itself:
658 my $stderr_anything_from_script = 0;
659 my $match_our_own_errors = '^(srun: error: |starting: \[)';
660 while ($installpid != waitpid(-1, WNOHANG)) {
661 freeze_if_want_freeze ($installpid);
662 # Wait up to 0.1 seconds for something to appear on stderr, then
663 # do a non-blocking read.
664 my $bits = fhbits($install_stderr_r);
665 select ($bits, undef, $bits, 0.1);
666 if (0 < sysread ($install_stderr_r, $stderr_buf, 8192, length($stderr_buf)))
668 while ($stderr_buf =~ /^(.*?)\n/) {
670 substr $stderr_buf, 0, 1+length($line), "";
671 Log(undef, "stderr $line");
672 if ($line !~ /$match_our_own_errors/) {
673 $stderr_anything_from_script = 1;
678 delete $proc{$installpid};
679 $install_exited = $?;
680 close($install_stderr_r);
681 if (length($stderr_buf) > 0) {
682 if ($stderr_buf !~ /$match_our_own_errors/) {
683 $stderr_anything_from_script = 1;
685 Log(undef, "stderr $stderr_buf")
688 Log (undef, "Install script exited ".exit_status_s($install_exited));
689 last if $install_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);
705 if ($install_exited != 0) {
710 foreach (qw (script script_version script_parameters runtime_constraints))
714 (ref($Job->{$_}) ? JSON::encode_json($Job->{$_}) : $Job->{$_}));
716 foreach (split (/\n/, $Job->{knobs}))
718 Log (undef, "knob " . $_);
723 $main::success = undef;
729 my $thisround_succeeded = 0;
730 my $thisround_failed = 0;
731 my $thisround_failed_multiple = 0;
732 my $working_slot_count = scalar(@slot);
734 @jobstep_todo = sort { $jobstep[$a]->{level} <=> $jobstep[$b]->{level}
735 or $a <=> $b } @jobstep_todo;
736 my $level = $jobstep[$jobstep_todo[0]]->{level};
738 my $initial_tasks_this_level = 0;
739 foreach my $id (@jobstep_todo) {
740 $initial_tasks_this_level++ if ($jobstep[$id]->{level} == $level);
743 # If the number of tasks scheduled at this level #T is smaller than the number
744 # of slots available #S, only use the first #T slots, or the first slot on
745 # each node, whichever number is greater.
747 # When we dispatch tasks later, we'll allocate whole-node resources like RAM
748 # based on these numbers. Using fewer slots makes more resources available
749 # to each individual task, which should normally be a better strategy when
750 # there are fewer of them running with less parallelism.
752 # Note that this calculation is not redone if the initial tasks at
753 # this level queue more tasks at the same level. This may harm
754 # overall task throughput for that level.
756 if ($initial_tasks_this_level < @node) {
757 @freeslot = (0..$#node);
758 } elsif ($initial_tasks_this_level < @slot) {
759 @freeslot = (0..$initial_tasks_this_level - 1);
761 @freeslot = (0..$#slot);
763 my $round_num_freeslots = scalar(@freeslot);
765 my %round_max_slots = ();
766 for (my $ii = $#freeslot; $ii >= 0; $ii--) {
767 my $this_slot = $slot[$freeslot[$ii]];
768 my $node_name = $this_slot->{node}->{name};
769 $round_max_slots{$node_name} ||= $this_slot->{cpu};
770 last if (scalar(keys(%round_max_slots)) >= @node);
773 Log(undef, "start level $level with $round_num_freeslots slots");
776 my $progress_is_dirty = 1;
777 my $progress_stats_updated = 0;
779 update_progress_stats();
783 for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
785 my $id = $jobstep_todo[$todo_ptr];
786 my $Jobstep = $jobstep[$id];
787 if ($Jobstep->{level} != $level)
792 pipe $reader{$id}, "writer" or croak("pipe() failed: $!");
793 set_nonblocking($reader{$id});
795 my $childslot = $freeslot[0];
796 my $childnode = $slot[$childslot]->{node};
797 my $childslotname = join (".",
798 $slot[$childslot]->{node}->{name},
799 $slot[$childslot]->{cpu});
801 my $childpid = fork();
804 $SIG{'INT'} = 'DEFAULT';
805 $SIG{'QUIT'} = 'DEFAULT';
806 $SIG{'TERM'} = 'DEFAULT';
808 foreach (values (%reader))
812 fcntl ("writer", F_SETFL, 0) or croak ($!); # no close-on-exec
813 open(STDOUT,">&writer");
814 open(STDERR,">&writer");
819 delete $ENV{"GNUPGHOME"};
820 $ENV{"TASK_UUID"} = $Jobstep->{'arvados_task'}->{'uuid'};
821 $ENV{"TASK_QSEQUENCE"} = $id;
822 $ENV{"TASK_SEQUENCE"} = $level;
823 $ENV{"JOB_SCRIPT"} = $Job->{script};
824 while (my ($param, $value) = each %{$Job->{script_parameters}}) {
825 $param =~ tr/a-z/A-Z/;
826 $ENV{"JOB_PARAMETER_$param"} = $value;
828 $ENV{"TASK_SLOT_NODE"} = $slot[$childslot]->{node}->{name};
829 $ENV{"TASK_SLOT_NUMBER"} = $slot[$childslot]->{cpu};
830 $ENV{"TASK_WORK"} = $ENV{"CRUNCH_TMP"}."/task/$childslotname";
831 $ENV{"HOME"} = $ENV{"TASK_WORK"};
832 $ENV{"TASK_KEEPMOUNT"} = $ENV{"TASK_WORK"}.".keep";
833 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
834 $ENV{"CRUNCH_NODE_SLOTS"} = $round_max_slots{$ENV{TASK_SLOT_NODE}};
835 $ENV{"PATH"} = $ENV{"CRUNCH_INSTALL"} . "/bin:" . $ENV{"PATH"};
841 "--nodelist=".$childnode->{name},
842 qw(-n1 -c1 -N1 -D), $ENV{'TMPDIR'},
843 "--job-name=$job_id.$id.$$",
846 "if [ -e $ENV{TASK_WORK} ]; then rm -rf $ENV{TASK_WORK}; fi; "
847 ."mkdir -p $ENV{CRUNCH_TMP} $ENV{JOB_WORK} $ENV{TASK_WORK} $ENV{TASK_KEEPMOUNT} "
848 ."&& cd $ENV{CRUNCH_TMP} "
849 # These environment variables get used explicitly later in
850 # $command. No tool is expected to read these values directly.
