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 $ENV{"HOST_CRUNCHRUNNER_BIN"} ||= `which crunchrunner`;
113 unless (defined($ENV{"HOST_CERTS"})) {
114 if (-f "/etc/ssl/certs/ca-certificates.crt") {
115 $ENV{"HOST_CERTS"} = "/etc/ssl/certs/ca-certificates.crt";
116 } elsif (-f "/etc/pki/tls/certs/ca-bundle.crt") {
117 $ENV{"HOST_CERTS"} = "/etc/pki/tls/certs/ca-bundle.crt";
121 # Create the tmp directory if it does not exist
122 if ( ! -d $ENV{"CRUNCH_TMP"} ) {
123 make_path $ENV{"CRUNCH_TMP"} or die "Failed to create temporary working directory: " . $ENV{"CRUNCH_TMP"};
126 $ENV{"JOB_WORK"} = $ENV{"CRUNCH_TMP"} . "/work";
127 $ENV{"CRUNCH_INSTALL"} = "$ENV{CRUNCH_TMP}/opt";
128 $ENV{"CRUNCH_WORK"} = $ENV{"JOB_WORK"}; # deprecated
129 mkdir ($ENV{"JOB_WORK"});
138 my $cgroup_root = "/sys/fs/cgroup";
139 my $docker_bin = "docker.io";
140 my $docker_run_args = "";
141 GetOptions('force-unlock' => \$force_unlock,
142 'git-dir=s' => \$git_dir,
143 'job=s' => \$jobspec,
144 'job-api-token=s' => \$job_api_token,
145 'no-clear-tmp' => \$no_clear_tmp,
146 'resume-stash=s' => \$resume_stash,
147 'cgroup-root=s' => \$cgroup_root,
148 'docker-bin=s' => \$docker_bin,
149 'docker-run-args=s' => \$docker_run_args,
152 if (defined $job_api_token) {
153 $ENV{ARVADOS_API_TOKEN} = $job_api_token;
156 my $have_slurm = exists $ENV{SLURM_JOB_ID} && exists $ENV{SLURM_NODELIST};
161 $main::ENV{CRUNCH_DEBUG} = 1;
165 $main::ENV{CRUNCH_DEBUG} = 0;
168 my $arv = Arvados->new('apiVersion' => 'v1');
177 if ($jobspec =~ /^[-a-z\d]+$/)
179 # $jobspec is an Arvados UUID, not a JSON job specification
180 $Job = api_call("jobs/get", uuid => $jobspec);
181 $job_id = $Job->{'uuid'};
182 log_writer_start("${job_id}.log.txt");
187 $local_job = JSON::decode_json($jobspec);
191 # Make sure our workers (our slurm nodes, localhost, or whatever) are
192 # at least able to run basic commands: they aren't down or severely
195 if (($Job || $local_job)->{docker_image_locator}) {
196 $cmd = [$docker_bin, 'ps', '-q'];
198 Log(undef, "Sanity check is `@$cmd`");
199 my ($exited, $stdout, $stderr) = srun_sync(
200 ["srun", "--nodes=\Q$ENV{SLURM_NNODES}\E", "--ntasks-per-node=1"],
202 {label => "sanity check"});
204 Log(undef, "Sanity check failed: ".exit_status_s($exited));
208 Log(undef, "Sanity check OK");
211 my $User = api_call("users/current");
214 if (!$force_unlock) {
215 # Claim this job, and make sure nobody else does
216 eval { api_call("jobs/lock", uuid => $Job->{uuid}); };
218 Log(undef, "Error while locking job, exiting ".EX_TEMPFAIL);
228 map { croak ("No $_ specified") unless $local_job->{$_} }
229 qw(script script_version script_parameters);
232 $local_job->{'is_locked_by_uuid'} = $User->{'uuid'};
233 $local_job->{'started_at'} = gmtime;
234 $local_job->{'state'} = 'Running';
236 $Job = api_call("jobs/create", job => $local_job);
237 $job_id = $Job->{'uuid'};
238 log_writer_start("${job_id}.log.txt");
241 $Job->{'runtime_constraints'} ||= {};
242 $Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0;
243 my $max_ncpus = $Job->{'runtime_constraints'}->{'max_tasks_per_node'};
245 my $gem_versions = `gem list --quiet arvados-cli 2>/dev/null`;
247 $gem_versions =~ s/^arvados-cli \(/ with arvados-cli Gem version(s) /;
248 chomp($gem_versions);
249 chop($gem_versions); # Closing parentheses
254 "running from " . ((-e $0) ? realpath($0) : "stdin") . $gem_versions);
256 Log (undef, "check slurm allocation");
259 # Should use $ENV{SLURM_TASKS_PER_NODE} instead of sinfo? (eg. "4(x3),2,4(x2)")
263 my $localcpus = 0 + `grep -cw ^processor /proc/cpuinfo` || 1;
264 push @sinfo, "$localcpus localhost";
266 if (exists $ENV{SLURM_NODELIST})
268 push @sinfo, `sinfo -h --format='%c %N' --nodes=\Q$ENV{SLURM_NODELIST}\E`;
272 my ($ncpus, $slurm_nodelist) = split;
273 $ncpus = $max_ncpus if $max_ncpus && $ncpus > $max_ncpus;
276 while ($slurm_nodelist =~ s/^([^\[,]+?(\[.*?\])?)(,|$)//)
279 if ($nodelist =~ /\[((\d+)(-(\d+))?(,(\d+)(-(\d+))?)*)\]/)
282 foreach (split (",", $ranges))
295 push @nodelist, map {
297 $n =~ s/\[[-,\d]+\]/$_/;
304 push @nodelist, $nodelist;
307 foreach my $nodename (@nodelist)
309 Log (undef, "node $nodename - $ncpus slots");
310 my $node = { name => $nodename,
312 # The number of consecutive times a task has been dispatched
313 # to this node and failed.
315 # The number of consecutive times that SLURM has reported
316 # a node failure since the last successful task.
318 # Don't dispatch work to this node until this time
319 # (in seconds since the epoch) has passed.
321 foreach my $cpu (1..$ncpus)
323 push @slot, { node => $node,
327 push @node, @nodelist;
332 # Ensure that we get one jobstep running on each allocated node before
333 # we start overloading nodes with concurrent steps
335 @slot = sort { $a->{cpu} <=> $b->{cpu} } @slot;
338 $Job->update_attributes(
339 'tasks_summary' => { 'failed' => 0,
344 Log (undef, "start");
345 $SIG{'INT'} = sub { $main::please_freeze = 1; };
346 $SIG{'QUIT'} = sub { $main::please_freeze = 1; };
347 $SIG{'TERM'} = \&croak;
348 $SIG{'TSTP'} = sub { $main::please_freeze = 1; };
349 $SIG{'ALRM'} = sub { $main::please_info = 1; };
350 $SIG{'CONT'} = sub { $main::please_continue = 1; };
351 $SIG{'HUP'} = sub { $main::please_refresh = 1; };
353 $main::please_freeze = 0;
354 $main::please_info = 0;
355 $main::please_continue = 0;
356 $main::please_refresh = 0;
357 my $jobsteps_must_output_keys = 0; # becomes 1 when any task outputs a key
359 grep { $ENV{$1} = $2 if /^(NOCACHE.*?)=(.*)/ } split ("\n", $$Job{knobs});
360 $ENV{"CRUNCH_JOB_UUID"} = $job_id;
361 $ENV{"JOB_UUID"} = $job_id;
364 my @jobstep_todo = ();
365 my @jobstep_done = ();
366 my @jobstep_tomerge = ();
367 my $jobstep_tomerge_level = 0;
368 my $squeue_checked = 0;
369 my $latest_refresh = scalar time;
373 if (defined $Job->{thawedfromkey})
375 thaw ($Job->{thawedfromkey});
379 my $first_task = api_call("job_tasks/create", job_task => {
380 'job_uuid' => $Job->{'uuid'},
385 push @jobstep, { 'level' => 0,
387 'arvados_task' => $first_task,
389 push @jobstep_todo, 0;
395 must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
398 my $build_script = handle_readall(\*DATA);
399 my $nodelist = join(",", @node);
400 my $git_tar_count = 0;
402 if (!defined $no_clear_tmp) {
403 # Find FUSE mounts under $CRUNCH_TMP and unmount them. Then clean
404 # up work directories crunch_tmp/work, crunch_tmp/opt,
407 # TODO: When #5036 is done and widely deployed, we can limit mount's
408 # -t option to simply fuse.keep.
409 my ($exited, $stdout, $stderr) = srun_sync(
410 ["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
411 ['bash', '-ec', '-o', 'pipefail', 'mount -t fuse,fuse.keep | awk "(index(\$3, \"$CRUNCH_TMP\") == 1){print \$3}" | xargs -r -n 1 fusermount -u -z; sleep 1; rm -rf $JOB_WORK $CRUNCH_INSTALL $CRUNCH_TMP/task $CRUNCH_TMP/src* $CRUNCH_TMP/*.cid'],
412 {label => "clean work dirs"});
415 exit(EX_RETRY_UNLOCKED);
419 # If this job requires a Docker image, install that.
