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;
102 $ENV{"TMPDIR"} ||= "/tmp";
103 unless (defined $ENV{"CRUNCH_TMP"}) {
104 $ENV{"CRUNCH_TMP"} = $ENV{"TMPDIR"} . "/crunch-job";
105 if ($ENV{"USER"} ne "crunch" && $< != 0) {
106 # use a tmp dir unique for my uid
107 $ENV{"CRUNCH_TMP"} .= "-$<";
111 # Create the tmp directory if it does not exist
112 if ( ! -d $ENV{"CRUNCH_TMP"} ) {
113 make_path $ENV{"CRUNCH_TMP"} or die "Failed to create temporary working directory: " . $ENV{"CRUNCH_TMP"};
116 $ENV{"JOB_WORK"} = $ENV{"CRUNCH_TMP"} . "/work";
117 $ENV{"CRUNCH_INSTALL"} = "$ENV{CRUNCH_TMP}/opt";
118 $ENV{"CRUNCH_WORK"} = $ENV{"JOB_WORK"}; # deprecated
119 mkdir ($ENV{"JOB_WORK"});
127 GetOptions('force-unlock' => \$force_unlock,
128 'git-dir=s' => \$git_dir,
129 'job=s' => \$jobspec,
130 'job-api-token=s' => \$job_api_token,
131 'no-clear-tmp' => \$no_clear_tmp,
132 'resume-stash=s' => \$resume_stash,
135 if (defined $job_api_token) {
136 $ENV{ARVADOS_API_TOKEN} = $job_api_token;
139 my $have_slurm = exists $ENV{SLURM_JOBID} && exists $ENV{SLURM_NODELIST};
145 $main::ENV{CRUNCH_DEBUG} = 1;
149 $main::ENV{CRUNCH_DEBUG} = 0;
154 my $arv = Arvados->new('apiVersion' => 'v1');
162 my $User = api_call("users/current");
164 if ($jobspec =~ /^[-a-z\d]+$/)
166 # $jobspec is an Arvados UUID, not a JSON job specification
167 $Job = api_call("jobs/get", uuid => $jobspec);
168 if (!$force_unlock) {
169 # Claim this job, and make sure nobody else does
170 eval { api_call("jobs/lock", uuid => $Job->{uuid}); };
172 Log(undef, "Error while locking job, exiting ".EX_TEMPFAIL);
179 $Job = JSON::decode_json($jobspec);
183 map { croak ("No $_ specified") unless $Job->{$_} }
184 qw(script script_version script_parameters);
187 $Job->{'is_locked_by_uuid'} = $User->{'uuid'};
188 $Job->{'started_at'} = gmtime;
189 $Job->{'state'} = 'Running';
191 $Job = api_call("jobs/create", job => $Job);
193 $job_id = $Job->{'uuid'};
195 my $keep_logfile = $job_id . '.log.txt';
196 log_writer_start($keep_logfile);
198 $Job->{'runtime_constraints'} ||= {};
199 $Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0;
200 my $max_ncpus = $Job->{'runtime_constraints'}->{'max_tasks_per_node'};
202 my $gem_versions = `gem list --quiet arvados-cli 2>/dev/null`;
204 $gem_versions =~ s/^arvados-cli \(/ with arvados-cli Gem version(s) /;
205 chomp($gem_versions);
206 chop($gem_versions); # Closing parentheses
211 "running from " . ((-e $0) ? realpath($0) : "stdin") . $gem_versions);
213 Log (undef, "check slurm allocation");
216 # Should use $ENV{SLURM_TASKS_PER_NODE} instead of sinfo? (eg. "4(x3),2,4(x2)")
220 my $localcpus = 0 + `grep -cw ^processor /proc/cpuinfo` || 1;
221 push @sinfo, "$localcpus localhost";
223 if (exists $ENV{SLURM_NODELIST})
225 push @sinfo, `sinfo -h --format='%c %N' --nodes=\Q$ENV{SLURM_NODELIST}\E`;
229 my ($ncpus, $slurm_nodelist) = split;
230 $ncpus = $max_ncpus if $max_ncpus && $ncpus > $max_ncpus;
233 while ($slurm_nodelist =~ s/^([^\[,]+?(\[.*?\])?)(,|$)//)
236 if ($nodelist =~ /\[((\d+)(-(\d+))?(,(\d+)(-(\d+))?)*)\]/)
239 foreach (split (",", $ranges))
252 push @nodelist, map {
254 $n =~ s/\[[-,\d]+\]/$_/;
261 push @nodelist, $nodelist;
264 foreach my $nodename (@nodelist)
266 Log (undef, "node $nodename - $ncpus slots");
267 my $node = { name => $nodename,
271 foreach my $cpu (1..$ncpus)
273 push @slot, { node => $node,
277 push @node, @nodelist;
282 # Ensure that we get one jobstep running on each allocated node before
283 # we start overloading nodes with concurrent steps
285 @slot = sort { $a->{cpu} <=> $b->{cpu} } @slot;
288 $Job->update_attributes(
289 'tasks_summary' => { 'failed' => 0,
294 Log (undef, "start");
295 $SIG{'INT'} = sub { $main::please_freeze = 1; };
296 $SIG{'QUIT'} = sub { $main::please_freeze = 1; };
297 $SIG{'TERM'} = \&croak;
298 $SIG{'TSTP'} = sub { $main::please_freeze = 1; };
299 $SIG{'ALRM'} = sub { $main::please_info = 1; };
300 $SIG{'CONT'} = sub { $main::please_continue = 1; };
301 $SIG{'HUP'} = sub { $main::please_refresh = 1; };
303 $main::please_freeze = 0;
304 $main::please_info = 0;
305 $main::please_continue = 0;
306 $main::please_refresh = 0;
307 my $jobsteps_must_output_keys = 0; # becomes 1 when any task outputs a key
309 grep { $ENV{$1} = $2 if /^(NOCACHE.*?)=(.*)/ } split ("\n", $$Job{knobs});
310 $ENV{"CRUNCH_JOB_UUID"} = $job_id;
311 $ENV{"JOB_UUID"} = $job_id;
314 my @jobstep_todo = ();
315 my @jobstep_done = ();
316 my @jobstep_tomerge = ();
317 my $jobstep_tomerge_level = 0;
319 my $squeue_kill_checked;
320 my $latest_refresh = scalar time;
324 if (defined $Job->{thawedfromkey})
326 thaw ($Job->{thawedfromkey});
330 my $first_task = api_call("job_tasks/create", job_task => {
331 'job_uuid' => $Job->{'uuid'},
336 push @jobstep, { 'level' => 0,
338 'arvados_task' => $first_task,
340 push @jobstep_todo, 0;
346 must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
349 my $build_script = handle_readall(\*DATA);
350 my $nodelist = join(",", @node);
351 my $git_tar_count = 0;
353 if (!defined $no_clear_tmp) {
354 # Clean out crunch_tmp/work, crunch_tmp/opt, crunch_tmp/src*
355 Log (undef, "Clean work dirs");
357 my $cleanpid = fork();
360 # Find FUSE mounts that look like Keep mounts (the mount path has the
361 # word "keep") and unmount them. Then clean up work directories.
362 # TODO: When #5036 is done and widely deployed, we can get rid of the
363 # regular expression and just unmount everything with type fuse.keep.
364 srun (["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
365 ['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']);
370 last if $cleanpid == waitpid (-1, WNOHANG);
371 freeze_if_want_freeze ($cleanpid);
372 select (undef, undef, undef, 0.1);
374 Log (undef, "Cleanup command exited ".exit_status_s($?));
377 # If this job requires a Docker image, install that.