851 .q{&& MEM=$(awk '($1 == "MemTotal:"){print $2}' </proc/meminfo) }
852 .q{&& SWAP=$(awk '($1 == "SwapTotal:"){print $2}' </proc/meminfo) }
853 ."&& MEMLIMIT=\$(( (\$MEM * 95) / ($ENV{CRUNCH_NODE_SLOTS} * 100) )) "
854 ."&& let SWAPLIMIT=\$MEMLIMIT+\$SWAP ";
855 $command .= "&& exec arv-mount --by-id --allow-other $ENV{TASK_KEEPMOUNT} --exec ";
858 my $cidfile = "$ENV{CRUNCH_TMP}/$Jobstep->{arvados_task}->{uuid}-$Jobstep->{failures}.cid";
859 $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -cgroup-parent=docker -cgroup-cid=$cidfile -poll=10000 ";
860 $command .= "$docker_bin run --rm=true --attach=stdout --attach=stderr --attach=stdin -i --user=crunch --cidfile=$cidfile --sig-proxy ";
861 # We only set memory limits if Docker lets us limit both memory and swap.
862 # Memory limits alone have been supported longer, but subprocesses tend
863 # to get SIGKILL if they exceed that without any swap limit set.
864 # See #5642 for additional background.
865 if ($docker_limitmem) {
866 $command .= "--memory=\${MEMLIMIT}k --memory-swap=\${SWAPLIMIT}k ";
869 # Dynamically configure the container to use the host system as its
870 # DNS server. Get the host's global addresses from the ip command,
871 # and turn them into docker --dns options using gawk.
873 q{$(ip -o address show scope global |
874 gawk 'match($4, /^([0-9\.:]+)\//, x){print "--dns", x[1]}') };
876 # The source tree and $destdir directory (which we have
877 # installed on the worker host) are available in the container,
878 # under the same path.
879 $command .= "--volume=\Q$ENV{CRUNCH_SRC}:$ENV{CRUNCH_SRC}:ro\E ";
880 $command .= "--volume=\Q$ENV{CRUNCH_INSTALL}:$ENV{CRUNCH_INSTALL}:ro\E ";
882 # Currently, we make arv-mount's mount point appear at /keep
883 # inside the container (instead of using the same path as the
884 # host like we do with CRUNCH_SRC and CRUNCH_INSTALL). However,
885 # crunch scripts and utilities must not rely on this. They must
886 # use $TASK_KEEPMOUNT.
887 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT}:/keep:ro\E ";
888 $ENV{TASK_KEEPMOUNT} = "/keep";
890 # TASK_WORK is almost exactly like a docker data volume: it
891 # starts out empty, is writable, and persists until no
892 # containers use it any more. We don't use --volumes-from to
893 # share it with other containers: it is only accessible to this
894 # task, and it goes away when this task stops.
896 # However, a docker data volume is writable only by root unless
897 # the mount point already happens to exist in the container with
898 # different permissions. Therefore, we [1] assume /tmp already
899 # exists in the image and is writable by the crunch user; [2]
900 # avoid putting TASK_WORK inside CRUNCH_TMP (which won't be
901 # writable if they are created by docker while setting up the
902 # other --volumes); and [3] create $TASK_WORK inside the
903 # container using $build_script.
904 $command .= "--volume=/tmp ";
905 $ENV{"TASK_WORK"} = "/tmp/crunch-job-task-work/$childslotname";
906 $ENV{"HOME"} = $ENV{"TASK_WORK"};
907 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
909 # TODO: Share a single JOB_WORK volume across all task
910 # containers on a given worker node, and delete it when the job
911 # ends (and, in case that doesn't work, when the next job
914 # For now, use the same approach as TASK_WORK above.
915 $ENV{"JOB_WORK"} = "/tmp/crunch-job-work";
917 while (my ($env_key, $env_val) = each %ENV)
919 if ($env_key =~ /^(ARVADOS|CRUNCH|JOB|TASK)_/) {
920 $command .= "--env=\Q$env_key=$env_val\E ";
923 $command .= "--env=\QHOME=$ENV{HOME}\E ";
924 $command .= "\Q$docker_hash\E ";
925 $command .= "stdbuf --output=0 --error=0 ";
926 $command .= "perl - $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
929 $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -poll=10000 ";
930 $command .= "stdbuf --output=0 --error=0 ";
931 $command .= "perl - $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
934 my @execargs = ('bash', '-c', $command);
935 srun (\@srunargs, \@execargs, undef, $build_script);
936 # exec() failed, we assume nothing happened.
937 die "srun() failed on build script\n";
940 if (!defined $childpid)
947 $proc{$childpid} = { jobstep => $id,
950 jobstepname => "$job_id.$id.$childpid",
952 croak ("assert failed: \$slot[$childslot]->{'pid'} exists") if exists $slot[$childslot]->{pid};
953 $slot[$childslot]->{pid} = $childpid;
955 Log ($id, "job_task ".$Jobstep->{'arvados_task'}->{'uuid'});
956 Log ($id, "child $childpid started on $childslotname");
957 $Jobstep->{starttime} = time;
958 $Jobstep->{node} = $childnode->{name};
959 $Jobstep->{slotindex} = $childslot;
960 delete $Jobstep->{stderr};