420 my ($docker_locator, $docker_stream, $docker_hash, $docker_limitmem, $dockeruserarg);
421 if ($docker_locator = $Job->{docker_image_locator}) {
422 Log (undef, "Install docker image $docker_locator");
423 ($docker_stream, $docker_hash) = find_docker_image($docker_locator);
426 croak("No Docker image hash found from locator $docker_locator");
428 Log (undef, "docker image hash is $docker_hash");
429 $docker_stream =~ s/^\.//;
430 my $docker_install_script = qq{
431 if ! $docker_bin images -q --no-trunc --all | grep -qxF \Q$docker_hash\E; then
432 arv-get \Q$docker_locator$docker_stream/$docker_hash.tar\E | $docker_bin load
436 my ($exited, $stdout, $stderr) = srun_sync(
437 ["srun", "--nodelist=" . join(',', @node)],
438 ["/bin/bash", "-o", "pipefail", "-ec", $docker_install_script],
439 {label => "load docker image"});
443 exit(EX_RETRY_UNLOCKED);
446 # Determine whether this version of Docker supports memory+swap limits.
447 ($exited, $stdout, $stderr) = srun_sync(
448 ["srun", "--nodes=1"],
449 [$docker_bin, 'run', '--help'],
450 {label => "check --memory-swap feature"});
451 $docker_limitmem = ($stdout =~ /--memory-swap/);
453 # Find a non-root Docker user to use.
454 # Tries the default user for the container, then 'crunch', then 'nobody',
455 # testing for whether the actual user id is non-zero. This defends against
456 # mistakes but not malice, but we intend to harden the security in the future
457 # so we don't want anyone getting used to their jobs running as root in their
459 my @tryusers = ("", "crunch", "nobody");
460 foreach my $try_user (@tryusers) {
463 if ($try_user eq "") {
464 $label = "check whether default user is UID 0";
467 $label = "check whether user '$try_user' is UID 0";
468 $try_user_arg = "--user=$try_user";
470 my ($exited, $stdout, $stderr) = srun_sync(
471 ["srun", "--nodes=1"],
473 "$docker_bin run $docker_run_args $try_user_arg $docker_hash id --user"],
476 if ($exited == 0 && $stdout =~ /^\d+$/ && $stdout > 0) {
477 $dockeruserarg = $try_user_arg;
478 if ($try_user eq "") {
479 Log(undef, "Container will run with default user");
481 Log(undef, "Container will run with $dockeruserarg");
487 if (!defined $dockeruserarg) {
488 croak("Could not find a user in container that is not UID 0 (tried default user, @tryusers) or there was a problem running 'id' in the container.");
491 if ($Job->{arvados_sdk_version}) {
492 # The job also specifies an Arvados SDK version. Add the SDKs to the
493 # tar file for the build script to install.
494 Log(undef, sprintf("Packing Arvados SDK version %s for installation",
495 $Job->{arvados_sdk_version}));
496 add_git_archive("git", "--git-dir=$git_dir", "archive",
497 "--prefix=.arvados.sdk/",
498 $Job->{arvados_sdk_version}, "sdk");
502 if (!defined $git_dir && $Job->{'script_version'} =~ m{^/}) {
503 # If script_version looks like an absolute path, *and* the --git-dir
504 # argument was not given -- which implies we were not invoked by
505 # crunch-dispatch -- we will use the given path as a working
506 # directory instead of resolving script_version to a git commit (or
507 # doing anything else with git).
508 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{'script_version'};
509 $ENV{"CRUNCH_SRC"} = $Job->{'script_version'};
512 # Resolve the given script_version to a git commit sha1. Also, if
513 # the repository is remote, clone it into our local filesystem: this
514 # ensures "git archive" will work, and is necessary to reliably
515 # resolve a symbolic script_version like "master^".
516 $ENV{"CRUNCH_SRC"} = "$ENV{CRUNCH_TMP}/src";
518 Log (undef, "Looking for version ".$Job->{script_version}." from repository ".$Job->{repository});
520 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{script_version};
522 # If we're running under crunch-dispatch, it will have already
523 # pulled the appropriate source tree into its own repository, and
524 # given us that repo's path as $git_dir.
526 # If we're running a "local" job, we might have to fetch content
527 # from a remote repository.
529 # (Currently crunch-dispatch gives a local path with --git-dir, but
530 # we might as well accept URLs there too in case it changes its
532 my $repo = $git_dir || $Job->{'repository'};
534 # Repository can be remote or local. If remote, we'll need to fetch it
535 # to a local dir before doing `git log` et al.
538 if ($repo =~ m{://|^[^/]*:}) {
539 # $repo is a git url we can clone, like git:// or https:// or
540 # file:/// or [user@]host:repo.git. Note "user/name@host:foo" is
541 # not recognized here because distinguishing that from a local
542 # path is too fragile. If you really need something strange here,
543 # use the ssh:// form.
544 $repo_location = 'remote';
545 } elsif ($repo =~ m{^\.*/}) {
546 # $repo is a local path to a git index. We'll also resolve ../foo
547 # to ../foo/.git if the latter is a directory. To help
548 # disambiguate local paths from named hosted repositories, this
549 # form must be given as ./ or ../ if it's a relative path.
550 if (-d "$repo/.git") {
551 $repo = "$repo/.git";
553 $repo_location = 'local';
555 # $repo is none of the above. It must be the name of a hosted
557 my $arv_repo_list = api_call("repositories/list",
558 'filters' => [['name','=',$repo]]);
559 my @repos_found = @{$arv_repo_list->{'items'}};
560 my $n_found = $arv_repo_list->{'serverResponse'}->{'items_available'};
562 Log(undef, "Repository '$repo' -> "
563 . join(", ", map { $_->{'uuid'} } @repos_found));
566 croak("Error: Found $n_found repositories with name '$repo'.");
568 $repo = $repos_found[0]->{'fetch_url'};
569 $repo_location = 'remote';
571 Log(undef, "Using $repo_location repository '$repo'");
572 $ENV{"CRUNCH_SRC_URL"} = $repo;
574 # Resolve given script_version (we'll call that $treeish here) to a
575 # commit sha1 ($commit).
576 my $treeish = $Job->{'script_version'};
578 if ($repo_location eq 'remote') {
579 # We minimize excess object-fetching by re-using the same bare
580 # repository in CRUNCH_TMP/.git for multiple crunch-jobs -- we
581 # just keep adding remotes to it as needed.
582 my $local_repo = $ENV{'CRUNCH_TMP'}."/.git";
583 my $gitcmd = "git --git-dir=\Q$local_repo\E";
585 # Set up our local repo for caching remote objects, making
587 if (!-d $local_repo) {
588 make_path($local_repo) or croak("Error: could not create $local_repo");
590 # This works (exits 0 and doesn't delete fetched objects) even
591 # if $local_repo is already initialized:
592 `$gitcmd init --bare`;
594 croak("Error: $gitcmd init --bare exited ".exit_status_s($?));
597 # If $treeish looks like a hash (or abbrev hash) we look it up in
598 # our local cache first, since that's cheaper. (We don't want to
599 # do that with tags/branches though -- those change over time, so
600 # they should always be resolved by the remote repo.)
601 if ($treeish =~ /^[0-9a-f]{7,40}$/s) {
602 # Hide stderr because it's normal for this to fail:
603 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E 2>/dev/null`;
605 # Careful not to resolve a branch named abcdeff to commit 1234567:
606 $sha1 =~ /^$treeish/ &&
607 $sha1 =~ /^([0-9a-f]{40})$/s) {
609 Log(undef, "Commit $commit already present in $local_repo");
613 if (!defined $commit) {
614 # If $treeish isn't just a hash or abbrev hash, or isn't here
615 # yet, we need to fetch the remote to resolve it correctly.
617 # First, remove all local heads. This prevents a name that does
618 # not exist on the remote from resolving to (or colliding with)
619 # a previously fetched branch or tag (possibly from a different
621 remove_tree("$local_repo/refs/heads", {keep_root => 1});
623 Log(undef, "Fetching objects from $repo to $local_repo");
624 `$gitcmd fetch --no-progress --tags ''\Q$repo\E \Q+refs/heads/*:refs/heads/*\E`;
626 croak("Error: `$gitcmd fetch` exited ".exit_status_s($?));
630 # Now that the data is all here, we will use our local repo for
631 # the rest of our git activities.
635 my $gitcmd = "git --git-dir=\Q$repo\E";
636 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E`;
637 unless ($? == 0 && $sha1 =~ /^([0-9a-f]{40})$/) {
638 croak("`$gitcmd rev-list` exited "
640 .", '$treeish' not found, giving up");
643 Log(undef, "Version $treeish is commit $commit");
645 if ($commit ne $Job->{'script_version'}) {
646 # Record the real commit id in the database, frozentokey, logs,
647 # etc. -- instead of an abbreviation or a branch name which can
648 # become ambiguous or point to a different commit in the future.