378 my $docker_bin = "/usr/bin/docker.io";
379 my ($docker_locator, $docker_stream, $docker_hash, $docker_limitmem);
380 if ($docker_locator = $Job->{docker_image_locator}) {
381 ($docker_stream, $docker_hash) = find_docker_image($docker_locator);
384 croak("No Docker image hash found from locator $docker_locator");
386 $docker_stream =~ s/^\.//;
387 my $docker_install_script = qq{
388 if ! $docker_bin images -q --no-trunc --all | grep -qxF \Q$docker_hash\E; then
389 arv-get \Q$docker_locator$docker_stream/$docker_hash.tar\E | $docker_bin load
392 my $docker_pid = fork();
393 if ($docker_pid == 0)
395 srun (["srun", "--nodelist=" . join(',', @node)],
396 ["/bin/sh", "-ec", $docker_install_script]);
401 last if $docker_pid == waitpid (-1, WNOHANG);
402 freeze_if_want_freeze ($docker_pid);
403 select (undef, undef, undef, 0.1);
407 croak("Installing Docker image from $docker_locator exited "
411 # Determine whether this version of Docker supports memory+swap limits.
412 srun(["srun", "--nodelist=" . $node[0]],
413 ["/bin/sh", "-ec", "$docker_bin run --help | grep -qe --memory-swap="],
415 $docker_limitmem = ($? == 0);
417 if ($Job->{arvados_sdk_version}) {
418 # The job also specifies an Arvados SDK version. Add the SDKs to the
419 # tar file for the build script to install.
420 Log(undef, sprintf("Packing Arvados SDK version %s for installation",
421 $Job->{arvados_sdk_version}));
422 add_git_archive("git", "--git-dir=$git_dir", "archive",
423 "--prefix=.arvados.sdk/",
424 $Job->{arvados_sdk_version}, "sdk");
428 if (!defined $git_dir && $Job->{'script_version'} =~ m{^/}) {
429 # If script_version looks like an absolute path, *and* the --git-dir
430 # argument was not given -- which implies we were not invoked by
431 # crunch-dispatch -- we will use the given path as a working
432 # directory instead of resolving script_version to a git commit (or
433 # doing anything else with git).
434 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{'script_version'};
435 $ENV{"CRUNCH_SRC"} = $Job->{'script_version'};
438 # Resolve the given script_version to a git commit sha1. Also, if
439 # the repository is remote, clone it into our local filesystem: this
440 # ensures "git archive" will work, and is necessary to reliably
441 # resolve a symbolic script_version like "master^".
442 $ENV{"CRUNCH_SRC"} = "$ENV{CRUNCH_TMP}/src";
444 Log (undef, "Looking for version ".$Job->{script_version}." from repository ".$Job->{repository});
446 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{script_version};
448 # If we're running under crunch-dispatch, it will have already
449 # pulled the appropriate source tree into its own repository, and
450 # given us that repo's path as $git_dir.
452 # If we're running a "local" job, we might have to fetch content
453 # from a remote repository.
455 # (Currently crunch-dispatch gives a local path with --git-dir, but
456 # we might as well accept URLs there too in case it changes its
458 my $repo = $git_dir || $Job->{'repository'};
460 # Repository can be remote or local. If remote, we'll need to fetch it
461 # to a local dir before doing `git log` et al.
464 if ($repo =~ m{://|^[^/]*:}) {
465 # $repo is a git url we can clone, like git:// or https:// or
466 # file:/// or [user@]host:repo.git. Note "user/name@host:foo" is
467 # not recognized here because distinguishing that from a local
468 # path is too fragile. If you really need something strange here,
469 # use the ssh:// form.
470 $repo_location = 'remote';
471 } elsif ($repo =~ m{^\.*/}) {
472 # $repo is a local path to a git index. We'll also resolve ../foo
473 # to ../foo/.git if the latter is a directory. To help
474 # disambiguate local paths from named hosted repositories, this
475 # form must be given as ./ or ../ if it's a relative path.
476 if (-d "$repo/.git") {
477 $repo = "$repo/.git";
479 $repo_location = 'local';
481 # $repo is none of the above. It must be the name of a hosted
483 my $arv_repo_list = api_call("repositories/list",
484 'filters' => [['name','=',$repo]]);
485 my @repos_found = @{$arv_repo_list->{'items'}};
486 my $n_found = $arv_repo_list->{'serverResponse'}->{'items_available'};
488 Log(undef, "Repository '$repo' -> "
489 . join(", ", map { $_->{'uuid'} } @repos_found));
492 croak("Error: Found $n_found repositories with name '$repo'.");
494 $repo = $repos_found[0]->{'fetch_url'};
495 $repo_location = 'remote';
497 Log(undef, "Using $repo_location repository '$repo'");
498 $ENV{"CRUNCH_SRC_URL"} = $repo;
500 # Resolve given script_version (we'll call that $treeish here) to a
501 # commit sha1 ($commit).
502 my $treeish = $Job->{'script_version'};
504 if ($repo_location eq 'remote') {
505 # We minimize excess object-fetching by re-using the same bare
506 # repository in CRUNCH_TMP/.git for multiple crunch-jobs -- we
507 # just keep adding remotes to it as needed.
508 my $local_repo = $ENV{'CRUNCH_TMP'}."/.git";
509 my $gitcmd = "git --git-dir=\Q$local_repo\E";
511 # Set up our local repo for caching remote objects, making
513 if (!-d $local_repo) {
514 make_path($local_repo) or croak("Error: could not create $local_repo");
516 # This works (exits 0 and doesn't delete fetched objects) even
517 # if $local_repo is already initialized:
518 `$gitcmd init --bare`;
520 croak("Error: $gitcmd init --bare exited ".exit_status_s($?));
523 # If $treeish looks like a hash (or abbrev hash) we look it up in
524 # our local cache first, since that's cheaper. (We don't want to
525 # do that with tags/branches though -- those change over time, so
526 # they should always be resolved by the remote repo.)
527 if ($treeish =~ /^[0-9a-f]{7,40}$/s) {
528 # Hide stderr because it's normal for this to fail:
529 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E 2>/dev/null`;
531 # Careful not to resolve a branch named abcdeff to commit 1234567:
532 $sha1 =~ /^$treeish/ &&
533 $sha1 =~ /^([0-9a-f]{40})$/s) {
535 Log(undef, "Commit $commit already present in $local_repo");
539 if (!defined $commit) {
540 # If $treeish isn't just a hash or abbrev hash, or isn't here
541 # yet, we need to fetch the remote to resolve it correctly.
543 # First, remove all local heads. This prevents a name that does
544 # not exist on the remote from resolving to (or colliding with)
545 # a previously fetched branch or tag (possibly from a different
547 remove_tree("$local_repo/refs/heads", {keep_root => 1});
549 Log(undef, "Fetching objects from $repo to $local_repo");
550 `$gitcmd fetch --no-progress --tags ''\Q$repo\E \Q+refs/heads/*:refs/heads/*\E`;
552 croak("Error: `$gitcmd fetch` exited ".exit_status_s($?));
556 # Now that the data is all here, we will use our local repo for
557 # the rest of our git activities.
561 my $gitcmd = "git --git-dir=\Q$repo\E";
562 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E`;
563 unless ($? == 0 && $sha1 =~ /^([0-9a-f]{40})$/) {
564 croak("`$gitcmd rev-list` exited "
566 .", '$treeish' not found. Giving up.");
569 Log(undef, "Version $treeish is commit $commit");
571 if ($commit ne $Job->{'script_version'}) {
572 # Record the real commit id in the database, frozentokey, logs,
573 # etc. -- instead of an abbreviation or a branch name which can
574 # become ambiguous or point to a different commit in the future.