961 delete $Jobstep->{finishtime};
962 delete $Jobstep->{tempfail};
964 $Jobstep->{'arvados_task'}->{started_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{starttime});
965 $Jobstep->{'arvados_task'}->save;
967 splice @jobstep_todo, $todo_ptr, 1;
970 $progress_is_dirty = 1;
974 ($round_num_freeslots > @freeslot && $todo_ptr+1 > $#jobstep_todo))
976 last THISROUND if $main::please_freeze || defined($main::success);
977 if ($main::please_info)
979 $main::please_info = 0;
981 create_output_collection();
983 update_progress_stats();
990 check_refresh_wanted();
992 update_progress_stats();
993 select (undef, undef, undef, 0.1);
995 elsif (time - $progress_stats_updated >= 30)
997 update_progress_stats();
999 $working_slot_count = scalar(grep { $_->{node}->{fail_count} == 0 &&
1000 $_->{node}->{hold_count} < 4 } @slot);
1001 if (($thisround_failed_multiple >= 8 && $thisround_succeeded == 0) ||
1002 ($thisround_failed_multiple >= 16 && $thisround_failed_multiple > $thisround_succeeded))
1004 my $message = "Repeated failure rate too high ($thisround_failed_multiple/"
1005 .($thisround_failed+$thisround_succeeded)
1006 .") -- giving up on this round";
1007 Log (undef, $message);
1011 # move slots from freeslot to holdslot (or back to freeslot) if necessary
1012 for (my $i=$#freeslot; $i>=0; $i--) {
1013 if ($slot[$freeslot[$i]]->{node}->{hold_until} > scalar time) {
1014 push @holdslot, (splice @freeslot, $i, 1);
1017 for (my $i=$#holdslot; $i>=0; $i--) {
1018 if ($slot[$holdslot[$i]]->{node}->{hold_until} <= scalar time) {
1019 push @freeslot, (splice @holdslot, $i, 1);
1023 # give up if no nodes are succeeding
1024 if ($working_slot_count < 1) {
1025 Log(undef, "Every node has failed -- giving up");
1032 push @freeslot, splice @holdslot;
1033 map { $slot[$freeslot[$_]]->{node}->{losing_streak} = 0 } (0..$#freeslot);
1036 Log (undef, "wait for last ".(scalar keys %proc)." children to finish");
1039 if ($main::please_continue) {
1040 $main::please_continue = 0;
1043 $main::please_info = 0, freeze(), create_output_collection(), save_meta(1) if $main::please_info;
1045 if (!reapchildren())
1047 check_refresh_wanted();
1049 update_progress_stats();
1050 select (undef, undef, undef, 0.1);
1051 killem (keys %proc) if $main::please_freeze;
1055 update_progress_stats();
1056 freeze_if_want_freeze();
1059 if (!defined $main::success)
1061 if (!@jobstep_todo) {
1063 } elsif ($working_slot_count < 1) {
1064 save_output_collection();
1066 exit(EX_RETRY_UNLOCKED);
1067 } elsif ($thisround_succeeded == 0 &&
1068 ($thisround_failed == 0 || $thisround_failed > 4)) {
1069 my $message = "stop because $thisround_failed tasks failed and none succeeded";
1070 Log (undef, $message);
1075 goto ONELEVEL if !defined $main::success;
1078 release_allocation();
1080 my $collated_output = save_output_collection();
1081 Log (undef, "finish");
1086 if ($collated_output && $main::success) {
1087 $final_state = 'Complete';
1089 $final_state = 'Failed';
1091 $Job->update_attributes('state' => $final_state);
1093 exit (($final_state eq 'Complete') ? 0 : 1);
1097 sub update_progress_stats
1099 $progress_stats_updated = time;
1100 return if !$progress_is_dirty;
1101 my ($todo, $done, $running) = (scalar @jobstep_todo,
1102 scalar @jobstep_done,
1103 scalar @slot - scalar @freeslot - scalar @holdslot);
1104 $Job->{'tasks_summary'} ||= {};
1105 $Job->{'tasks_summary'}->{'todo'} = $todo;
1106 $Job->{'tasks_summary'}->{'done'} = $done;
1107 $Job->{'tasks_summary'}->{'running'} = $running;
1108 $Job->update_attributes('tasks_summary' => $Job->{'tasks_summary'});
1109 Log (undef, "status: $done done, $running running, $todo todo");
1110 $progress_is_dirty = 0;
1117 my $pid = waitpid (-1, WNOHANG);
1118 return 0 if $pid <= 0;
1120 my $whatslot = ($slot[$proc{$pid}->{slot}]->{node}->{name}
1122 . $slot[$proc{$pid}->{slot}]->{cpu});
1123 my $jobstepid = $proc{$pid}->{jobstep};
1124 my $elapsed = time - $proc{$pid}->{time};
1125 my $Jobstep = $jobstep[$jobstepid];
1127 my $childstatus = $?;
1128 my $exitvalue = $childstatus >> 8;
1129 my $exitinfo = "exit ".exit_status_s($childstatus);
1130 $Jobstep->{'arvados_task'}->reload;
1131 my $task_success = $Jobstep->{'arvados_task'}->{success};
1133 Log ($jobstepid, "child $pid on $whatslot $exitinfo success=$task_success");
1135 if (!defined $task_success) {
1136 # task did not indicate one way or the other --> fail
1137 $Jobstep->{'arvados_task'}->{success} = 0;
1138 $Jobstep->{'arvados_task'}->save;
1145 $temporary_fail ||= $Jobstep->{tempfail};
1146 $temporary_fail ||= ($exitvalue == TASK_TEMPFAIL);
1148 ++$thisround_failed;
1149 ++$thisround_failed_multiple if $Jobstep->{'failures'} >= 1;