649 if (!$Job->update_attributes('script_version' => $commit)) {
650 croak("Error: failed to update job's script_version attribute");
654 $ENV{"CRUNCH_SRC_COMMIT"} = $commit;
655 add_git_archive("$gitcmd archive ''\Q$commit\E");
658 my $git_archive = combined_git_archive();
659 if (!defined $git_archive) {
660 Log(undef, "Skip install phase (no git archive)");
662 Log(undef, "Warning: This probably means workers have no source tree!");
667 my $install_script_tries_left = 3;
668 for (my $attempts = 0; $attempts < 3; $attempts++) {
669 my @srunargs = ("srun",
670 "--nodelist=$nodelist",
671 "-D", $ENV{'TMPDIR'}, "--job-name=$job_id");
672 my @execargs = ("sh", "-c",
673 "mkdir -p $ENV{CRUNCH_INSTALL} && cd $ENV{CRUNCH_TMP} && perl -");
675 $ENV{"CRUNCH_GIT_ARCHIVE_HASH"} = md5_hex($git_archive);
676 my ($stdout, $stderr);
677 ($exited, $stdout, $stderr) = srun_sync(
678 \@srunargs, \@execargs,
679 {label => "run install script on all workers"},
680 $build_script . $git_archive);
682 my $stderr_anything_from_script = 0;
683 for my $line (split(/\n/, $stderr)) {
684 if ($line !~ /^(srun: error: |starting: \[)/) {
685 $stderr_anything_from_script = 1;
689 last if $exited == 0 || $main::please_freeze;
691 # If the install script fails but doesn't print an error message,
692 # the next thing anyone is likely to do is just run it again in
693 # case it was a transient problem like "slurm communication fails
694 # because the network isn't reliable enough". So we'll just do
695 # that ourselves (up to 3 attempts in total). OTOH, if there is an
696 # error message, the problem is more likely to have a real fix and
697 # we should fail the job so the fixing process can start, instead
698 # of doing 2 more attempts.
699 last if $stderr_anything_from_script;
702 foreach my $tar_filename (map { tar_filename_n($_); } (1..$git_tar_count)) {
703 unlink($tar_filename);
711 foreach (qw (script script_version script_parameters runtime_constraints))
715 (ref($Job->{$_}) ? JSON::encode_json($Job->{$_}) : $Job->{$_}));
717 foreach (split (/\n/, $Job->{knobs}))
719 Log (undef, "knob " . $_);
724 $main::success = undef;
730 my $thisround_succeeded = 0;
731 my $thisround_failed = 0;
732 my $thisround_failed_multiple = 0;
733 my $working_slot_count = scalar(@slot);
735 @jobstep_todo = sort { $jobstep[$a]->{level} <=> $jobstep[$b]->{level}
736 or $a <=> $b } @jobstep_todo;
737 my $level = $jobstep[$jobstep_todo[0]]->{level};
739 my $initial_tasks_this_level = 0;
740 foreach my $id (@jobstep_todo) {
741 $initial_tasks_this_level++ if ($jobstep[$id]->{level} == $level);
744 # If the number of tasks scheduled at this level #T is smaller than the number
745 # of slots available #S, only use the first #T slots, or the first slot on
746 # each node, whichever number is greater.
748 # When we dispatch tasks later, we'll allocate whole-node resources like RAM
749 # based on these numbers. Using fewer slots makes more resources available
750 # to each individual task, which should normally be a better strategy when
751 # there are fewer of them running with less parallelism.
753 # Note that this calculation is not redone if the initial tasks at
754 # this level queue more tasks at the same level. This may harm
755 # overall task throughput for that level.
757 if ($initial_tasks_this_level < @node) {
758 @freeslot = (0..$#node);
759 } elsif ($initial_tasks_this_level < @slot) {
760 @freeslot = (0..$initial_tasks_this_level - 1);
762 @freeslot = (0..$#slot);
764 my $round_num_freeslots = scalar(@freeslot);
765 print STDERR "crunch-job have ${round_num_freeslots} free slots for ${initial_tasks_this_level} initial tasks at this level, ".scalar(@node)." nodes, and ".scalar(@slot)." slots\n";
767 my %round_max_slots = ();
768 for (my $ii = $#freeslot; $ii >= 0; $ii--) {
769 my $this_slot = $slot[$freeslot[$ii]];
770 my $node_name = $this_slot->{node}->{name};
771 $round_max_slots{$node_name} ||= $this_slot->{cpu};
772 last if (scalar(keys(%round_max_slots)) >= @node);
775 Log(undef, "start level $level with $round_num_freeslots slots");
778 my $progress_is_dirty = 1;
779 my $progress_stats_updated = 0;
781 update_progress_stats();
785 for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
787 # Don't create new tasks if we already know the job's final result.
788 last if defined($main::success);
790 my $id = $jobstep_todo[$todo_ptr];
791 my $Jobstep = $jobstep[$id];
792 if ($Jobstep->{level} != $level)
797 pipe $reader{$id}, "writer" or croak("pipe() failed: $!");
798 set_nonblocking($reader{$id});
800 my $childslot = $freeslot[0];
801 my $childnode = $slot[$childslot]->{node};
802 my $childslotname = join (".",
803 $slot[$childslot]->{node}->{name},
804 $slot[$childslot]->{cpu});
806 my $childpid = fork();
809 $SIG{'INT'} = 'DEFAULT';
810 $SIG{'QUIT'} = 'DEFAULT';
811 $SIG{'TERM'} = 'DEFAULT';
813 foreach (values (%reader))
817 fcntl ("writer", F_SETFL, 0) or croak ($!); # no close-on-exec
818 open(STDOUT,">&writer");
819 open(STDERR,">&writer");
824 delete $ENV{"GNUPGHOME"};
825 $ENV{"TASK_UUID"} = $Jobstep->{'arvados_task'}->{'uuid'};
826 $ENV{"TASK_QSEQUENCE"} = $id;
827 $ENV{"TASK_SEQUENCE"} = $level;
828 $ENV{"JOB_SCRIPT"} = $Job->{script};
829 while (my ($param, $value) = each %{$Job->{script_parameters}}) {
830 $param =~ tr/a-z/A-Z/;
831 $ENV{"JOB_PARAMETER_$param"} = $value;
833 $ENV{"TASK_SLOT_NODE"} = $slot[$childslot]->{node}->{name};
834 $ENV{"TASK_SLOT_NUMBER"} = $slot[$childslot]->{cpu};
835 $ENV{"TASK_WORK"} = $ENV{"CRUNCH_TMP"}."/task/$childslotname";
836 $ENV{"HOME"} = $ENV{"TASK_WORK"};
837 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
838 $ENV{"CRUNCH_NODE_SLOTS"} = $round_max_slots{$ENV{TASK_SLOT_NODE}};
839 $ENV{"PATH"} = $ENV{"CRUNCH_INSTALL"} . "/bin:" . $ENV{"PATH"};
841 my $keep_mnt = $ENV{"TASK_WORK"}.".keep";
847 "--nodelist=".$childnode->{name},
848 qw(-n1 -c1 -N1 -D), $ENV{'TMPDIR'},
849 "--job-name=$job_id.$id.$$",
852 my $stdbuf = " stdbuf --output=0 --error=0 ";
854 my $arv_file_cache = "";
855 if (defined($Job->{'runtime_constraints'}->{'keep_cache_mb_per_task'})) {
856 $arv_file_cache = "--file-cache=" . ($Job->{'runtime_constraints'}->{'keep_cache_mb_per_task'} * 1024 * 1024);
860 "if [ -e \Q$ENV{TASK_WORK}\E ]; then rm -rf \Q$ENV{TASK_WORK}\E; fi; "
861 ."mkdir -p \Q$ENV{CRUNCH_TMP}\E \Q$ENV{JOB_WORK}\E \Q$ENV{TASK_WORK}\E \Q$keep_mnt\E "
862 ."&& cd \Q$ENV{CRUNCH_TMP}\E "
863 # These environment variables get used explicitly later in
864 # $command. No tool is expected to read these values directly.
865 .q{&& MEM=$(awk '($1 == "MemTotal:"){print $2}' </proc/meminfo) }
866 .q{&& SWAP=$(awk '($1 == "SwapTotal:"){print $2}' </proc/meminfo) }
867 ."&& MEMLIMIT=\$(( (\$MEM * 95) / ($ENV{CRUNCH_NODE_SLOTS} * 100) )) "
868 ."&& let SWAPLIMIT=\$MEMLIMIT+\$SWAP ";
870 $command .= "&& exec arv-mount --read-write --mount-by-pdh=by_pdh --mount-tmp=tmp --crunchstat-interval=10 --allow-other $arv_file_cache \Q$keep_mnt\E --exec ";
871 $ENV{TASK_KEEPMOUNT} = "$keep_mnt/by_pdh";
872 $ENV{TASK_KEEPMOUNT_TMP} = "$keep_mnt/tmp";
876 my $containername = "$Jobstep->{arvados_task}->{uuid}-$Jobstep->{failures}";
877 my $cidfile = "$ENV{CRUNCH_TMP}/$containername.cid";
878 $command .= "crunchstat -cgroup-root=\Q$cgroup_root\E -cgroup-parent=docker -cgroup-cid=$cidfile -poll=10000 ";
879 $command .= "$docker_bin run $docker_run_args --name=$containername --attach=stdout --attach=stderr --attach=stdin -i \Q$dockeruserarg\E --cidfile=$cidfile --sig-proxy ";
880 # We only set memory limits if Docker lets us limit both memory and swap.