575 if (!$Job->update_attributes('script_version' => $commit)) {
576 croak("Error: failed to update job's script_version attribute");
580 $ENV{"CRUNCH_SRC_COMMIT"} = $commit;
581 add_git_archive("$gitcmd archive ''\Q$commit\E");
584 my $git_archive = combined_git_archive();
585 if (!defined $git_archive) {
586 Log(undef, "Skip install phase (no git archive)");
588 Log(undef, "Warning: This probably means workers have no source tree!");
592 Log(undef, "Run install script on all workers");
594 my @srunargs = ("srun",
595 "--nodelist=$nodelist",
596 "-D", $ENV{'TMPDIR'}, "--job-name=$job_id");
597 my @execargs = ("sh", "-c",
598 "mkdir -p $ENV{CRUNCH_INSTALL} && cd $ENV{CRUNCH_TMP} && perl -");
600 $ENV{"CRUNCH_GIT_ARCHIVE_HASH"} = md5_hex($git_archive);
601 my ($install_stderr_r, $install_stderr_w);
602 pipe $install_stderr_r, $install_stderr_w or croak("pipe() failed: $!");
603 set_nonblocking($install_stderr_r);
604 my $installpid = fork();
605 if ($installpid == 0)
607 close($install_stderr_r);
608 fcntl($install_stderr_w, F_SETFL, 0) or croak($!); # no close-on-exec
609 open(STDOUT, ">&", $install_stderr_w);
610 open(STDERR, ">&", $install_stderr_w);
611 srun (\@srunargs, \@execargs, {}, $build_script . $git_archive);
614 close($install_stderr_w);
616 while ($installpid != waitpid(-1, WNOHANG)) {
617 freeze_if_want_freeze ($installpid);
618 # Wait up to 0.1 seconds for something to appear on stderr, then
619 # do a non-blocking read.
620 my $bits = fhbits($install_stderr_r);
621 select ($bits, undef, $bits, 0.1);
622 if (0 < sysread ($install_stderr_r, $stderr_buf, 8192, length($stderr_buf)))
624 while ($stderr_buf =~ /^(.*?)\n/) {
626 substr $stderr_buf, 0, 1+length($line), "";
627 Log(undef, "stderr $line");
631 my $install_exited = $?;
632 close($install_stderr_r);
633 if (length($stderr_buf) > 0) {
634 Log(undef, "stderr $stderr_buf")
637 Log (undef, "Install script exited ".exit_status_s($install_exited));
638 foreach my $tar_filename (map { tar_filename_n($_); } (1..$git_tar_count)) {
639 unlink($tar_filename);
641 exit (1) if $install_exited != 0;
644 foreach (qw (script script_version script_parameters runtime_constraints))
648 (ref($Job->{$_}) ? JSON::encode_json($Job->{$_}) : $Job->{$_}));
650 foreach (split (/\n/, $Job->{knobs}))
652 Log (undef, "knob " . $_);
657 $main::success = undef;
663 my $thisround_succeeded = 0;
664 my $thisround_failed = 0;
665 my $thisround_failed_multiple = 0;
667 @jobstep_todo = sort { $jobstep[$a]->{level} <=> $jobstep[$b]->{level}
668 or $a <=> $b } @jobstep_todo;
669 my $level = $jobstep[$jobstep_todo[0]]->{level};
671 my $initial_tasks_this_level = 0;
672 foreach my $id (@jobstep_todo) {
673 $initial_tasks_this_level++ if ($jobstep[$id]->{level} == $level);
676 # If the number of tasks scheduled at this level #T is smaller than the number
677 # of slots available #S, only use the first #T slots, or the first slot on
678 # each node, whichever number is greater.
680 # When we dispatch tasks later, we'll allocate whole-node resources like RAM
681 # based on these numbers. Using fewer slots makes more resources available
682 # to each individual task, which should normally be a better strategy when
683 # there are fewer of them running with less parallelism.
685 # Note that this calculation is not redone if the initial tasks at
686 # this level queue more tasks at the same level. This may harm
687 # overall task throughput for that level.
689 if ($initial_tasks_this_level < @node) {
690 @freeslot = (0..$#node);
691 } elsif ($initial_tasks_this_level < @slot) {
692 @freeslot = (0..$initial_tasks_this_level - 1);
694 @freeslot = (0..$#slot);
696 my $round_num_freeslots = scalar(@freeslot);
698 my %round_max_slots = ();
699 for (my $ii = $#freeslot; $ii >= 0; $ii--) {
700 my $this_slot = $slot[$freeslot[$ii]];
701 my $node_name = $this_slot->{node}->{name};
702 $round_max_slots{$node_name} ||= $this_slot->{cpu};
703 last if (scalar(keys(%round_max_slots)) >= @node);
706 Log(undef, "start level $level with $round_num_freeslots slots");
710 my $progress_is_dirty = 1;
711 my $progress_stats_updated = 0;
713 update_progress_stats();
717 for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
719 my $id = $jobstep_todo[$todo_ptr];
720 my $Jobstep = $jobstep[$id];
721 if ($Jobstep->{level} != $level)
726 pipe $reader{$id}, "writer" or croak("pipe() failed: $!");
727 set_nonblocking($reader{$id});
729 my $childslot = $freeslot[0];
730 my $childnode = $slot[$childslot]->{node};
731 my $childslotname = join (".",
732 $slot[$childslot]->{node}->{name},
733 $slot[$childslot]->{cpu});
735 my $childpid = fork();
738 $SIG{'INT'} = 'DEFAULT';
739 $SIG{'QUIT'} = 'DEFAULT';
740 $SIG{'TERM'} = 'DEFAULT';
742 foreach (values (%reader))
746 fcntl ("writer", F_SETFL, 0) or croak ($!); # no close-on-exec
747 open(STDOUT,">&writer");
748 open(STDERR,">&writer");
753 delete $ENV{"GNUPGHOME"};
754 $ENV{"TASK_UUID"} = $Jobstep->{'arvados_task'}->{'uuid'};
755 $ENV{"TASK_QSEQUENCE"} = $id;
756 $ENV{"TASK_SEQUENCE"} = $level;
757 $ENV{"JOB_SCRIPT"} = $Job->{script};
758 while (my ($param, $value) = each %{$Job->{script_parameters}}) {
759 $param =~ tr/a-z/A-Z/;
760 $ENV{"JOB_PARAMETER_$param"} = $value;
762 $ENV{"TASK_SLOT_NODE"} = $slot[$childslot]->{node}->{name};
763 $ENV{"TASK_SLOT_NUMBER"} = $slot[$childslot]->{cpu};
764 $ENV{"TASK_WORK"} = $ENV{"CRUNCH_TMP"}."/task/$childslotname";
765 $ENV{"HOME"} = $ENV{"TASK_WORK"};
766 $ENV{"TASK_KEEPMOUNT"} = $ENV{"TASK_WORK"}.".keep";
767 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
768 $ENV{"CRUNCH_NODE_SLOTS"} = $round_max_slots{$ENV{TASK_SLOT_NODE}};
769 $ENV{"PATH"} = $ENV{"CRUNCH_INSTALL"} . "/bin:" . $ENV{"PATH"};
775 "--nodelist=".$childnode->{name},
776 qw(-n1 -c1 -N1 -D), $ENV{'TMPDIR'},
777 "--job-name=$job_id.$id.$$",
780 "if [ -e $ENV{TASK_WORK} ]; then rm -rf $ENV{TASK_WORK}; fi; "
781 ."mkdir -p $ENV{CRUNCH_TMP} $ENV{JOB_WORK} $ENV{TASK_WORK} $ENV{TASK_KEEPMOUNT} "
782 ."&& cd $ENV{CRUNCH_TMP} "
783 # These environment variables get used explicitly later in
784 # $command. No tool is expected to read these values directly.
785 .q{&& MEM=$(awk '($1 == "MemTotal:"){print $2}' </proc/meminfo) }
786 .q{&& SWAP=$(awk '($1 == "SwapTotal:"){print $2}' </proc/meminfo) }
787 ."&& MEMLIMIT=\$(( (\$MEM * 95) / ($ENV{CRUNCH_NODE_SLOTS} * 100) )) "
788 ."&& let SWAPLIMIT=\$MEMLIMIT+\$SWAP ";
789 $command .= "&& exec arv-mount --by-id --allow-other $ENV{TASK_KEEPMOUNT} --exec ";
792 my $cidfile = "$ENV{CRUNCH_TMP}/$Jobstep->{arvados_task}->{uuid}-$Jobstep->{failures}.cid";
793 $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -cgroup-parent=docker -cgroup-cid=$cidfile -poll=10000 ";
794 $command .= "$docker_bin run --rm=true --attach=stdout --attach=stderr --attach=stdin -i --user=crunch --cidfile=$cidfile --sig-proxy ";
795 # We only set memory limits if Docker lets us limit both memory and swap.