1151 # Check for signs of a failed or misconfigured node
1152 if (++$slot[$proc{$pid}->{slot}]->{node}->{losing_streak} >=
1153 2+$slot[$proc{$pid}->{slot}]->{node}->{ncpus}) {
1154 # Don't count this against jobstep failure thresholds if this
1155 # node is already suspected faulty and srun exited quickly
1156 if ($slot[$proc{$pid}->{slot}]->{node}->{hold_until} &&
1158 Log ($jobstepid, "blaming failure on suspect node " .
1159 $slot[$proc{$pid}->{slot}]->{node}->{name});
1160 $temporary_fail ||= 1;
1162 ban_node_by_slot($proc{$pid}->{slot});
1165 Log ($jobstepid, sprintf('failure (#%d, %s) after %d seconds',
1166 ++$Jobstep->{'failures'},
1167 $temporary_fail ? 'temporary' : 'permanent',
1170 if (!$temporary_fail || $Jobstep->{'failures'} >= 3) {
1171 # Give up on this task, and the whole job
1174 # Put this task back on the todo queue
1175 push @jobstep_todo, $jobstepid;
1176 $Job->{'tasks_summary'}->{'failed'}++;
1180 ++$thisround_succeeded;
1181 $slot[$proc{$pid}->{slot}]->{node}->{losing_streak} = 0;
1182 $slot[$proc{$pid}->{slot}]->{node}->{hold_until} = 0;
1183 $slot[$proc{$pid}->{slot}]->{node}->{fail_count} = 0;
1184 push @jobstep_done, $jobstepid;
1185 Log ($jobstepid, "success in $elapsed seconds");
1187 $Jobstep->{exitcode} = $childstatus;
1188 $Jobstep->{finishtime} = time;
1189 $Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
1190 $Jobstep->{'arvados_task'}->save;
1191 process_stderr ($jobstepid, $task_success);
1192 Log ($jobstepid, sprintf("task output (%d bytes): %s",
1193 length($Jobstep->{'arvados_task'}->{output}),
1194 $Jobstep->{'arvados_task'}->{output}));
1196 close $reader{$jobstepid};
1197 delete $reader{$jobstepid};
1198 delete $slot[$proc{$pid}->{slot}]->{pid};
1199 push @freeslot, $proc{$pid}->{slot};
1202 if ($task_success) {
1204 my $newtask_list = [];
1205 my $newtask_results;
1207 $newtask_results = api_call(
1210 'created_by_job_task_uuid' => $Jobstep->{'arvados_task'}->{uuid}
1212 'order' => 'qsequence',
1213 'offset' => scalar(@$newtask_list),
1215 push(@$newtask_list, @{$newtask_results->{items}});
1216 } while (@{$newtask_results->{items}});
1217 foreach my $arvados_task (@$newtask_list) {
1219 'level' => $arvados_task->{'sequence'},
1221 'arvados_task' => $arvados_task
1223 push @jobstep, $jobstep;
1224 push @jobstep_todo, $#jobstep;
1228 $progress_is_dirty = 1;
1232 sub check_refresh_wanted
1234 my @stat = stat $ENV{"CRUNCH_REFRESH_TRIGGER"};
1235 if (@stat && $stat[9] > $latest_refresh) {
1236 $latest_refresh = scalar time;
1237 my $Job2 = api_call("jobs/get", uuid => $jobspec);
1238 for my $attr ('cancelled_at',
1239 'cancelled_by_user_uuid',
1240 'cancelled_by_client_uuid',
1242 $Job->{$attr} = $Job2->{$attr};
1244 if ($Job->{'state'} ne "Running") {
1245 if ($Job->{'state'} eq "Cancelled") {
1246 Log (undef, "Job cancelled at " . $Job->{'cancelled_at'} . " by user " . $Job->{'cancelled_by_user_uuid'});
1248 Log (undef, "Job state unexpectedly changed to " . $Job->{'state'});
1251 $main::please_freeze = 1;
1258 my $last_squeue_check = $squeue_checked;
1260 # Do not call `squeue` or check the kill list more than once every
1262 return if $last_squeue_check > time - 15;
1263 $squeue_checked = time;
1265 # Look for children from which we haven't received stderr data since
1266 # the last squeue check. If no such children exist, all procs are
1267 # alive and there's no need to even look at squeue.
1269 # As long as the crunchstat poll interval (10s) is shorter than the
1270 # squeue check interval (15s) this should make the squeue check an
1272 my $silent_procs = 0;
1273 for my $jobstep (values %proc)
1275 if ($jobstep->{stderr_at} < $last_squeue_check)
1280 return if $silent_procs == 0;
1282 # use killem() on procs whose killtime is reached
1283 while (my ($pid, $jobstep) = each %proc)
1285 if (exists $jobstep->{killtime}
1286 && $jobstep->{killtime} <= time
1287 && $jobstep->{stderr_at} < $last_squeue_check)
1290 if ($jobstep->{stderr_at}) {
1291 $sincewhen = " in last " . (time - $jobstep->{stderr_at}) . "s";
1293 Log($jobstep->{jobstep}, "killing orphaned srun process $pid (task not in slurm queue, no stderr received$sincewhen)");
1300 # here is an opportunity to check for mysterious problems with local procs
1304 # Get a list of steps still running. Note: squeue(1) says --steps
1305 # selects a format (which we override anyway) and allows us to
1306 # specify which steps we're interested in (which we don't).
1307 # Importantly, it also changes the meaning of %j from "job name" to
1308 # "step name" and (although this isn't mentioned explicitly in the
1309 # docs) switches from "one line per job" mode to "one line per step"
1310 # mode. Without it, we'd just get a list of one job, instead of a
1312 my @squeue = `squeue --jobs=\Q$ENV{SLURM_JOB_ID}\E --steps --format='%j' --noheader`;
1315 Log(undef, "warning: squeue exit status $? ($!)");
1320 # which of my jobsteps are running, according to squeue?
1322 for my $jobstepname (@squeue)
1324 $ok{$jobstepname} = 1;
1327 # Check for child procs >60s old and not mentioned by squeue.
1328 while (my ($pid, $jobstep) = each %proc)
1330 if ($jobstep->{time} < time - 60
1331 && $jobstep->{jobstepname}
1332 && !exists $ok{$jobstep->{jobstepname}}
1333 && !exists $jobstep->{killtime})
1335 # According to slurm, this task has ended (successfully or not)
1336 # -- but our srun child hasn't exited. First we must wait (30
1337 # seconds) in case this is just a race between communication
1338 # channels. Then, if our srun child process still hasn't
1339 # terminated, we'll conclude some slurm communication
1340 # error/delay has caused the task to die without notifying srun,
1341 # and we'll kill srun ourselves.
1342 $jobstep->{killtime} = time + 30;
1343 Log($jobstep->{jobstep}, "notice: task is not in slurm queue but srun process $pid has not exited");
1349 sub release_allocation
1353 Log (undef, "release job allocation");
1354 system "scancel $ENV{SLURM_JOB_ID}";
1362 foreach my $job (keys %reader)
1365 while (0 < sysread ($reader{$job}, $buf, 8192))
1367 print STDERR $buf if $ENV{CRUNCH_DEBUG};
1368 $jobstep[$job]->{stderr_at} = time;
1369 $jobstep[$job]->{stderr} .= $buf;
1370 preprocess_stderr ($job);
1371 if (length ($jobstep[$job]->{stderr}) > 16384)
1373 substr ($jobstep[$job]->{stderr}, 0, 8192) = "";
1382 sub preprocess_stderr
1386 while ($jobstep[$job]->{stderr} =~ /^(.*?)\n/) {
1388 substr $jobstep[$job]->{stderr}, 0, 1+length($line), "";
1389 Log ($job, "stderr $line");
1390 if ($line =~ /srun: error: (SLURM job $ENV{SLURM_JOB_ID} has expired|Unable to confirm allocation for job $ENV{SLURM_JOB_ID})/) {
1392 $main::please_freeze = 1;
1394 elsif ($line =~ /srun: error: Node failure on/) {
1395 my $job_slot_index = $jobstep[$job]->{slotindex};
1396 $slot[$job_slot_index]->{node}->{fail_count}++;
1397 $jobstep[$job]->{tempfail} = 1;
1398 ban_node_by_slot($job_slot_index);
1400 elsif ($line =~ /srun: error: (Unable to create job step|.*: Communication connection failure)/) {
1401 $jobstep[$job]->{tempfail} = 1;
1402 ban_node_by_slot($jobstep[$job]->{slotindex});
1404 elsif ($line =~ /arvados\.errors\.Keep/) {
1405 $jobstep[$job]->{tempfail} = 1;
1414 my $task_success = shift;
1415 preprocess_stderr ($job);
1418 Log ($job, "stderr $_");
1419 } split ("\n", $jobstep[$job]->{stderr});
1426 if (!open($keep, "-|", "arv-get", "--retries", retry_count(), $hash)) {
1427 Log(undef, "fetch_block run error from arv-get $hash: $!");
1430 my $output_block = "";
1433 my $bytes = sysread($keep, $buf, 1024 * 1024);
1434 if (!defined $bytes) {
1435 Log(undef, "fetch_block read error from arv-get: $!");
1436 $output_block = undef;
1438 } elsif ($bytes == 0) {
1439 # sysread returns 0 at the end of the pipe.