881 # Memory limits alone have been supported longer, but subprocesses tend
882 # to get SIGKILL if they exceed that without any swap limit set.
883 # See #5642 for additional background.
884 if ($docker_limitmem) {
885 $command .= "--memory=\${MEMLIMIT}k --memory-swap=\${SWAPLIMIT}k ";
888 # The source tree and $destdir directory (which we have
889 # installed on the worker host) are available in the container,
890 # under the same path.
891 $command .= "--volume=\Q$ENV{CRUNCH_SRC}:$ENV{CRUNCH_SRC}:ro\E ";
892 $command .= "--volume=\Q$ENV{CRUNCH_INSTALL}:$ENV{CRUNCH_INSTALL}:ro\E ";
894 # Currently, we make the "by_pdh" directory in arv-mount's mount
895 # point appear at /keep inside the container (instead of using
896 # the same path as the host like we do with CRUNCH_SRC and
897 # CRUNCH_INSTALL). However, crunch scripts and utilities must
898 # not rely on this. They must use $TASK_KEEPMOUNT.
899 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT}:/keep:ro\E ";
900 $ENV{TASK_KEEPMOUNT} = "/keep";
902 # Ditto TASK_KEEPMOUNT_TMP, as /keep_tmp.
903 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT_TMP}:/keep_tmp\E ";
904 $ENV{TASK_KEEPMOUNT_TMP} = "/keep_tmp";
906 # TASK_WORK is almost exactly like a docker data volume: it
907 # starts out empty, is writable, and persists until no
908 # containers use it any more. We don't use --volumes-from to
909 # share it with other containers: it is only accessible to this
910 # task, and it goes away when this task stops.
912 # However, a docker data volume is writable only by root unless
913 # the mount point already happens to exist in the container with
914 # different permissions. Therefore, we [1] assume /tmp already
915 # exists in the image and is writable by the crunch user; [2]
916 # avoid putting TASK_WORK inside CRUNCH_TMP (which won't be
917 # writable if they are created by docker while setting up the
918 # other --volumes); and [3] create $TASK_WORK inside the
919 # container using $build_script.
920 $command .= "--volume=/tmp ";
921 $ENV{"TASK_WORK"} = "/tmp/crunch-job-task-work/$childslotname";
922 $ENV{"HOME"} = $ENV{"TASK_WORK"};
923 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
925 # TODO: Share a single JOB_WORK volume across all task
926 # containers on a given worker node, and delete it when the job
927 # ends (and, in case that doesn't work, when the next job
930 # For now, use the same approach as TASK_WORK above.
931 $ENV{"JOB_WORK"} = "/tmp/crunch-job-work";
933 # Bind mount the crunchrunner binary and host TLS certificates file into
935 $command .= "--volume=\Q$ENV{HOST_CRUNCHRUNNER_BIN}:/usr/local/bin/crunchrunner\E ";
936 $command .= "--volume=\Q$ENV{HOST_CERTS}:/etc/arvados/ca-certificates.crt\E ";
938 while (my ($env_key, $env_val) = each %ENV)
940 if ($env_key =~ /^(ARVADOS|CRUNCH|JOB|TASK)_/) {
941 $command .= "--env=\Q$env_key=$env_val\E ";
944 $command .= "--env=\QHOME=$ENV{HOME}\E ";
945 $command .= "\Q$docker_hash\E ";
947 if ($Job->{arvados_sdk_version}) {
949 $command .= "perl - \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E";
951 $command .= "/bin/sh -c \'python -c " .
952 '"from pkg_resources import get_distribution as get; print \"Using Arvados SDK version\", get(\"arvados-python-client\").version"' .
953 ">&2 2>/dev/null; " .
954 "mkdir -p \"$ENV{JOB_WORK}\" \"$ENV{TASK_WORK}\" && " .
955 "if which stdbuf >/dev/null ; then " .
956 " exec $stdbuf \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E ;" .
958 " exec \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E ;" .
963 $command .= "crunchstat -cgroup-root=\Q$cgroup_root\E -poll=10000 ";
965 $command .= "perl - $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
968 my @execargs = ('bash', '-c', $command);
969 srun (\@srunargs, \@execargs, undef, $build_script);
970 # exec() failed, we assume nothing happened.
971 die "srun() failed on build script\n";
974 if (!defined $childpid)
985 jobstepname => "$job_id.$id.$childpid",
987 croak ("assert failed: \$slot[$childslot]->{'pid'} exists") if exists $slot[$childslot]->{pid};
988 $slot[$childslot]->{pid} = $childpid;
990 Log ($id, "job_task ".$Jobstep->{'arvados_task'}->{'uuid'});
991 Log ($id, "child $childpid started on $childslotname");
992 $Jobstep->{starttime} = time;
993 $Jobstep->{node} = $childnode->{name};
994 $Jobstep->{slotindex} = $childslot;
995 delete $Jobstep->{stderr};
996 delete $Jobstep->{finishtime};
997 delete $Jobstep->{tempfail};
999 $Jobstep->{'arvados_task'}->{started_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{starttime});
1000 $Jobstep->{'arvados_task'}->save;
1002 splice @jobstep_todo, $todo_ptr, 1;
1005 $progress_is_dirty = 1;
1009 ($round_num_freeslots > @freeslot && $todo_ptr+1 > $#jobstep_todo))
1011 last THISROUND if $main::please_freeze;
1012 if ($main::please_info)
1014 $main::please_info = 0;
1016 create_output_collection();
1018 update_progress_stats();
1023 if (!$gotsome || ($latest_refresh + 2 < scalar time))
1025 check_refresh_wanted();
1027 update_progress_stats();
1029 elsif (time - $progress_stats_updated >= 30 || $progress_is_dirty)
1031 update_progress_stats();
1034 select (undef, undef, undef, 0.1);
1036 $working_slot_count = scalar(grep { $_->{node}->{fail_count} == 0 &&
1037 $_->{node}->{hold_count} < 4 } @slot);
1038 if (($thisround_failed_multiple >= 8 && $thisround_succeeded == 0) ||
1039 ($thisround_failed_multiple >= 16 && $thisround_failed_multiple > $thisround_succeeded))
1041 my $message = "Repeated failure rate too high ($thisround_failed_multiple/"
1042 .($thisround_failed+$thisround_succeeded)
1043 .") -- giving up on this round";
1044 Log (undef, $message);
1048 # move slots from freeslot to holdslot (or back to freeslot) if necessary
1049 for (my $i=$#freeslot; $i>=0; $i--) {
1050 if ($slot[$freeslot[$i]]->{node}->{hold_until} > scalar time) {
1051 push @holdslot, (splice @freeslot, $i, 1);
1054 for (my $i=$#holdslot; $i>=0; $i--) {
1055 if ($slot[$holdslot[$i]]->{node}->{hold_until} <= scalar time) {
1056 push @freeslot, (splice @holdslot, $i, 1);
1060 # give up if no nodes are succeeding
1061 if ($working_slot_count < 1) {
1062 Log(undef, "Every node has failed -- giving up");
1069 push @freeslot, splice @holdslot;
1070 map { $slot[$freeslot[$_]]->{node}->{losing_streak} = 0 } (0..$#freeslot);
1073 Log (undef, "wait for last ".(scalar keys %proc)." children to finish");
1076 if ($main::please_continue) {
1077 $main::please_continue = 0;
1080 $main::please_info = 0, freeze(), create_output_collection(), save_meta(1) if $main::please_info;
1082 if (!reapchildren())
1084 check_refresh_wanted();
1086 update_progress_stats();
1087 select (undef, undef, undef, 0.1);
1088 killem (keys %proc) if $main::please_freeze;
1092 update_progress_stats();
1093 freeze_if_want_freeze();
1096 if (!defined $main::success)
1098 if (!@jobstep_todo) {
1100 } elsif ($working_slot_count < 1) {
1101 save_output_collection();
1103 exit(EX_RETRY_UNLOCKED);
1104 } elsif ($thisround_succeeded == 0 &&
1105 ($thisround_failed == 0 || $thisround_failed > 4)) {
1106 my $message = "stop because $thisround_failed tasks failed and none succeeded";
1107 Log (undef, $message);
1112 goto ONELEVEL if !defined $main::success;
1115 release_allocation();
1117 my $collated_output = save_output_collection();