796 # Memory limits alone have been supported longer, but subprocesses tend
797 # to get SIGKILL if they exceed that without any swap limit set.
798 # See #5642 for additional background.
799 if ($docker_limitmem) {
800 $command .= "--memory=\${MEMLIMIT}k --memory-swap=\${SWAPLIMIT}k ";
803 # Dynamically configure the container to use the host system as its
804 # DNS server. Get the host's global addresses from the ip command,
805 # and turn them into docker --dns options using gawk.
807 q{$(ip -o address show scope global |
808 gawk 'match($4, /^([0-9\.:]+)\//, x){print "--dns", x[1]}') };
810 # The source tree and $destdir directory (which we have
811 # installed on the worker host) are available in the container,
812 # under the same path.
813 $command .= "--volume=\Q$ENV{CRUNCH_SRC}:$ENV{CRUNCH_SRC}:ro\E ";
814 $command .= "--volume=\Q$ENV{CRUNCH_INSTALL}:$ENV{CRUNCH_INSTALL}:ro\E ";
816 # Currently, we make arv-mount's mount point appear at /keep
817 # inside the container (instead of using the same path as the
818 # host like we do with CRUNCH_SRC and CRUNCH_INSTALL). However,
819 # crunch scripts and utilities must not rely on this. They must
820 # use $TASK_KEEPMOUNT.
821 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT}:/keep:ro\E ";
822 $ENV{TASK_KEEPMOUNT} = "/keep";
824 # TASK_WORK is almost exactly like a docker data volume: it
825 # starts out empty, is writable, and persists until no
826 # containers use it any more. We don't use --volumes-from to
827 # share it with other containers: it is only accessible to this
828 # task, and it goes away when this task stops.
830 # However, a docker data volume is writable only by root unless
831 # the mount point already happens to exist in the container with
832 # different permissions. Therefore, we [1] assume /tmp already
833 # exists in the image and is writable by the crunch user; [2]
834 # avoid putting TASK_WORK inside CRUNCH_TMP (which won't be
835 # writable if they are created by docker while setting up the
836 # other --volumes); and [3] create $TASK_WORK inside the
837 # container using $build_script.
838 $command .= "--volume=/tmp ";
839 $ENV{"TASK_WORK"} = "/tmp/crunch-job-task-work/$childslotname";
840 $ENV{"HOME"} = $ENV{"TASK_WORK"};
841 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
843 # TODO: Share a single JOB_WORK volume across all task
844 # containers on a given worker node, and delete it when the job
845 # ends (and, in case that doesn't work, when the next job
848 # For now, use the same approach as TASK_WORK above.
849 $ENV{"JOB_WORK"} = "/tmp/crunch-job-work";
851 while (my ($env_key, $env_val) = each %ENV)
853 if ($env_key =~ /^(ARVADOS|CRUNCH|JOB|TASK)_/) {
854 $command .= "--env=\Q$env_key=$env_val\E ";
857 $command .= "--env=\QHOME=$ENV{HOME}\E ";
858 $command .= "\Q$docker_hash\E ";
859 $command .= "stdbuf --output=0 --error=0 ";
860 $command .= "perl - $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
863 $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -poll=10000 ";
864 $command .= "stdbuf --output=0 --error=0 ";
865 $command .= "perl - $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
868 my @execargs = ('bash', '-c', $command);
869 srun (\@srunargs, \@execargs, undef, $build_script);
870 # exec() failed, we assume nothing happened.
871 die "srun() failed on build script\n";
874 if (!defined $childpid)
881 $proc{$childpid} = { jobstep => $id,
884 jobstepname => "$job_id.$id.$childpid",
886 croak ("assert failed: \$slot[$childslot]->{'pid'} exists") if exists $slot[$childslot]->{pid};
887 $slot[$childslot]->{pid} = $childpid;
889 Log ($id, "job_task ".$Jobstep->{'arvados_task'}->{'uuid'});
890 Log ($id, "child $childpid started on $childslotname");
891 $Jobstep->{starttime} = time;
892 $Jobstep->{node} = $childnode->{name};
893 $Jobstep->{slotindex} = $childslot;
894 delete $Jobstep->{stderr};
895 delete $Jobstep->{finishtime};
897 $Jobstep->{'arvados_task'}->{started_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{starttime});
898 $Jobstep->{'arvados_task'}->save;
900 splice @jobstep_todo, $todo_ptr, 1;
903 $progress_is_dirty = 1;
907 ($round_num_freeslots > @freeslot && $todo_ptr+1 > $#jobstep_todo))
909 last THISROUND if $main::please_freeze || defined($main::success);
910 if ($main::please_info)
912 $main::please_info = 0;
914 create_output_collection();
916 update_progress_stats();
923 check_refresh_wanted();
925 update_progress_stats();
926 select (undef, undef, undef, 0.1);
928 elsif (time - $progress_stats_updated >= 30)
930 update_progress_stats();
932 if (($thisround_failed_multiple >= 8 && $thisround_succeeded == 0) ||
933 ($thisround_failed_multiple >= 16 && $thisround_failed_multiple > $thisround_succeeded))
935 my $message = "Repeated failure rate too high ($thisround_failed_multiple/"
936 .($thisround_failed+$thisround_succeeded)
937 .") -- giving up on this round";
938 Log (undef, $message);
942 # move slots from freeslot to holdslot (or back to freeslot) if necessary
943 for (my $i=$#freeslot; $i>=0; $i--) {
944 if ($slot[$freeslot[$i]]->{node}->{hold_until} > scalar time) {
945 push @holdslot, (splice @freeslot, $i, 1);
948 for (my $i=$#holdslot; $i>=0; $i--) {
949 if ($slot[$holdslot[$i]]->{node}->{hold_until} <= scalar time) {
950 push @freeslot, (splice @holdslot, $i, 1);
954 # give up if no nodes are succeeding
955 if (!grep { $_->{node}->{losing_streak} == 0 &&
956 $_->{node}->{hold_count} < 4 } @slot) {
957 my $message = "Every node has failed -- giving up on this round";
958 Log (undef, $message);
965 push @freeslot, splice @holdslot;
966 map { $slot[$freeslot[$_]]->{node}->{losing_streak} = 0 } (0..$#freeslot);
969 Log (undef, "wait for last ".(scalar keys %proc)." children to finish");
972 if ($main::please_continue) {
973 $main::please_continue = 0;
976 $main::please_info = 0, freeze(), create_output_collection(), save_meta(1) if $main::please_info;
980 check_refresh_wanted();
982 update_progress_stats();
983 select (undef, undef, undef, 0.1);
984 killem (keys %proc) if $main::please_freeze;
988 update_progress_stats();
989 freeze_if_want_freeze();
992 if (!defined $main::success)