1442 # some bytes were read into buf.
1443 $output_block .= $buf;
1448 Log(undef, "fetch_block arv-get exited " . exit_status_s($?));
1449 $output_block = undef;
1451 return $output_block;
1454 # Create a collection by concatenating the output of all tasks (each
1455 # task's output is either a manifest fragment, a locator for a
1456 # manifest fragment stored in Keep, or nothing at all). Return the
1457 # portable_data_hash of the new collection.
1458 sub create_output_collection
1460 Log (undef, "collate");
1462 my ($child_out, $child_in);
1463 my $pid = open2($child_out, $child_in, 'python', '-c', q{
1466 print (arvados.api("v1").collections().
1467 create(body={"manifest_text": sys.stdin.read()}).
1468 execute(num_retries=int(sys.argv[1]))["portable_data_hash"])
1472 my $manifest_size = 0;
1476 my $output = $_->{'arvados_task'}->{output};
1477 next if (!defined($output));
1479 if ($output =~ /^[0-9a-f]{32}(\+\S+)*$/) {
1480 $next_write = fetch_block($output);
1482 $next_write = $output;
1484 if (defined($next_write)) {
1485 if (!defined(syswrite($child_in, $next_write))) {
1486 # There's been an error writing. Stop the loop.
1487 # We'll log details about the exit code later.
1490 $manifest_size += length($next_write);
1493 my $uuid = $_->{'arvados_task'}->{'uuid'};
1494 Log (undef, "Error retrieving '$output' output by task $task_idx ($uuid)");
1499 Log(undef, "collated output manifest text to send to API server is $manifest_size bytes with access tokens");
1502 my $s = IO::Select->new($child_out);
1503 if ($s->can_read(120)) {
1504 sysread($child_out, $joboutput, 1024 * 1024);
1507 Log(undef, "output collection creation exited " . exit_status_s($?));
1513 Log (undef, "timed out while creating output collection");
1514 foreach my $signal (2, 2, 2, 15, 15, 9) {
1515 kill($signal, $pid);
1516 last if waitpid($pid, WNOHANG) == -1;
1525 # Calls create_output_collection, logs the result, and returns it.
1526 # If that was successful, save that as the output in the job record.
1527 sub save_output_collection {
1528 my $collated_output = create_output_collection();
1530 if (!$collated_output) {
1531 Log(undef, "Failed to write output collection");
1534 Log(undef, "job output $collated_output");
1535 $Job->update_attributes('output' => $collated_output);
1537 return $collated_output;
1544 my $sig = 2; # SIGINT first
1545 if (exists $proc{$_}->{"sent_$sig"} &&
1546 time - $proc{$_}->{"sent_$sig"} > 4)
1548 $sig = 15; # SIGTERM if SIGINT doesn't work
1550 if (exists $proc{$_}->{"sent_$sig"} &&
1551 time - $proc{$_}->{"sent_$sig"} > 4)
1553 $sig = 9; # SIGKILL if SIGTERM doesn't work
1555 if (!exists $proc{$_}->{"sent_$sig"})
1557 Log ($proc{$_}->{jobstep}, "sending 2x signal $sig to pid $_");
1559 select (undef, undef, undef, 0.1);
1562 kill $sig, $_; # srun wants two SIGINT to really interrupt
1564 $proc{$_}->{"sent_$sig"} = time;
1565 $proc{$_}->{"killedafter"} = time - $proc{$_}->{"time"};
1575 vec($bits,fileno($_),1) = 1;
1581 # Send log output to Keep via arv-put.
1583 # $log_pipe_in and $log_pipe_out are the input and output filehandles to the arv-put pipe.
1584 # $log_pipe_out_buf is a string containing all output read from arv-put so far.
1585 # $log_pipe_out_select is an IO::Select object around $log_pipe_out.
1586 # $log_pipe_pid is the pid of the arv-put subprocess.
1588 # The only functions that should access these variables directly are:
1590 # log_writer_start($logfilename)
1591 # Starts an arv-put pipe, reading data on stdin and writing it to
1592 # a $logfilename file in an output collection.
1594 # log_writer_read_output([$timeout])
1595 # Read output from $log_pipe_out and append it to $log_pipe_out_buf.
1596 # Passes $timeout to the select() call, with a default of 0.01.
1597 # Returns the result of the last read() call on $log_pipe_out, or
1598 # -1 if read() wasn't called because select() timed out.
1599 # Only other log_writer_* functions should need to call this.
1601 # log_writer_send($txt)
1602 # Writes $txt to the output log collection.
1604 # log_writer_finish()
1605 # Closes the arv-put pipe and returns the output that it produces.
1607 # log_writer_is_active()
1608 # Returns a true value if there is currently a live arv-put
1609 # process, false otherwise.
1611 my ($log_pipe_in, $log_pipe_out, $log_pipe_out_buf, $log_pipe_out_select,
1614 sub log_writer_start($)
1616 my $logfilename = shift;
1617 $log_pipe_pid = open2($log_pipe_out, $log_pipe_in,
1621 '--filename', $logfilename,
1623 $log_pipe_out_buf = "";
1624 $log_pipe_out_select = IO::Select->new($log_pipe_out);
1627 sub log_writer_read_output {
1628 my $timeout = shift || 0.01;
1630 while ($read && $log_pipe_out_select->can_read($timeout)) {
1631 $read = read($log_pipe_out, $log_pipe_out_buf, 65536,
1632 length($log_pipe_out_buf));
1634 if (!defined($read)) {
1635 Log(undef, "error reading log manifest from arv-put: $!");
1640 sub log_writer_send($)
1643 print $log_pipe_in $txt;
1644 log_writer_read_output();
1647 sub log_writer_finish()
1649 return unless $log_pipe_pid;
1651 close($log_pipe_in);
1653 my $read_result = log_writer_read_output(120);
1654 if ($read_result == -1) {
1655 Log (undef, "timed out reading from 'arv-put'");
1656 } elsif ($read_result != 0) {
1657 Log(undef, "failed to read arv-put log manifest to EOF");
1660 waitpid($log_pipe_pid, 0);