1118 Log (undef, "finish");
1123 if ($collated_output && $main::success) {
1124 $final_state = 'Complete';
1126 $final_state = 'Failed';
1128 $Job->update_attributes('state' => $final_state);
1130 exit (($final_state eq 'Complete') ? 0 : 1);
1134 sub update_progress_stats
1136 $progress_stats_updated = time;
1137 return if !$progress_is_dirty;
1138 my ($todo, $done, $running) = (scalar @jobstep_todo,
1139 scalar @jobstep_done,
1140 scalar keys(%proc));
1141 $Job->{'tasks_summary'} ||= {};
1142 $Job->{'tasks_summary'}->{'todo'} = $todo;
1143 $Job->{'tasks_summary'}->{'done'} = $done;
1144 $Job->{'tasks_summary'}->{'running'} = $running;
1145 $Job->update_attributes('tasks_summary' => $Job->{'tasks_summary'});
1146 Log (undef, "status: $done done, $running running, $todo todo");
1147 $progress_is_dirty = 0;
1154 my $children_reaped = 0;
1155 my @successful_task_uuids = ();
1157 while((my $pid = waitpid (-1, WNOHANG)) > 0)
1159 my $childstatus = $?;
1161 my $whatslot = ($slot[$proc{$pid}->{slot}]->{node}->{name}
1163 . $slot[$proc{$pid}->{slot}]->{cpu});
1164 my $jobstepidx = $proc{$pid}->{jobstepidx};
1166 if (!WIFEXITED($childstatus))
1168 # child did not exit (may be temporarily stopped)
1169 Log ($jobstepidx, "child $pid did not actually exit in reapchildren, ignoring for now.");
1174 my $elapsed = time - $proc{$pid}->{time};
1175 my $Jobstep = $jobstep[$jobstepidx];
1177 my $exitvalue = $childstatus >> 8;
1178 my $exitinfo = "exit ".exit_status_s($childstatus);
1179 $Jobstep->{'arvados_task'}->reload;
1180 my $task_success = $Jobstep->{'arvados_task'}->{success};
1182 Log ($jobstepidx, "child $pid on $whatslot $exitinfo success=$task_success");
1184 if (!defined $task_success) {
1185 # task did not indicate one way or the other --> fail
1186 Log($jobstepidx, sprintf(
1187 "ERROR: Task process exited %s, but never updated its task record to indicate success and record its output.",
1188 exit_status_s($childstatus)));
1189 $Jobstep->{'arvados_task'}->{success} = 0;
1190 $Jobstep->{'arvados_task'}->save;
1197 $temporary_fail ||= $Jobstep->{tempfail};
1198 $temporary_fail ||= ($exitvalue == TASK_TEMPFAIL);
1200 ++$thisround_failed;
1201 ++$thisround_failed_multiple if $Jobstep->{'failures'} >= 1;
1203 # Check for signs of a failed or misconfigured node
1204 if (++$slot[$proc{$pid}->{slot}]->{node}->{losing_streak} >=
1205 2+$slot[$proc{$pid}->{slot}]->{node}->{ncpus}) {
1206 # Don't count this against jobstep failure thresholds if this
1207 # node is already suspected faulty and srun exited quickly
1208 if ($slot[$proc{$pid}->{slot}]->{node}->{hold_until} &&
1210 Log ($jobstepidx, "blaming failure on suspect node " .
1211 $slot[$proc{$pid}->{slot}]->{node}->{name});
1212 $temporary_fail ||= 1;
1214 ban_node_by_slot($proc{$pid}->{slot});
1217 Log ($jobstepidx, sprintf('failure (#%d, %s) after %d seconds',
1218 ++$Jobstep->{'failures'},
1219 $temporary_fail ? 'temporary' : 'permanent',
1222 if (!$temporary_fail || $Jobstep->{'failures'} >= 3) {
1223 # Give up on this task, and the whole job
1226 # Put this task back on the todo queue
1227 push @jobstep_todo, $jobstepidx;
1228 $Job->{'tasks_summary'}->{'failed'}++;
1232 push @successful_task_uuids, $Jobstep->{'arvados_task'}->{uuid};
1233 ++$thisround_succeeded;
1234 $slot[$proc{$pid}->{slot}]->{node}->{losing_streak} = 0;
1235 $slot[$proc{$pid}->{slot}]->{node}->{hold_until} = 0;
1236 $slot[$proc{$pid}->{slot}]->{node}->{fail_count} = 0;
1237 push @jobstep_done, $jobstepidx;
1238 Log ($jobstepidx, "success in $elapsed seconds");
1240 $Jobstep->{exitcode} = $childstatus;
1241 $Jobstep->{finishtime} = time;
1242 $Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
1243 $Jobstep->{'arvados_task'}->save;
1244 process_stderr_final ($jobstepidx);
1245 Log ($jobstepidx, sprintf("task output (%d bytes): %s",
1246 length($Jobstep->{'arvados_task'}->{output}),
1247 $Jobstep->{'arvados_task'}->{output}));
1249 close $reader{$jobstepidx};
1250 delete $reader{$jobstepidx};
1251 delete $slot[$proc{$pid}->{slot}]->{pid};
1252 push @freeslot, $proc{$pid}->{slot};
1255 $progress_is_dirty = 1;
1258 if (scalar(@successful_task_uuids) > 0)
1260 Log (undef, sprintf("%d tasks exited (%d succeeded), checking for new tasks from API server.", $children_reaped, scalar(@successful_task_uuids)));
1262 my $newtask_list = [];
1263 my $newtask_results;
1265 $newtask_results = api_call(
1267 'filters' => [["created_by_job_task_uuid","in",\@successful_task_uuids]],
1268 'order' => 'qsequence',
1269 'offset' => scalar(@$newtask_list),
1271 push(@$newtask_list, @{$newtask_results->{items}});
1272 } while (@{$newtask_results->{items}});
1273 Log (undef, sprintf("Got %d new tasks from API server.", scalar(@$newtask_list)));
1274 foreach my $arvados_task (@$newtask_list) {
1276 'level' => $arvados_task->{'sequence'},
1278 'arvados_task' => $arvados_task
1280 push @jobstep, $jobstep;
1281 push @jobstep_todo, $#jobstep;
1285 return $children_reaped;
1288 sub check_refresh_wanted
1290 my @stat = stat $ENV{"CRUNCH_REFRESH_TRIGGER"};
1292 $stat[9] > $latest_refresh &&
1293 # ...and we have actually locked the job record...
1294 $job_id eq $Job->{'uuid'}) {
1295 $latest_refresh = scalar time;
1296 my $Job2 = api_call("jobs/get", uuid => $jobspec);
1297 for my $attr ('cancelled_at',
1298 'cancelled_by_user_uuid',
1299 'cancelled_by_client_uuid',
1301 $Job->{$attr} = $Job2->{$attr};
1303 if ($Job->{'state'} ne "Running") {
1304 if ($Job->{'state'} eq "Cancelled") {
1305 Log (undef, "Job cancelled at " . $Job->{'cancelled_at'} . " by user " . $Job->{'cancelled_by_user_uuid'});
1307 Log (undef, "Job state unexpectedly changed to " . $Job->{'state'});
1310 $main::please_freeze = 1;
1317 my $last_squeue_check = $squeue_checked;
1319 # Do not call `squeue` or check the kill list more than once every
1321 return if $last_squeue_check > time - 15;
1322 $squeue_checked = time;
1324 # Look for children from which we haven't received stderr data since
1325 # the last squeue check. If no such children exist, all procs are
1326 # alive and there's no need to even look at squeue.
1328 # As long as the crunchstat poll interval (10s) is shorter than the
1329 # squeue check interval (15s) this should make the squeue check an
1331 my $silent_procs = 0;
1332 for my $js (map {$jobstep[$_->{jobstepidx}]} values %proc)
1334 if (!exists($js->{stderr_at}))
1336 $js->{stderr_at} = 0;
1338 if ($js->{stderr_at} < $last_squeue_check)
1343 return if $silent_procs == 0;
1345 # use killem() on procs whose killtime is reached
1346 while (my ($pid, $procinfo) = each %proc)
1348 my $js = $jobstep[$procinfo->{jobstepidx}];
1349 if (exists $procinfo->{killtime}
1350 && $procinfo->{killtime} <= time
1351 && $js->{stderr_at} < $last_squeue_check)
1354 if ($js->{stderr_at}) {
1355 $sincewhen = " in last " . (time - $js->{stderr_at}) . "s";
1357 Log($procinfo->{jobstepidx}, "killing orphaned srun process $pid (task not in slurm queue, no stderr received$sincewhen)");
1364 # here is an opportunity to check for mysterious problems with local procs
1368 # Get a list of steps still running. Note: squeue(1) says --steps
1369 # selects a format (which we override anyway) and allows us to
1370 # specify which steps we're interested in (which we don't).