995 $thisround_succeeded == 0 &&
996 ($thisround_failed == 0 || $thisround_failed > 4))
998 my $message = "stop because $thisround_failed tasks failed and none succeeded";
999 Log (undef, $message);
1008 goto ONELEVEL if !defined $main::success;
1011 release_allocation();
1013 my $collated_output = &create_output_collection();
1015 if (!$collated_output) {
1016 Log (undef, "Failed to write output collection");
1019 Log(undef, "job output $collated_output");
1020 $Job->update_attributes('output' => $collated_output);
1023 Log (undef, "finish");
1028 if ($collated_output && $main::success) {
1029 $final_state = 'Complete';
1031 $final_state = 'Failed';
1033 $Job->update_attributes('state' => $final_state);
1035 exit (($final_state eq 'Complete') ? 0 : 1);
1039 sub update_progress_stats
1041 $progress_stats_updated = time;
1042 return if !$progress_is_dirty;
1043 my ($todo, $done, $running) = (scalar @jobstep_todo,
1044 scalar @jobstep_done,
1045 scalar @slot - scalar @freeslot - scalar @holdslot);
1046 $Job->{'tasks_summary'} ||= {};
1047 $Job->{'tasks_summary'}->{'todo'} = $todo;
1048 $Job->{'tasks_summary'}->{'done'} = $done;
1049 $Job->{'tasks_summary'}->{'running'} = $running;
1050 $Job->update_attributes('tasks_summary' => $Job->{'tasks_summary'});
1051 Log (undef, "status: $done done, $running running, $todo todo");
1052 $progress_is_dirty = 0;
1059 my $pid = waitpid (-1, WNOHANG);
1060 return 0 if $pid <= 0;
1062 my $whatslot = ($slot[$proc{$pid}->{slot}]->{node}->{name}
1064 . $slot[$proc{$pid}->{slot}]->{cpu});
1065 my $jobstepid = $proc{$pid}->{jobstep};
1066 my $elapsed = time - $proc{$pid}->{time};
1067 my $Jobstep = $jobstep[$jobstepid];
1069 my $childstatus = $?;
1070 my $exitvalue = $childstatus >> 8;
1071 my $exitinfo = "exit ".exit_status_s($childstatus);
1072 $Jobstep->{'arvados_task'}->reload;
1073 my $task_success = $Jobstep->{'arvados_task'}->{success};
1075 Log ($jobstepid, "child $pid on $whatslot $exitinfo success=$task_success");
1077 if (!defined $task_success) {
1078 # task did not indicate one way or the other --> fail
1079 $Jobstep->{'arvados_task'}->{success} = 0;
1080 $Jobstep->{'arvados_task'}->save;
1087 $temporary_fail ||= $Jobstep->{node_fail};
1088 $temporary_fail ||= ($exitvalue == TASK_TEMPFAIL);
1090 ++$thisround_failed;
1091 ++$thisround_failed_multiple if $Jobstep->{'failures'} >= 1;
1093 # Check for signs of a failed or misconfigured node
1094 if (++$slot[$proc{$pid}->{slot}]->{node}->{losing_streak} >=
1095 2+$slot[$proc{$pid}->{slot}]->{node}->{ncpus}) {
1096 # Don't count this against jobstep failure thresholds if this
1097 # node is already suspected faulty and srun exited quickly
1098 if ($slot[$proc{$pid}->{slot}]->{node}->{hold_until} &&
1100 Log ($jobstepid, "blaming failure on suspect node " .
1101 $slot[$proc{$pid}->{slot}]->{node}->{name});
1102 $temporary_fail ||= 1;
1104 ban_node_by_slot($proc{$pid}->{slot});
1107 Log ($jobstepid, sprintf('failure (#%d, %s) after %d seconds',
1108 ++$Jobstep->{'failures'},
1109 $temporary_fail ? 'temporary ' : 'permanent',
1112 if (!$temporary_fail || $Jobstep->{'failures'} >= 3) {
1113 # Give up on this task, and the whole job
1116 # Put this task back on the todo queue
1117 push @jobstep_todo, $jobstepid;
1118 $Job->{'tasks_summary'}->{'failed'}++;
1122 ++$thisround_succeeded;
1123 $slot[$proc{$pid}->{slot}]->{node}->{losing_streak} = 0;
1124 $slot[$proc{$pid}->{slot}]->{node}->{hold_until} = 0;
1125 push @jobstep_done, $jobstepid;
1126 Log ($jobstepid, "success in $elapsed seconds");
1128 $Jobstep->{exitcode} = $childstatus;
1129 $Jobstep->{finishtime} = time;
1130 $Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
1131 $Jobstep->{'arvados_task'}->save;
1132 process_stderr ($jobstepid, $task_success);
1133 Log ($jobstepid, sprintf("task output (%d bytes): %s",
1134 length($Jobstep->{'arvados_task'}->{output}),
1135 $Jobstep->{'arvados_task'}->{output}));
1137 close $reader{$jobstepid};
1138 delete $reader{$jobstepid};
1139 delete $slot[$proc{$pid}->{slot}]->{pid};
1140 push @freeslot, $proc{$pid}->{slot};
1143 if ($task_success) {
1145 my $newtask_list = [];
1146 my $newtask_results;
1148 $newtask_results = api_call(
1151 'created_by_job_task_uuid' => $Jobstep->{'arvados_task'}->{uuid}
1153 'order' => 'qsequence',
1154 'offset' => scalar(@$newtask_list),
1156 push(@$newtask_list, @{$newtask_results->{items}});
1157 } while (@{$newtask_results->{items}});
1158 foreach my $arvados_task (@$newtask_list) {
1160 'level' => $arvados_task->{'sequence'},
1162 'arvados_task' => $arvados_task
1164 push @jobstep, $jobstep;
1165 push @jobstep_todo, $#jobstep;
1169 $progress_is_dirty = 1;
1173 sub check_refresh_wanted
1175 my @stat = stat $ENV{"CRUNCH_REFRESH_TRIGGER"};
1176 if (@stat && $stat[9] > $latest_refresh) {
1177 $latest_refresh = scalar time;
1178 my $Job2 = api_call("jobs/get", uuid => $jobspec);
1179 for my $attr ('cancelled_at',
1180 'cancelled_by_user_uuid',
1181 'cancelled_by_client_uuid',
1183 $Job->{$attr} = $Job2->{$attr};
1185 if ($Job->{'state'} ne "Running") {
1186 if ($Job->{'state'} eq "Cancelled") {
1187 Log (undef, "Job cancelled at " . $Job->{'cancelled_at'} . " by user " . $Job->{'cancelled_by_user_uuid'});
1189 Log (undef, "Job state unexpectedly changed to " . $Job->{'state'});
1192 $main::please_freeze = 1;
1199 # return if the kill list was checked <4 seconds ago
1200 if (defined $squeue_kill_checked && $squeue_kill_checked > time - 4)
1204 $squeue_kill_checked = time;
1206 # use killem() on procs whose killtime is reached
1209 if (exists $proc{$_}->{killtime}
1210 && $proc{$_}->{killtime} <= time)
1216 # return if the squeue was checked <60 seconds ago
1217 if (defined $squeue_checked && $squeue_checked > time - 60)
1221 $squeue_checked = time;
1225 # here is an opportunity to check for mysterious problems with local procs
1229 # get a list of steps still running
1230 my @squeue = `squeue -s -h -o '%i %j' && echo ok`;
1232 if ($squeue[-1] ne "ok")
1238 # which of my jobsteps are running, according to squeue?
1242 if (/^(\d+)\.(\d+) (\S+)/)
1244 if ($1 eq $ENV{SLURM_JOBID})
1251 # which of my active child procs (>60s old) were not mentioned by squeue?