1662 Log(undef, "log_writer_finish: arv-put exited " . exit_status_s($?))
1665 close($log_pipe_out);
1666 my $arv_put_output = $log_pipe_out_buf;
1667 $log_pipe_pid = $log_pipe_in = $log_pipe_out = $log_pipe_out_buf =
1668 $log_pipe_out_select = undef;
1670 return $arv_put_output;
1673 sub log_writer_is_active() {
1674 return $log_pipe_pid;
1677 sub Log # ($jobstep_id, $logmessage)
1679 if ($_[1] =~ /\n/) {
1680 for my $line (split (/\n/, $_[1])) {
1685 my $fh = select STDERR; $|=1; select $fh;
1686 my $message = sprintf ("%s %d %s %s", $job_id, $$, @_);
1687 $message =~ s{([^ -\176])}{"\\" . sprintf ("%03o", ord($1))}ge;
1690 if (log_writer_is_active() || -t STDERR) {
1691 my @gmtime = gmtime;
1692 $datetime = sprintf ("%04d-%02d-%02d_%02d:%02d:%02d",
1693 $gmtime[5]+1900, $gmtime[4]+1, @gmtime[3,2,1,0]);
1695 print STDERR ((-t STDERR) ? ($datetime." ".$message) : $message);
1697 if (log_writer_is_active()) {
1698 log_writer_send($datetime . " " . $message);
1705 my ($package, $file, $line) = caller;
1706 my $message = "@_ at $file line $line\n";
1707 Log (undef, $message);
1708 freeze() if @jobstep_todo;
1709 create_output_collection() if @jobstep_todo;
1719 if ($Job->{'state'} eq 'Cancelled') {
1720 $Job->update_attributes('finished_at' => scalar gmtime);
1722 $Job->update_attributes('state' => 'Failed');
1729 my $justcheckpoint = shift; # false if this will be the last meta saved
1730 return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm
1731 return unless log_writer_is_active();
1733 my $log_manifest = "";
1735 my $prev_log_coll = api_call("collections/get", uuid => $Job->{log});
1736 $log_manifest .= $prev_log_coll->{manifest_text};
1738 $log_manifest .= log_writer_finish();
1740 my $log_coll = api_call(
1741 "collections/create", ensure_unique_name => 1, collection => {
1742 manifest_text => $log_manifest,
1743 owner_uuid => $Job->{owner_uuid},
1744 name => sprintf("Log from %s job %s", $Job->{script}, $Job->{uuid}),
1746 Log(undef, "log collection is " . $log_coll->{portable_data_hash});
1747 $Job->update_attributes('log' => $log_coll->{portable_data_hash});
1751 sub freeze_if_want_freeze
1753 if ($main::please_freeze)
1755 release_allocation();
1758 # kill some srun procs before freeze+stop
1759 map { $proc{$_} = {} } @_;
1762 killem (keys %proc);
1763 select (undef, undef, undef, 0.1);
1765 while (($died = waitpid (-1, WNOHANG)) > 0)
1767 delete $proc{$died};
1772 create_output_collection();
1782 Log (undef, "Freeze not implemented");
1789 croak ("Thaw not implemented");
1805 $s =~ s{\\(.)}{$1 eq "n" ? "\n" : $1}ge;
1812 my $srunargs = shift;
1813 my $execargs = shift;
1814 my $opts = shift || {};
1816 my $args = $have_slurm ? [@$srunargs, @$execargs] : $execargs;
1818 $Data::Dumper::Terse = 1;
1819 $Data::Dumper::Indent = 0;
1820 my $show_cmd = Dumper($args);
1821 $show_cmd =~ s/(TOKEN\\*=)[^\s\']+/${1}[...]/g;
1822 $show_cmd =~ s/\n/ /g;
1823 if ($opts->{fork}) {
1824 Log(undef, "starting: $show_cmd");
1826 # This is a child process: parent is in charge of reading our
1827 # stderr and copying it to Log() if needed.
1828 warn "starting: $show_cmd\n";
1831 if (defined $stdin) {
1832 my $child = open STDIN, "-|";
1833 defined $child or die "no fork: $!";
1835 print $stdin or die $!;
1836 close STDOUT or die $!;
1841 return system (@$args) if $opts->{fork};
1844 warn "ENV size is ".length(join(" ",%ENV));
1845 die "exec failed: $!: @$args";
1849 sub ban_node_by_slot {
1850 # Don't start any new jobsteps on this node for 60 seconds
1852 $slot[$slotid]->{node}->{hold_until} = 60 + scalar time;
1853 $slot[$slotid]->{node}->{hold_count}++;
1854 Log (undef, "backing off node " . $slot[$slotid]->{node}->{name} . " for 60 seconds");
1859 my ($lockfile, $error_message) = @_;
1860 open L, ">", $lockfile or croak("$lockfile: $!");
1861 if (!flock L, LOCK_EX|LOCK_NB) {
1862 croak("Can't lock $lockfile: $error_message\n");
1866 sub find_docker_image {
1867 # Given a Keep locator, check to see if it contains a Docker image.
1868 # If so, return its stream name and Docker hash.
1869 # If not, return undef for both values.
1870 my $locator = shift;
1871 my ($streamname, $filename);
1872 my $image = api_call("collections/get", uuid => $locator);
1874 foreach my $line (split(/\n/, $image->{manifest_text})) {
1875 my @tokens = split(/\s+/, $line);
1877 $streamname = shift(@tokens);
1878 foreach my $filedata (grep(/^\d+:\d+:/, @tokens)) {
1879 if (defined($filename)) {
1880 return (undef, undef); # More than one file in the Collection.
1882 $filename = (split(/:/, $filedata, 3))[2];
1887 if (defined($filename) and ($filename =~ /^([0-9A-Fa-f]{64})\.tar$/)) {
1888 return ($streamname, $1);
1890 return (undef, undef);
1895 # Calculate the number of times an operation should be retried,
1896 # assuming exponential backoff, and that we're willing to retry as
1897 # long as tasks have been running. Enforce a minimum of 3 retries.
1898 my ($starttime, $endtime, $timediff, $retries);
1900 $starttime = $jobstep[0]->{starttime};
1901 $endtime = $jobstep[-1]->{finishtime};
1903 if (!defined($starttime)) {
1905 } elsif (!defined($endtime)) {
1906 $timediff = time - $starttime;
1908 $timediff = ($endtime - $starttime) - (time - $endtime);
1910 if ($timediff > 0) {
1911 $retries = int(log($timediff) / log(2));
1913 $retries = 1; # Use the minimum.