1371 # Importantly, it also changes the meaning of %j from "job name" to
1372 # "step name" and (although this isn't mentioned explicitly in the
1373 # docs) switches from "one line per job" mode to "one line per step"
1374 # mode. Without it, we'd just get a list of one job, instead of a
1376 my @squeue = `squeue --jobs=\Q$ENV{SLURM_JOB_ID}\E --steps --format='%j' --noheader`;
1379 Log(undef, "warning: squeue exit status $? ($!)");
1384 # which of my jobsteps are running, according to squeue?
1386 for my $jobstepname (@squeue)
1388 $ok{$jobstepname} = 1;
1391 # Check for child procs >60s old and not mentioned by squeue.
1392 while (my ($pid, $procinfo) = each %proc)
1394 if ($procinfo->{time} < time - 60
1395 && $procinfo->{jobstepname}
1396 && !exists $ok{$procinfo->{jobstepname}}
1397 && !exists $procinfo->{killtime})
1399 # According to slurm, this task has ended (successfully or not)
1400 # -- but our srun child hasn't exited. First we must wait (30
1401 # seconds) in case this is just a race between communication
1402 # channels. Then, if our srun child process still hasn't
1403 # terminated, we'll conclude some slurm communication
1404 # error/delay has caused the task to die without notifying srun,
1405 # and we'll kill srun ourselves.
1406 $procinfo->{killtime} = time + 30;
1407 Log($procinfo->{jobstepidx}, "notice: task is not in slurm queue but srun process $pid has not exited");
1413 sub release_allocation
1417 Log (undef, "release job allocation");
1418 system "scancel $ENV{SLURM_JOB_ID}";
1427 my $sel = IO::Select->new();
1428 foreach my $jobstepidx (keys %reader)
1430 my $fd = $reader{$jobstepidx};
1432 $fd_job{$fd} = $jobstepidx;
1434 if (my $stdout_fd = $jobstep[$jobstepidx]->{stdout_r}) {
1435 $sel->add($stdout_fd);
1436 $fd_job{$stdout_fd} = $jobstepidx;
1439 # select on all reader fds with 0.1s timeout
1440 my @ready_fds = $sel->can_read(0.1);
1441 foreach my $fd (@ready_fds)
1444 if (0 < sysread ($fd, $buf, 65536))
1447 print STDERR $buf if $ENV{CRUNCH_DEBUG};
1449 my $jobstepidx = $fd_job{$fd};
1450 if ($jobstep[$jobstepidx]->{stdout_r} == $fd) {
1451 $jobstep[$jobstepidx]->{stdout_captured} .= $buf;
1455 $jobstep[$jobstepidx]->{stderr_at} = time;
1456 $jobstep[$jobstepidx]->{stderr} .= $buf;
1458 # Consume everything up to the last \n
1459 preprocess_stderr ($jobstepidx);
1461 if (length ($jobstep[$jobstepidx]->{stderr}) > 16384)
1463 # If we get a lot of stderr without a newline, chop off the
1464 # front to avoid letting our buffer grow indefinitely.
1465 substr ($jobstep[$jobstepidx]->{stderr},
1466 0, length($jobstep[$jobstepidx]->{stderr}) - 8192) = "";
1474 # Consume all full lines of stderr for a jobstep. Everything after the
1475 # last newline will remain in $jobstep[$jobstepidx]->{stderr} after
1477 sub preprocess_stderr
1479 my $jobstepidx = shift;
1481 while ($jobstep[$jobstepidx]->{stderr} =~ /^(.*?)\n/) {
1483 substr $jobstep[$jobstepidx]->{stderr}, 0, 1+length($line), "";
1484 Log ($jobstepidx, "stderr $line");
1485 if ($line =~ /srun: error: (SLURM job $ENV{SLURM_JOB_ID} has expired|Unable to confirm allocation for job $ENV{SLURM_JOB_ID})/) {
1487 $main::please_freeze = 1;
1489 elsif (!exists $jobstep[$jobstepidx]->{slotindex}) {
1490 # Skip the following tempfail checks if this srun proc isn't
1491 # attached to a particular worker slot.
1493 elsif ($line =~ /srun: error: (Node failure on|Aborting, .*\bio error\b)/) {
1494 my $job_slot_index = $jobstep[$jobstepidx]->{slotindex};
1495 $slot[$job_slot_index]->{node}->{fail_count}++;
1496 $jobstep[$jobstepidx]->{tempfail} = 1;
1497 ban_node_by_slot($job_slot_index);
1499 elsif ($line =~ /srun: error: (Unable to create job step|.*: Communication connection failure)/) {
1500 $jobstep[$jobstepidx]->{tempfail} = 1;
1501 ban_node_by_slot($jobstep[$jobstepidx]->{slotindex});
1503 elsif ($line =~ /\bKeep(Read|Write|Request)Error:/) {
1504 $jobstep[$jobstepidx]->{tempfail} = 1;
1510 sub process_stderr_final
1512 my $jobstepidx = shift;
1513 preprocess_stderr ($jobstepidx);
1516 Log ($jobstepidx, "stderr $_");
1517 } split ("\n", $jobstep[$jobstepidx]->{stderr});
1518 $jobstep[$jobstepidx]->{stderr} = '';
1525 if (!open($keep, "-|", "arv-get", "--retries", retry_count(), $hash)) {
1526 Log(undef, "fetch_block run error from arv-get $hash: $!");
1529 my $output_block = "";
1532 my $bytes = sysread($keep, $buf, 1024 * 1024);
1533 if (!defined $bytes) {
1534 Log(undef, "fetch_block read error from arv-get: $!");
1535 $output_block = undef;
1537 } elsif ($bytes == 0) {
1538 # sysread returns 0 at the end of the pipe.
1541 # some bytes were read into buf.
1542 $output_block .= $buf;
1547 Log(undef, "fetch_block arv-get exited " . exit_status_s($?));
1548 $output_block = undef;
1550 return $output_block;
1553 # Create a collection by concatenating the output of all tasks (each
1554 # task's output is either a manifest fragment, a locator for a
1555 # manifest fragment stored in Keep, or nothing at all). Return the
1556 # portable_data_hash of the new collection.
1557 sub create_output_collection
1559 Log (undef, "collate");
1561 my ($child_out, $child_in);
1562 my $pid = open2($child_out, $child_in, 'python', '-c', q{
1565 print (arvados.api("v1").collections().
1566 create(body={"manifest_text": sys.stdin.read()}).
1567 execute(num_retries=int(sys.argv[1]))["portable_data_hash"])
1571 my $manifest_size = 0;
1575 my $output = $_->{'arvados_task'}->{output};
1576 next if (!defined($output));
1578 if ($output =~ /^[0-9a-f]{32}(\+\S+)*$/) {
1579 $next_write = fetch_block($output);
1581 $next_write = $output;
1583 if (defined($next_write)) {
1584 if (!defined(syswrite($child_in, $next_write))) {
1585 # There's been an error writing. Stop the loop.
1586 # We'll log details about the exit code later.
1589 $manifest_size += length($next_write);
1592 my $uuid = $_->{'arvados_task'}->{'uuid'};
1593 Log (undef, "Error retrieving '$output' output by task $task_idx ($uuid)");
1598 Log(undef, "collated output manifest text to send to API server is $manifest_size bytes with access tokens");
1601 my $s = IO::Select->new($child_out);
1602 if ($s->can_read(120)) {
1603 sysread($child_out, $joboutput, 1024 * 1024);
1606 Log(undef, "output collection creation exited " . exit_status_s($?));
1612 Log (undef, "timed out while creating output collection");
1613 foreach my $signal (2, 2, 2, 15, 15, 9) {
1614 kill($signal, $pid);
1615 last if waitpid($pid, WNOHANG) == -1;
1624 # Calls create_output_collection, logs the result, and returns it.
1625 # If that was successful, save that as the output in the job record.
1626 sub save_output_collection {
1627 my $collated_output = create_output_collection();
1629 if (!$collated_output) {
1630 Log(undef, "Failed to write output collection");
1633 Log(undef, "job output $collated_output");
1634 $Job->update_attributes('output' => $collated_output);
1636 return $collated_output;
1643 my $sig = 2; # SIGINT first
1644 if (exists $proc{$_}->{"sent_$sig"} &&
1645 time - $proc{$_}->{"sent_$sig"} > 4)
1647 $sig = 15; # SIGTERM if SIGINT doesn't work
1649 if (exists $proc{$_}->{"sent_$sig"} &&
1650 time - $proc{$_}->{"sent_$sig"} > 4)
1652 $sig = 9; # SIGKILL if SIGTERM doesn't work
1654 if (!exists $proc{$_}->{"sent_$sig"})
1656 Log ($proc{$_}->{jobstepidx}, "sending 2x signal $sig to pid $_");
1658 select (undef, undef, undef, 0.1);
1661 kill $sig, $_; # srun wants two SIGINT to really interrupt
1663 $proc{$_}->{"sent_$sig"} = time;
1664 $proc{$_}->{"killedafter"} = time - $proc{$_}->{"time"};
1674 vec($bits,fileno($_),1) = 1;
1680 # Send log output to Keep via arv-put.
1682 # $log_pipe_in and $log_pipe_out are the input and output filehandles to the arv-put pipe.