1252 foreach (keys %proc)
1254 if ($proc{$_}->{time} < time - 60
1255 && !exists $ok{$proc{$_}->{jobstepname}}
1256 && !exists $proc{$_}->{killtime})
1258 # kill this proc if it hasn't exited in 30 seconds
1259 $proc{$_}->{killtime} = time + 30;
1265 sub release_allocation
1269 Log (undef, "release job allocation");
1270 system "scancel $ENV{SLURM_JOBID}";
1278 foreach my $job (keys %reader)
1281 while (0 < sysread ($reader{$job}, $buf, 8192))
1283 print STDERR $buf if $ENV{CRUNCH_DEBUG};
1284 $jobstep[$job]->{stderr} .= $buf;
1285 preprocess_stderr ($job);
1286 if (length ($jobstep[$job]->{stderr}) > 16384)
1288 substr ($jobstep[$job]->{stderr}, 0, 8192) = "";
1297 sub preprocess_stderr
1301 while ($jobstep[$job]->{stderr} =~ /^(.*?)\n/) {
1303 substr $jobstep[$job]->{stderr}, 0, 1+length($line), "";
1304 Log ($job, "stderr $line");
1305 if ($line =~ /srun: error: (SLURM job $ENV{SLURM_JOB_ID} has expired|Unable to confirm allocation for job $ENV{SLURM_JOB_ID})/) {
1307 $main::please_freeze = 1;
1309 elsif ($line =~ /(srun: error: (Node failure on|Unable to create job step|.*: Communication connection failure))|arvados.errors.Keep/) {
1310 $jobstep[$job]->{node_fail} = 1;
1311 ban_node_by_slot($jobstep[$job]->{slotindex});
1320 my $task_success = shift;
1321 preprocess_stderr ($job);
1324 Log ($job, "stderr $_");
1325 } split ("\n", $jobstep[$job]->{stderr});
1332 if (!open($keep, "-|", "arv-get", "--retries", retry_count(), $hash)) {
1333 Log(undef, "fetch_block run error from arv-get $hash: $!");
1336 my $output_block = "";
1339 my $bytes = sysread($keep, $buf, 1024 * 1024);
1340 if (!defined $bytes) {
1341 Log(undef, "fetch_block read error from arv-get: $!");
1342 $output_block = undef;
1344 } elsif ($bytes == 0) {
1345 # sysread returns 0 at the end of the pipe.
1348 # some bytes were read into buf.
1349 $output_block .= $buf;
1354 Log(undef, "fetch_block arv-get exited " . exit_status_s($?));
1355 $output_block = undef;
1357 return $output_block;
1360 # Create a collection by concatenating the output of all tasks (each
1361 # task's output is either a manifest fragment, a locator for a
1362 # manifest fragment stored in Keep, or nothing at all). Return the
1363 # portable_data_hash of the new collection.
1364 sub create_output_collection
1366 Log (undef, "collate");
1368 my ($child_out, $child_in);
1369 my $pid = open2($child_out, $child_in, 'python', '-c', q{
1372 print (arvados.api("v1").collections().
1373 create(body={"manifest_text": sys.stdin.read()}).
1374 execute(num_retries=int(sys.argv[1]))["portable_data_hash"])
1378 my $manifest_size = 0;
1382 my $output = $_->{'arvados_task'}->{output};
1383 next if (!defined($output));
1385 if ($output =~ /^[0-9a-f]{32}(\+\S+)*$/) {
1386 $next_write = fetch_block($output);
1388 $next_write = $output;
1390 if (defined($next_write)) {
1391 if (!defined(syswrite($child_in, $next_write))) {
1392 # There's been an error writing. Stop the loop.
1393 # We'll log details about the exit code later.
1396 $manifest_size += length($next_write);
1399 my $uuid = $_->{'arvados_task'}->{'uuid'};
1400 Log (undef, "Error retrieving '$output' output by task $task_idx ($uuid)");
1405 Log(undef, "collated output manifest text to send to API server is $manifest_size bytes with access tokens");
1408 my $s = IO::Select->new($child_out);
1409 if ($s->can_read(120)) {
1410 sysread($child_out, $joboutput, 1024 * 1024);
1413 Log(undef, "output collection creation exited " . exit_status_s($?));
1419 Log (undef, "timed out while creating output collection");
1420 foreach my $signal (2, 2, 2, 15, 15, 9) {
1421 kill($signal, $pid);
1422 last if waitpid($pid, WNOHANG) == -1;
1436 my $sig = 2; # SIGINT first
1437 if (exists $proc{$_}->{"sent_$sig"} &&
1438 time - $proc{$_}->{"sent_$sig"} > 4)
1440 $sig = 15; # SIGTERM if SIGINT doesn't work
1442 if (exists $proc{$_}->{"sent_$sig"} &&
1443 time - $proc{$_}->{"sent_$sig"} > 4)
1445 $sig = 9; # SIGKILL if SIGTERM doesn't work
1447 if (!exists $proc{$_}->{"sent_$sig"})
1449 Log ($proc{$_}->{jobstep}, "sending 2x signal $sig to pid $_");
1451 select (undef, undef, undef, 0.1);
1454 kill $sig, $_; # srun wants two SIGINT to really interrupt
1456 $proc{$_}->{"sent_$sig"} = time;
1457 $proc{$_}->{"killedafter"} = time - $proc{$_}->{"time"};
1467 vec($bits,fileno($_),1) = 1;
1473 # Send log output to Keep via arv-put.
1475 # $log_pipe_in and $log_pipe_out are the input and output filehandles to the arv-put pipe.
1476 # $log_pipe_pid is the pid of the arv-put subprocess.
1478 # The only functions that should access these variables directly are:
1480 # log_writer_start($logfilename)
1481 # Starts an arv-put pipe, reading data on stdin and writing it to
1482 # a $logfilename file in an output collection.
1484 # log_writer_send($txt)
1485 # Writes $txt to the output log collection.
1487 # log_writer_finish()
1488 # Closes the arv-put pipe and returns the output that it produces.
1490 # log_writer_is_active()
1491 # Returns a true value if there is currently a live arv-put
1492 # process, false otherwise.
1494 my ($log_pipe_in, $log_pipe_out, $log_pipe_pid);
1496 sub log_writer_start($)
1498 my $logfilename = shift;
1499 $log_pipe_pid = open2($log_pipe_out, $log_pipe_in,
1501 '--portable-data-hash',
1502 '--project-uuid', $Job->{owner_uuid},
1504 '--name', $logfilename,
1505 '--filename', $logfilename,
1509 sub log_writer_send($)
1512 print $log_pipe_in $txt;
1515 sub log_writer_finish()
1517 return unless $log_pipe_pid;
1519 close($log_pipe_in);
1522 my $s = IO::Select->new($log_pipe_out);
1523 if ($s->can_read(120)) {
1524 sysread($log_pipe_out, $arv_put_output, 1024);
1525 chomp($arv_put_output);
1527 Log (undef, "timed out reading from 'arv-put'");
1530 waitpid($log_pipe_pid, 0);
1531 $log_pipe_pid = $log_pipe_in = $log_pipe_out = undef;