1915 return ($retries > 3) ? $retries : 3;
1919 # Pass in two function references.
1920 # This method will be called with the remaining arguments.
1921 # If it dies, retry it with exponential backoff until it succeeds,
1922 # or until the current retry_count is exhausted. After each failure
1923 # that can be retried, the second function will be called with
1924 # the current try count (0-based), next try time, and error message.
1925 my $operation = shift;
1926 my $retry_callback = shift;
1927 my $retries = retry_count();
1928 foreach my $try_count (0..$retries) {
1929 my $next_try = time + (2 ** $try_count);
1930 my $result = eval { $operation->(@_); };
1933 } elsif ($try_count < $retries) {
1934 $retry_callback->($try_count, $next_try, $@);
1935 my $sleep_time = $next_try - time;
1936 sleep($sleep_time) if ($sleep_time > 0);
1939 # Ensure the error message ends in a newline, so Perl doesn't add
1940 # retry_op's line number to it.
1946 # Pass in a /-separated API method name, and arguments for it.
1947 # This function will call that method, retrying as needed until
1948 # the current retry_count is exhausted, with a log on the first failure.
1949 my $method_name = shift;
1950 my $log_api_retry = sub {
1951 my ($try_count, $next_try_at, $errmsg) = @_;
1952 $errmsg =~ s/\s*\bat \Q$0\E line \d+\.?\s*//;
1953 $errmsg =~ s/\s/ /g;
1954 $errmsg =~ s/\s+$//;
1956 if ($next_try_at < time) {
1957 $retry_msg = "Retrying.";
1959 my $next_try_fmt = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($next_try_at);
1960 $retry_msg = "Retrying at $next_try_fmt.";
1962 Log(undef, "API method $method_name failed: $errmsg. $retry_msg");
1965 foreach my $key (split(/\//, $method_name)) {
1966 $method = $method->{$key};
1968 return retry_op(sub { $method->execute(@_); }, $log_api_retry, @_);
1972 # Given a $?, return a human-readable exit code string like "0" or
1973 # "1" or "0 with signal 1" or "1 with signal 11".
1974 my $exitcode = shift;
1975 my $s = $exitcode >> 8;
1976 if ($exitcode & 0x7f) {
1977 $s .= " with signal " . ($exitcode & 0x7f);
1979 if ($exitcode & 0x80) {
1980 $s .= " with core dump";
1985 sub handle_readall {
1986 # Pass in a glob reference to a file handle.
1987 # Read all its contents and return them as a string.
1988 my $fh_glob_ref = shift;
1990 return <$fh_glob_ref>;
1993 sub tar_filename_n {
1995 return sprintf("%s/git.%s.%d.tar", $ENV{CRUNCH_TMP}, $job_id, $n);
1998 sub add_git_archive {
1999 # Pass in a git archive command as a string or list, a la system().
2000 # This method will save its output to be included in the archive sent to the
2004 if (!open(GIT_ARCHIVE, ">", tar_filename_n($git_tar_count))) {
2005 croak("Failed to save git archive: $!");
2007 my $git_pid = open2(">&GIT_ARCHIVE", $git_input, @_);
2009 waitpid($git_pid, 0);
2012 croak("Failed to save git archive: git exited " . exit_status_s($?));
2016 sub combined_git_archive {
2017 # Combine all saved tar archives into a single archive, then return its
2018 # contents in a string. Return undef if no archives have been saved.
2019 if ($git_tar_count < 1) {
2022 my $base_tar_name = tar_filename_n(1);
2023 foreach my $tar_to_append (map { tar_filename_n($_); } (2..$git_tar_count)) {
2024 my $tar_exit = system("tar", "-Af", $base_tar_name, $tar_to_append);
2025 if ($tar_exit != 0) {
2026 croak("Error preparing build archive: tar -A exited " .
2027 exit_status_s($tar_exit));
2030 if (!open(GIT_TAR, "<", $base_tar_name)) {
2031 croak("Could not open build archive: $!");
2033 my $tar_contents = handle_readall(\*GIT_TAR);
2035 return $tar_contents;
2038 sub set_nonblocking {
2040 my $flags = fcntl ($fh, F_GETFL, 0) or croak ($!);
2041 fcntl ($fh, F_SETFL, $flags | O_NONBLOCK) or croak ($!);
2047 # This is crunch-job's internal dispatch script. crunch-job running on the API
2048 # server invokes this script on individual compute nodes, or localhost if we're
2049 # running a job locally. It gets called in two modes:
2051 # * No arguments: Installation mode. Read a tar archive from the DATA
2052 # file handle; it includes the Crunch script's source code, and
2053 # maybe SDKs as well. Those should be installed in the proper
2054 # locations. This runs outside of any Docker container, so don't try to
2055 # introspect Crunch's runtime environment.
2057 # * With arguments: Crunch script run mode. This script should set up the
2058 # environment, then run the command specified in the arguments. This runs
2059 # inside any Docker container.
2062 use File::Path qw( make_path remove_tree );
2063 use POSIX qw(getcwd);
2065 use constant TASK_TEMPFAIL => 111;
2067 # Map SDK subdirectories to the path environments they belong to.
2068 my %SDK_ENVVARS = ("perl/lib" => "PERLLIB", "ruby/lib" => "RUBYLIB");
2070 my $destdir = $ENV{"CRUNCH_SRC"};
2071 my $archive_hash = $ENV{"CRUNCH_GIT_ARCHIVE_HASH"};
2072 my $repo = $ENV{"CRUNCH_SRC_URL"};
2073 my $install_dir = $ENV{"CRUNCH_INSTALL"} || (getcwd() . "/opt");
2074 my $job_work = $ENV{"JOB_WORK"};
2075 my $task_work = $ENV{"TASK_WORK"};
2077 open(STDOUT_ORIG, ">&", STDOUT);
2078 open(STDERR_ORIG, ">&", STDERR);
2080 for my $dir ($destdir, $job_work, $task_work) {
2083 -e $dir or die "Failed to create temporary directory ($dir): $!";
2088 remove_tree($task_work, {keep_root => 1});
2091 ### Crunch script run mode
2093 # We want to do routine logging during task 0 only. This gives the user
2094 # the information they need, but avoids repeating the information for every
2097 if ($ENV{TASK_SEQUENCE} eq "0") {
2100 printf STDERR_ORIG "[Crunch] $msg\n", @_;
2106 my $python_src = "$install_dir/python";
2107 my $venv_dir = "$job_work/.arvados.venv";
2108 my $venv_built = -e "$venv_dir/bin/activate";
2109 if ((!$venv_built) and (-d $python_src) and can_run("virtualenv")) {
2110 shell_or_die(undef, "virtualenv", "--quiet", "--system-site-packages",
2111 "--python=python2.7", $venv_dir);
2112 shell_or_die(TASK_TEMPFAIL, "$venv_dir/bin/pip", "--quiet", "install", "-I", $python_src);
2114 $Log->("Built Python SDK virtualenv");
2117 my $pip_bin = "pip";
2119 $Log->("Running in Python SDK virtualenv");
2120 $pip_bin = "$venv_dir/bin/pip";
2121 my $orig_argv = join(" ", map { quotemeta($_); } @ARGV);
2122 @ARGV = ("/bin/sh", "-ec",
2123 ". \Q$venv_dir/bin/activate\E; exec $orig_argv");
2124 } elsif (-d $python_src) {
2125 $Log->("Warning: virtualenv not found inside Docker container default " .