1683 # $log_pipe_out_buf is a string containing all output read from arv-put so far.
1684 # $log_pipe_out_select is an IO::Select object around $log_pipe_out.
1685 # $log_pipe_pid is the pid of the arv-put subprocess.
1687 # The only functions that should access these variables directly are:
1689 # log_writer_start($logfilename)
1690 # Starts an arv-put pipe, reading data on stdin and writing it to
1691 # a $logfilename file in an output collection.
1693 # log_writer_read_output([$timeout])
1694 # Read output from $log_pipe_out and append it to $log_pipe_out_buf.
1695 # Passes $timeout to the select() call, with a default of 0.01.
1696 # Returns the result of the last read() call on $log_pipe_out, or
1697 # -1 if read() wasn't called because select() timed out.
1698 # Only other log_writer_* functions should need to call this.
1700 # log_writer_send($txt)
1701 # Writes $txt to the output log collection.
1703 # log_writer_finish()
1704 # Closes the arv-put pipe and returns the output that it produces.
1706 # log_writer_is_active()
1707 # Returns a true value if there is currently a live arv-put
1708 # process, false otherwise.
1710 my ($log_pipe_in, $log_pipe_out, $log_pipe_out_buf, $log_pipe_out_select,
1713 sub log_writer_start($)
1715 my $logfilename = shift;
1716 $log_pipe_pid = open2($log_pipe_out, $log_pipe_in,
1720 '--filename', $logfilename,
1722 $log_pipe_out_buf = "";
1723 $log_pipe_out_select = IO::Select->new($log_pipe_out);
1726 sub log_writer_read_output {
1727 my $timeout = shift || 0.01;
1729 while ($read && $log_pipe_out_select->can_read($timeout)) {
1730 $read = read($log_pipe_out, $log_pipe_out_buf, 65536,
1731 length($log_pipe_out_buf));
1733 if (!defined($read)) {
1734 Log(undef, "error reading log manifest from arv-put: $!");
1739 sub log_writer_send($)
1742 print $log_pipe_in $txt;
1743 log_writer_read_output();
1746 sub log_writer_finish()
1748 return unless $log_pipe_pid;
1750 close($log_pipe_in);
1752 my $logger_failed = 0;
1753 my $read_result = log_writer_read_output(120);
1754 if ($read_result == -1) {
1755 $logger_failed = -1;
1756 Log (undef, "timed out reading from 'arv-put'");
1757 } elsif ($read_result != 0) {
1758 $logger_failed = -2;
1759 Log(undef, "failed to read arv-put log manifest to EOF");
1762 waitpid($log_pipe_pid, 0);
1764 $logger_failed ||= $?;
1765 Log(undef, "log_writer_finish: arv-put exited " . exit_status_s($?))
1768 close($log_pipe_out);
1769 my $arv_put_output = $logger_failed ? undef : $log_pipe_out_buf;
1770 $log_pipe_pid = $log_pipe_in = $log_pipe_out = $log_pipe_out_buf =
1771 $log_pipe_out_select = undef;
1773 return $arv_put_output;
1776 sub log_writer_is_active() {
1777 return $log_pipe_pid;
1780 sub Log # ($jobstepidx, $logmessage)
1782 my ($jobstepidx, $logmessage) = @_;
1783 if ($logmessage =~ /\n/) {
1784 for my $line (split (/\n/, $_[1])) {
1785 Log ($jobstepidx, $line);
1789 my $fh = select STDERR; $|=1; select $fh;
1791 if (defined($jobstepidx) && exists($jobstep[$jobstepidx]->{arvados_task})) {
1792 $task_qseq = $jobstepidx;
1794 my $message = sprintf ("%s %d %s %s", $job_id, $$, $task_qseq, $logmessage);
1795 $message =~ s{([^ -\176])}{"\\" . sprintf ("%03o", ord($1))}ge;
1798 if (log_writer_is_active() || -t STDERR) {
1799 my @gmtime = gmtime;
1800 $datetime = sprintf ("%04d-%02d-%02d_%02d:%02d:%02d",
1801 $gmtime[5]+1900, $gmtime[4]+1, @gmtime[3,2,1,0]);
1803 print STDERR ((-t STDERR) ? ($datetime." ".$message) : $message);
1805 if (log_writer_is_active()) {
1806 log_writer_send($datetime . " " . $message);
1813 my ($package, $file, $line) = caller;
1814 my $message = "@_ at $file line $line\n";
1815 Log (undef, $message);
1816 freeze() if @jobstep_todo;
1817 create_output_collection() if @jobstep_todo;
1827 if ($Job->{'state'} eq 'Cancelled') {
1828 $Job->update_attributes('finished_at' => scalar gmtime);
1830 $Job->update_attributes('state' => 'Failed');
1837 my $justcheckpoint = shift; # false if this will be the last meta saved
1838 return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm
1839 return unless log_writer_is_active();
1840 my $log_manifest = log_writer_finish();
1841 return unless defined($log_manifest);
1844 my $prev_log_coll = api_call("collections/get", uuid => $Job->{log});
1845 $log_manifest = $prev_log_coll->{manifest_text} . $log_manifest;
1848 my $log_coll = api_call(
1849 "collections/create", ensure_unique_name => 1, collection => {
1850 manifest_text => $log_manifest,
1851 owner_uuid => $Job->{owner_uuid},
1852 name => sprintf("Log from %s job %s", $Job->{script}, $Job->{uuid}),
1854 Log(undef, "log collection is " . $log_coll->{portable_data_hash});
1855 $Job->update_attributes('log' => $log_coll->{portable_data_hash});
1859 sub freeze_if_want_freeze
1861 if ($main::please_freeze)
1863 release_allocation();
1866 # kill some srun procs before freeze+stop
1867 map { $proc{$_} = {} } @_;
1870 killem (keys %proc);
1871 select (undef, undef, undef, 0.1);
1873 while (($died = waitpid (-1, WNOHANG)) > 0)
1875 delete $proc{$died};
1880 create_output_collection();
1890 Log (undef, "Freeze not implemented");
1897 croak ("Thaw not implemented");
1913 $s =~ s{\\(.)}{$1 eq "n" ? "\n" : $1}ge;
1920 my $srunargs = shift;
1921 my $execargs = shift;
1922 my $opts = shift || {};
1925 my $label = exists $opts->{label} ? $opts->{label} : "@$execargs";
1926 Log (undef, "$label: start");
1928 my ($stderr_r, $stderr_w);
1929 pipe $stderr_r, $stderr_w or croak("pipe() failed: $!");
1931 my ($stdout_r, $stdout_w);
1932 pipe $stdout_r, $stdout_w or croak("pipe() failed: $!");
1934 my $srunpid = fork();
1939 fcntl($stderr_w, F_SETFL, 0) or croak($!); # no close-on-exec
1940 fcntl($stdout_w, F_SETFL, 0) or croak($!);
1941 open(STDERR, ">&", $stderr_w);
1942 open(STDOUT, ">&", $stdout_w);
1943 srun ($srunargs, $execargs, $opts, $stdin);
1949 set_nonblocking($stderr_r);
1950 set_nonblocking($stdout_r);
1952 # Add entries to @jobstep and %proc so check_squeue() and
1953 # freeze_if_want_freeze() can treat it like a job task process.
1957 stderr_captured => '',
1958 stdout_r => $stdout_r,
1959 stdout_captured => '',
1961 my $jobstepidx = $#jobstep;
1963 jobstepidx => $jobstepidx,
1965 $reader{$jobstepidx} = $stderr_r;
1967 while ($srunpid != waitpid ($srunpid, WNOHANG)) {
1968 my $busy = readfrompipes();
1969 if (!$busy || ($latest_refresh + 2 < scalar time)) {
1970 check_refresh_wanted();
1974 select(undef, undef, undef, 0.1);
1976 killem(keys %proc) if $main::please_freeze;
1980 1 while readfrompipes();
1981 process_stderr_final ($jobstepidx);
1983 Log (undef, "$label: exit ".exit_status_s($exited));
1987 delete $proc{$srunpid};
1988 delete $reader{$jobstepidx};
1990 my $j = pop @jobstep;
1991 return ($exited, $j->{stdout_captured}, $j->{stderr_captured});
1997 my $srunargs = shift;
1998 my $execargs = shift;
1999 my $opts = shift || {};
2001 my $args = $have_slurm ? [@$srunargs, @$execargs] : $execargs;
2003 $Data::Dumper::Terse = 1;
2004 $Data::Dumper::Indent = 0;
2005 my $show_cmd = Dumper($args);
2006 $show_cmd =~ s/(TOKEN\\*=)[^\s\']+/${1}[...]/g;
2007 $show_cmd =~ s/\n/ /g;
2008 if ($opts->{fork}) {
2009 Log(undef, "starting: $show_cmd");
2011 # This is a child process: parent is in charge of reading our
2012 # stderr and copying it to Log() if needed.