1533 Log("log_writer_finish: arv-put exited ".exit_status_s($?))
1536 return $arv_put_output;
1539 sub log_writer_is_active() {
1540 return $log_pipe_pid;
1543 sub Log # ($jobstep_id, $logmessage)
1545 if ($_[1] =~ /\n/) {
1546 for my $line (split (/\n/, $_[1])) {
1551 my $fh = select STDERR; $|=1; select $fh;
1552 my $message = sprintf ("%s %d %s %s", $job_id, $$, @_);
1553 $message =~ s{([^ -\176])}{"\\" . sprintf ("%03o", ord($1))}ge;
1556 if (log_writer_is_active() || -t STDERR) {
1557 my @gmtime = gmtime;
1558 $datetime = sprintf ("%04d-%02d-%02d_%02d:%02d:%02d",
1559 $gmtime[5]+1900, $gmtime[4]+1, @gmtime[3,2,1,0]);
1561 print STDERR ((-t STDERR) ? ($datetime." ".$message) : $message);
1563 if (log_writer_is_active()) {
1564 log_writer_send($datetime . " " . $message);
1571 my ($package, $file, $line) = caller;
1572 my $message = "@_ at $file line $line\n";
1573 Log (undef, $message);
1574 freeze() if @jobstep_todo;
1575 create_output_collection() if @jobstep_todo;
1585 if ($Job->{'state'} eq 'Cancelled') {
1586 $Job->update_attributes('finished_at' => scalar gmtime);
1588 $Job->update_attributes('state' => 'Failed');
1595 my $justcheckpoint = shift; # false if this will be the last meta saved
1596 return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm
1597 return unless log_writer_is_active();
1599 my $loglocator = log_writer_finish();
1600 Log (undef, "log manifest is $loglocator");
1601 $Job->{'log'} = $loglocator;
1602 $Job->update_attributes('log', $loglocator);
1606 sub freeze_if_want_freeze
1608 if ($main::please_freeze)
1610 release_allocation();
1613 # kill some srun procs before freeze+stop
1614 map { $proc{$_} = {} } @_;
1617 killem (keys %proc);
1618 select (undef, undef, undef, 0.1);
1620 while (($died = waitpid (-1, WNOHANG)) > 0)
1622 delete $proc{$died};
1627 create_output_collection();
1637 Log (undef, "Freeze not implemented");
1644 croak ("Thaw not implemented");
1660 $s =~ s{\\(.)}{$1 eq "n" ? "\n" : $1}ge;
1667 my $srunargs = shift;
1668 my $execargs = shift;
1669 my $opts = shift || {};
1671 my $args = $have_slurm ? [@$srunargs, @$execargs] : $execargs;
1673 $Data::Dumper::Terse = 1;
1674 $Data::Dumper::Indent = 0;
1675 my $show_cmd = Dumper($args);
1676 $show_cmd =~ s/(TOKEN\\*=)[^\s\']+/${1}[...]/g;
1677 $show_cmd =~ s/\n/ /g;
1678 warn "starting: $show_cmd\n";
1680 if (defined $stdin) {
1681 my $child = open STDIN, "-|";
1682 defined $child or die "no fork: $!";
1684 print $stdin or die $!;
1685 close STDOUT or die $!;
1690 return system (@$args) if $opts->{fork};
1693 warn "ENV size is ".length(join(" ",%ENV));
1694 die "exec failed: $!: @$args";
1698 sub ban_node_by_slot {
1699 # Don't start any new jobsteps on this node for 60 seconds
1701 $slot[$slotid]->{node}->{hold_until} = 60 + scalar time;
1702 $slot[$slotid]->{node}->{hold_count}++;
1703 Log (undef, "backing off node " . $slot[$slotid]->{node}->{name} . " for 60 seconds");
1708 my ($lockfile, $error_message) = @_;
1709 open L, ">", $lockfile or croak("$lockfile: $!");
1710 if (!flock L, LOCK_EX|LOCK_NB) {
1711 croak("Can't lock $lockfile: $error_message\n");
1715 sub find_docker_image {
1716 # Given a Keep locator, check to see if it contains a Docker image.
1717 # If so, return its stream name and Docker hash.
1718 # If not, return undef for both values.
1719 my $locator = shift;
1720 my ($streamname, $filename);
1721 my $image = api_call("collections/get", uuid => $locator);
1723 foreach my $line (split(/\n/, $image->{manifest_text})) {
1724 my @tokens = split(/\s+/, $line);
1726 $streamname = shift(@tokens);
1727 foreach my $filedata (grep(/^\d+:\d+:/, @tokens)) {
1728 if (defined($filename)) {
1729 return (undef, undef); # More than one file in the Collection.
1731 $filename = (split(/:/, $filedata, 3))[2];
1736 if (defined($filename) and ($filename =~ /^([0-9A-Fa-f]{64})\.tar$/)) {
1737 return ($streamname, $1);
1739 return (undef, undef);
1744 # Calculate the number of times an operation should be retried,
1745 # assuming exponential backoff, and that we're willing to retry as
1746 # long as tasks have been running. Enforce a minimum of 3 retries.
1747 my ($starttime, $endtime, $timediff, $retries);
1749 $starttime = $jobstep[0]->{starttime};
1750 $endtime = $jobstep[-1]->{finishtime};
1752 if (!defined($starttime)) {
1754 } elsif (!defined($endtime)) {
1755 $timediff = time - $starttime;
1757 $timediff = ($endtime - $starttime) - (time - $endtime);
1759 if ($timediff > 0) {
1760 $retries = int(log($timediff) / log(2));
1762 $retries = 1; # Use the minimum.
1764 return ($retries > 3) ? $retries : 3;
1768 # Pass in two function references.
1769 # This method will be called with the remaining arguments.
1770 # If it dies, retry it with exponential backoff until it succeeds,
1771 # or until the current retry_count is exhausted. After each failure
1772 # that can be retried, the second function will be called with
1773 # the current try count (0-based), next try time, and error message.
1774 my $operation = shift;
1775 my $retry_callback = shift;
1776 my $retries = retry_count();
1777 foreach my $try_count (0..$retries) {
1778 my $next_try = time + (2 ** $try_count);
1779 my $result = eval { $operation->(@_); };
1782 } elsif ($try_count < $retries) {
1783 $retry_callback->($try_count, $next_try, $@);
1784 my $sleep_time = $next_try - time;
1785 sleep($sleep_time) if ($sleep_time > 0);
1788 # Ensure the error message ends in a newline, so Perl doesn't add
1789 # retry_op's line number to it.
1795 # Pass in a /-separated API method name, and arguments for it.
1796 # This function will call that method, retrying as needed until
1797 # the current retry_count is exhausted, with a log on the first failure.
1798 my $method_name = shift;
1799 my $log_api_retry = sub {
1800 my ($try_count, $next_try_at, $errmsg) = @_;
1801 $errmsg =~ s/\s*\bat \Q$0\E line \d+\.?\s*//;
1802 $errmsg =~ s/\s/ /g;
1803 $errmsg =~ s/\s+$//;
1805 if ($next_try_at < time) {
1806 $retry_msg = "Retrying.";
1808 my $next_try_fmt = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($next_try_at);
1809 $retry_msg = "Retrying at $next_try_fmt.";
1811 Log(undef, "API method $method_name failed: $errmsg. $retry_msg");
1814 foreach my $key (split(/\//, $method_name)) {
1815 $method = $method->{$key};
1817 return retry_op(sub { $method->execute(@_); }, $log_api_retry, @_);
1821 # Given a $?, return a human-readable exit code string like "0" or
1822 # "1" or "0 with signal 1" or "1 with signal 11".
1823 my $exitcode = shift;
1824 my $s = $exitcode >> 8;
1825 if ($exitcode & 0x7f) {
1826 $s .= " with signal " . ($exitcode & 0x7f);
1828 if ($exitcode & 0x80) {
1829 $s .= " with core dump";
1834 sub handle_readall {
1835 # Pass in a glob reference to a file handle.
1836 # Read all its contents and return them as a string.
1837 my $fh_glob_ref = shift;
1839 return <$fh_glob_ref>;
1842 sub tar_filename_n {
1844 return sprintf("%s/git.%s.%d.tar", $ENV{CRUNCH_TMP}, $job_id, $n);
1847 sub add_git_archive {
1848 # Pass in a git archive command as a string or list, a la system().
1849 # This method will save its output to be included in the archive sent to the
1853 if (!open(GIT_ARCHIVE, ">", tar_filename_n($git_tar_count))) {
1854 croak("Failed to save git archive: $!");
1856 my $git_pid = open2(">&GIT_ARCHIVE", $git_input, @_);
1858 waitpid($git_pid, 0);
1861 croak("Failed to save git archive: git exited " . exit_status_s($?));
1865 sub combined_git_archive {
1866 # Combine all saved tar archives into a single archive, then return its
1867 # contents in a string. Return undef if no archives have been saved.
1868 if ($git_tar_count < 1) {
1871 my $base_tar_name = tar_filename_n(1);
1872 foreach my $tar_to_append (map { tar_filename_n($_); } (2..$git_tar_count)) {
1873 my $tar_exit = system("tar", "-Af", $base_tar_name, $tar_to_append);
1874 if ($tar_exit != 0) {
1875 croak("Error preparing build archive: tar -A exited " .