2126 "\$PATH. Can't install Python SDK.");
2129 my $pkgs = `(\Q$pip_bin\E freeze 2>/dev/null | grep arvados) || dpkg-query --show '*arvados*'`;
2131 $Log->("Using Arvados SDK:");
2132 foreach my $line (split /\n/, $pkgs) {
2136 $Log->("Arvados SDK packages not found");
2139 while (my ($sdk_dir, $sdk_envkey) = each(%SDK_ENVVARS)) {
2140 my $sdk_path = "$install_dir/$sdk_dir";
2142 if ($ENV{$sdk_envkey}) {
2143 $ENV{$sdk_envkey} = "$sdk_path:" . $ENV{$sdk_envkey};
2145 $ENV{$sdk_envkey} = $sdk_path;
2147 $Log->("Arvados SDK added to %s", $sdk_envkey);
2152 die "Cannot exec `@ARGV`: $!";
2155 ### Installation mode
2156 open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
2158 if (readlink ("$destdir.archive_hash") eq $archive_hash && -d $destdir) {
2159 # This exact git archive (source + arvados sdk) is already installed
2160 # here, so there's no need to reinstall it.
2162 # We must consume our DATA section, though: otherwise the process
2163 # feeding it to us will get SIGPIPE.
2165 while (read(DATA, $buf, 65536)) { }
2170 unlink "$destdir.archive_hash";
2174 # Ignore SIGPIPE: we check retval of close() instead. See perlipc(1).
2175 local $SIG{PIPE} = "IGNORE";
2176 warn "Extracting archive: $archive_hash\n";
2177 # --ignore-zeros is necessary sometimes: depending on how much NUL
2178 # padding tar -A put on our combined archive (which in turn depends
2179 # on the length of the component archives) tar without
2180 # --ignore-zeros will exit before consuming stdin and cause close()
2181 # to fail on the resulting SIGPIPE.
2182 if (!open(TARX, "|-", "tar", "--ignore-zeros", "-xC", $destdir)) {
2183 die "Error launching 'tar -xC $destdir': $!";
2185 # If we send too much data to tar in one write (> 4-5 MiB), it stops, and we
2186 # get SIGPIPE. We must feed it data incrementally.
2188 while (read(DATA, $tar_input, 65536)) {
2189 print TARX $tar_input;
2192 die "'tar -xC $destdir' exited $?: $!";
2198 my $sdk_root = "$destdir/.arvados.sdk/sdk";
2200 foreach my $sdk_lang (("python",
2201 map { (split /\//, $_, 2)[0]; } keys(%SDK_ENVVARS))) {
2202 if (-d "$sdk_root/$sdk_lang") {
2203 if (!rename("$sdk_root/$sdk_lang", "$install_dir/$sdk_lang")) {
2204 die "Failed to install $sdk_lang SDK: $!";
2210 my $python_dir = "$install_dir/python";
2211 if ((-d $python_dir) and can_run("python2.7")) {
2212 open(my $egg_info_pipe, "-|",
2213 "python2.7 \Q$python_dir/setup.py\E --quiet egg_info 2>&1 >/dev/null");
2214 my @egg_info_errors = <$egg_info_pipe>;
2215 close($egg_info_pipe);
2217 if (@egg_info_errors and ($egg_info_errors[-1] =~ /\bgit\b/)) {
2218 # egg_info apparently failed because it couldn't ask git for a build tag.
2219 # Specify no build tag.
2220 open(my $pysdk_cfg, ">>", "$python_dir/setup.cfg");
2221 print $pysdk_cfg "\n[egg_info]\ntag_build =\n";
2224 my $egg_info_exit = $? >> 8;
2225 foreach my $errline (@egg_info_errors) {
2226 print STDERR_ORIG $errline;
2228 warn "python setup.py egg_info failed: exit $egg_info_exit";
2229 exit ($egg_info_exit || 1);
2234 # Hide messages from the install script (unless it fails: shell_or_die
2235 # will show $destdir.log in that case).
2236 open(STDOUT, ">>", "$destdir.log");
2237 open(STDERR, ">&", STDOUT);
2239 if (-e "$destdir/crunch_scripts/install") {
2240 shell_or_die (undef, "$destdir/crunch_scripts/install", $install_dir);
2241 } elsif (!-e "./install.sh" && -e "./tests/autotests.sh") {
2243 shell_or_die (undef, "./tests/autotests.sh", $install_dir);
2244 } elsif (-e "./install.sh") {
2245 shell_or_die (undef, "./install.sh", $install_dir);
2248 if ($archive_hash) {
2249 unlink "$destdir.archive_hash.new";
2250 symlink ($archive_hash, "$destdir.archive_hash.new") or die "$destdir.archive_hash.new: $!";
2251 rename ("$destdir.archive_hash.new", "$destdir.archive_hash") or die "$destdir.archive_hash: $!";
2257 my $command_name = shift;
2258 open(my $which, "-|", "which", $command_name);
2259 while (<$which>) { }
2266 my $exitcode = shift;
2268 if ($ENV{"DEBUG"}) {
2269 print STDERR "@_\n";
2271 if (system (@_) != 0) {
2274 my $exitstatus = sprintf("exit %d signal %d", $code >> 8, $code & 0x7f);
2275 open STDERR, ">&STDERR_ORIG";
2276 system ("cat $destdir.log >&2");
2277 warn "@_ failed ($err): $exitstatus";
2278 if (defined($exitcode)) {
2282 exit (($code >> 8) || 1);