1876 exit_status_s($tar_exit));
1879 if (!open(GIT_TAR, "<", $base_tar_name)) {
1880 croak("Could not open build archive: $!");
1882 my $tar_contents = handle_readall(\*GIT_TAR);
1884 return $tar_contents;
1887 sub set_nonblocking {
1889 my $flags = fcntl ($fh, F_GETFL, 0) or croak ($!);
1890 fcntl ($fh, F_SETFL, $flags | O_NONBLOCK) or croak ($!);
1896 # This is crunch-job's internal dispatch script. crunch-job running on the API
1897 # server invokes this script on individual compute nodes, or localhost if we're
1898 # running a job locally. It gets called in two modes:
1900 # * No arguments: Installation mode. Read a tar archive from the DATA
1901 # file handle; it includes the Crunch script's source code, and
1902 # maybe SDKs as well. Those should be installed in the proper
1903 # locations. This runs outside of any Docker container, so don't try to
1904 # introspect Crunch's runtime environment.
1906 # * With arguments: Crunch script run mode. This script should set up the
1907 # environment, then run the command specified in the arguments. This runs
1908 # inside any Docker container.
1911 use File::Path qw( make_path remove_tree );
1912 use POSIX qw(getcwd);
1914 use constant TASK_TEMPFAIL => 111;
1916 # Map SDK subdirectories to the path environments they belong to.
1917 my %SDK_ENVVARS = ("perl/lib" => "PERLLIB", "ruby/lib" => "RUBYLIB");
1919 my $destdir = $ENV{"CRUNCH_SRC"};
1920 my $archive_hash = $ENV{"CRUNCH_GIT_ARCHIVE_HASH"};
1921 my $repo = $ENV{"CRUNCH_SRC_URL"};
1922 my $install_dir = $ENV{"CRUNCH_INSTALL"} || (getcwd() . "/opt");
1923 my $job_work = $ENV{"JOB_WORK"};
1924 my $task_work = $ENV{"TASK_WORK"};
1926 open(STDOUT_ORIG, ">&", STDOUT);
1927 open(STDERR_ORIG, ">&", STDERR);
1929 for my $dir ($destdir, $job_work, $task_work) {
1932 -e $dir or die "Failed to create temporary directory ($dir): $!";
1937 remove_tree($task_work, {keep_root => 1});
1940 ### Crunch script run mode
1942 # We want to do routine logging during task 0 only. This gives the user
1943 # the information they need, but avoids repeating the information for every
1946 if ($ENV{TASK_SEQUENCE} eq "0") {
1949 printf STDERR_ORIG "[Crunch] $msg\n", @_;
1955 my $python_src = "$install_dir/python";
1956 my $venv_dir = "$job_work/.arvados.venv";
1957 my $venv_built = -e "$venv_dir/bin/activate";
1958 if ((!$venv_built) and (-d $python_src) and can_run("virtualenv")) {
1959 shell_or_die(undef, "virtualenv", "--quiet", "--system-site-packages",
1960 "--python=python2.7", $venv_dir);
1961 shell_or_die(TASK_TEMPFAIL, "$venv_dir/bin/pip", "--quiet", "install", "-I", $python_src);
1963 $Log->("Built Python SDK virtualenv");
1966 my $pip_bin = "pip";
1968 $Log->("Running in Python SDK virtualenv");
1969 $pip_bin = "$venv_dir/bin/pip";
1970 my $orig_argv = join(" ", map { quotemeta($_); } @ARGV);
1971 @ARGV = ("/bin/sh", "-ec",
1972 ". \Q$venv_dir/bin/activate\E; exec $orig_argv");
1973 } elsif (-d $python_src) {
1974 $Log->("Warning: virtualenv not found inside Docker container default " .
1975 "\$PATH. Can't install Python SDK.");
1978 my $pkgs = `(\Q$pip_bin\E freeze 2>/dev/null | grep arvados) || dpkg-query --show '*arvados*'`;
1980 $Log->("Using Arvados SDK:");
1981 foreach my $line (split /\n/, $pkgs) {
1985 $Log->("Arvados SDK packages not found");
1988 while (my ($sdk_dir, $sdk_envkey) = each(%SDK_ENVVARS)) {
1989 my $sdk_path = "$install_dir/$sdk_dir";
1991 if ($ENV{$sdk_envkey}) {
1992 $ENV{$sdk_envkey} = "$sdk_path:" . $ENV{$sdk_envkey};
1994 $ENV{$sdk_envkey} = $sdk_path;
1996 $Log->("Arvados SDK added to %s", $sdk_envkey);
2001 die "Cannot exec `@ARGV`: $!";
2004 ### Installation mode
2005 open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
2007 if (readlink ("$destdir.archive_hash") eq $archive_hash && -d $destdir) {
2008 # This exact git archive (source + arvados sdk) is already installed
2009 # here, so there's no need to reinstall it.
2011 # We must consume our DATA section, though: otherwise the process
2012 # feeding it to us will get SIGPIPE.
2014 while (read(DATA, $buf, 65536)) { }
2019 unlink "$destdir.archive_hash";
2022 if (!open(TARX, "|-", "tar", "-xC", $destdir)) {
2023 die "Error launching 'tar -xC $destdir': $!";
2025 # If we send too much data to tar in one write (> 4-5 MiB), it stops, and we
2026 # get SIGPIPE. We must feed it data incrementally.
2028 while (read(DATA, $tar_input, 65536)) {
2029 print TARX $tar_input;
2032 die "'tar -xC $destdir' exited $?: $!";
2037 my $sdk_root = "$destdir/.arvados.sdk/sdk";
2039 foreach my $sdk_lang (("python",
2040 map { (split /\//, $_, 2)[0]; } keys(%SDK_ENVVARS))) {
2041 if (-d "$sdk_root/$sdk_lang") {
2042 if (!rename("$sdk_root/$sdk_lang", "$install_dir/$sdk_lang")) {
2043 die "Failed to install $sdk_lang SDK: $!";
2049 my $python_dir = "$install_dir/python";
2050 if ((-d $python_dir) and can_run("python2.7") and
2051 (system("python2.7", "$python_dir/setup.py", "--quiet", "egg_info") != 0)) {
2052 # egg_info failed, probably when it asked git for a build tag.
2053 # Specify no build tag.
2054 open(my $pysdk_cfg, ">>", "$python_dir/setup.cfg");
2055 print $pysdk_cfg "\n[egg_info]\ntag_build =\n";
2059 # Hide messages from the install script (unless it fails: shell_or_die
2060 # will show $destdir.log in that case).
2061 open(STDOUT, ">>", "$destdir.log");
2062 open(STDERR, ">&", STDOUT);
2064 if (-e "$destdir/crunch_scripts/install") {
2065 shell_or_die (undef, "$destdir/crunch_scripts/install", $install_dir);
2066 } elsif (!-e "./install.sh" && -e "./tests/autotests.sh") {
2068 shell_or_die (undef, "./tests/autotests.sh", $install_dir);
2069 } elsif (-e "./install.sh") {
2070 shell_or_die (undef, "./install.sh", $install_dir);
2073 if ($archive_hash) {
2074 unlink "$destdir.archive_hash.new";
2075 symlink ($archive_hash, "$destdir.archive_hash.new") or die "$destdir.archive_hash.new: $!";
2076 rename ("$destdir.archive_hash.new", "$destdir.archive_hash") or die "$destdir.archive_hash: $!";
2082 my $command_name = shift;
2083 open(my $which, "-|", "which", $command_name);
2084 while (<$which>) { }
2091 my $exitcode = shift;
2093 if ($ENV{"DEBUG"}) {
2094 print STDERR "@_\n";
2096 if (system (@_) != 0) {
2099 my $exitstatus = sprintf("exit %d signal %d", $code >> 8, $code & 0x7f);
2100 open STDERR, ">&STDERR_ORIG";
2101 system ("cat $destdir.log >&2");
2102 warn "@_ failed ($err): $exitstatus";
2103 if (defined($exitcode)) {
2107 exit (($code >> 8) || 1);