2013 warn "starting: $show_cmd\n";
2016 if (defined $stdin) {
2017 my $child = open STDIN, "-|";
2018 defined $child or die "no fork: $!";
2020 print $stdin or die $!;
2021 close STDOUT or die $!;
2026 return system (@$args) if $opts->{fork};
2029 warn "ENV size is ".length(join(" ",%ENV));
2030 die "exec failed: $!: @$args";
2034 sub ban_node_by_slot {
2035 # Don't start any new jobsteps on this node for 60 seconds
2037 $slot[$slotid]->{node}->{hold_until} = 60 + scalar time;
2038 $slot[$slotid]->{node}->{hold_count}++;
2039 Log (undef, "backing off node " . $slot[$slotid]->{node}->{name} . " for 60 seconds");
2044 my ($lockfile, $error_message) = @_;
2045 open L, ">", $lockfile or croak("$lockfile: $!");
2046 if (!flock L, LOCK_EX|LOCK_NB) {
2047 croak("Can't lock $lockfile: $error_message\n");
2051 sub find_docker_image {
2052 # Given a Keep locator, check to see if it contains a Docker image.
2053 # If so, return its stream name and Docker hash.
2054 # If not, return undef for both values.
2055 my $locator = shift;
2056 my ($streamname, $filename);
2057 my $image = api_call("collections/get", uuid => $locator);
2059 foreach my $line (split(/\n/, $image->{manifest_text})) {
2060 my @tokens = split(/\s+/, $line);
2062 $streamname = shift(@tokens);
2063 foreach my $filedata (grep(/^\d+:\d+:/, @tokens)) {
2064 if (defined($filename)) {
2065 return (undef, undef); # More than one file in the Collection.
2067 $filename = (split(/:/, $filedata, 3))[2];
2072 if (defined($filename) and ($filename =~ /^([0-9A-Fa-f]{64})\.tar$/)) {
2073 return ($streamname, $1);
2075 return (undef, undef);
2080 # Calculate the number of times an operation should be retried,
2081 # assuming exponential backoff, and that we're willing to retry as
2082 # long as tasks have been running. Enforce a minimum of 3 retries.
2083 my ($starttime, $endtime, $timediff, $retries);
2085 $starttime = $jobstep[0]->{starttime};
2086 $endtime = $jobstep[-1]->{finishtime};
2088 if (!defined($starttime)) {
2090 } elsif (!defined($endtime)) {
2091 $timediff = time - $starttime;
2093 $timediff = ($endtime - $starttime) - (time - $endtime);
2095 if ($timediff > 0) {
2096 $retries = int(log($timediff) / log(2));
2098 $retries = 1; # Use the minimum.
2100 return ($retries > 3) ? $retries : 3;
2104 # Pass in two function references.
2105 # This method will be called with the remaining arguments.
2106 # If it dies, retry it with exponential backoff until it succeeds,
2107 # or until the current retry_count is exhausted. After each failure
2108 # that can be retried, the second function will be called with
2109 # the current try count (0-based), next try time, and error message.
2110 my $operation = shift;
2111 my $retry_callback = shift;
2112 my $retries = retry_count();
2113 foreach my $try_count (0..$retries) {
2114 my $next_try = time + (2 ** $try_count);
2115 my $result = eval { $operation->(@_); };
2118 } elsif ($try_count < $retries) {
2119 $retry_callback->($try_count, $next_try, $@);
2120 my $sleep_time = $next_try - time;
2121 sleep($sleep_time) if ($sleep_time > 0);
2124 # Ensure the error message ends in a newline, so Perl doesn't add
2125 # retry_op's line number to it.
2131 # Pass in a /-separated API method name, and arguments for it.
2132 # This function will call that method, retrying as needed until
2133 # the current retry_count is exhausted, with a log on the first failure.
2134 my $method_name = shift;
2135 my $log_api_retry = sub {
2136 my ($try_count, $next_try_at, $errmsg) = @_;
2137 $errmsg =~ s/\s*\bat \Q$0\E line \d+\.?\s*//;
2138 $errmsg =~ s/\s/ /g;
2139 $errmsg =~ s/\s+$//;
2141 if ($next_try_at < time) {
2142 $retry_msg = "Retrying.";
2144 my $next_try_fmt = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($next_try_at);
2145 $retry_msg = "Retrying at $next_try_fmt.";
2147 Log(undef, "API method $method_name failed: $errmsg. $retry_msg");
2150 foreach my $key (split(/\//, $method_name)) {
2151 $method = $method->{$key};
2153 return retry_op(sub { $method->execute(@_); }, $log_api_retry, @_);
2157 # Given a $?, return a human-readable exit code string like "0" or
2158 # "1" or "0 with signal 1" or "1 with signal 11".
2159 my $exitcode = shift;
2160 my $s = $exitcode >> 8;
2161 if ($exitcode & 0x7f) {
2162 $s .= " with signal " . ($exitcode & 0x7f);
2164 if ($exitcode & 0x80) {
2165 $s .= " with core dump";
2170 sub handle_readall {
2171 # Pass in a glob reference to a file handle.
2172 # Read all its contents and return them as a string.
2173 my $fh_glob_ref = shift;
2175 return <$fh_glob_ref>;
2178 sub tar_filename_n {
2180 return sprintf("%s/git.%s.%d.tar", $ENV{CRUNCH_TMP}, $job_id, $n);
2183 sub add_git_archive {
2184 # Pass in a git archive command as a string or list, a la system().
2185 # This method will save its output to be included in the archive sent to the
2189 if (!open(GIT_ARCHIVE, ">", tar_filename_n($git_tar_count))) {
2190 croak("Failed to save git archive: $!");
2192 my $git_pid = open2(">&GIT_ARCHIVE", $git_input, @_);
2194 waitpid($git_pid, 0);
2197 croak("Failed to save git archive: git exited " . exit_status_s($?));
2201 sub combined_git_archive {
2202 # Combine all saved tar archives into a single archive, then return its
2203 # contents in a string. Return undef if no archives have been saved.
2204 if ($git_tar_count < 1) {
2207 my $base_tar_name = tar_filename_n(1);
2208 foreach my $tar_to_append (map { tar_filename_n($_); } (2..$git_tar_count)) {
2209 my $tar_exit = system("tar", "-Af", $base_tar_name, $tar_to_append);
2210 if ($tar_exit != 0) {
2211 croak("Error preparing build archive: tar -A exited " .
2212 exit_status_s($tar_exit));
2215 if (!open(GIT_TAR, "<", $base_tar_name)) {
2216 croak("Could not open build archive: $!");
2218 my $tar_contents = handle_readall(\*GIT_TAR);
2220 return $tar_contents;
2223 sub set_nonblocking {
2225 my $flags = fcntl ($fh, F_GETFL, 0) or croak ($!);
2226 fcntl ($fh, F_SETFL, $flags | O_NONBLOCK) or croak ($!);
2232 # This is crunch-job's internal dispatch script. crunch-job running on the API
2233 # server invokes this script on individual compute nodes, or localhost if we're
2234 # running a job locally. It gets called in two modes:
2236 # * No arguments: Installation mode. Read a tar archive from the DATA
2237 # file handle; it includes the Crunch script's source code, and
2238 # maybe SDKs as well. Those should be installed in the proper
2239 # locations. This runs outside of any Docker container, so don't try to
2240 # introspect Crunch's runtime environment.
2242 # * With arguments: Crunch script run mode. This script should set up the
2243 # environment, then run the command specified in the arguments. This runs
2244 # inside any Docker container.
2247 use File::Path qw( make_path remove_tree );
2248 use POSIX qw(getcwd);
2250 use constant TASK_TEMPFAIL => 111;
2252 # Map SDK subdirectories to the path environments they belong to.
2253 my %SDK_ENVVARS = ("perl/lib" => "PERLLIB", "ruby/lib" => "RUBYLIB");
2255 my $destdir = $ENV{"CRUNCH_SRC"};
2256 my $archive_hash = $ENV{"CRUNCH_GIT_ARCHIVE_HASH"};
2257 my $repo = $ENV{"CRUNCH_SRC_URL"};
2258 my $install_dir = $ENV{"CRUNCH_INSTALL"} || (getcwd() . "/opt");
2259 my $job_work = $ENV{"JOB_WORK"};
2260 my $task_work = $ENV{"TASK_WORK"};
2262 open(STDOUT_ORIG, ">&", STDOUT);
2263 open(STDERR_ORIG, ">&", STDERR);
2265 for my $dir ($destdir, $job_work, $task_work) {
2268 -e $dir or die "Failed to create temporary directory ($dir): $!";
2273 remove_tree($task_work, {keep_root => 1});
2276 ### Crunch script run mode
2278 # We want to do routine logging during task 0 only. This gives the user
2279 # the information they need, but avoids repeating the information for every
2282 if ($ENV{TASK_SEQUENCE} eq "0") {
2285 printf STDERR_ORIG "[Crunch] $msg\n", @_;
2291 my $python_src = "$install_dir/python";
2292 my $venv_dir = "$job_work/.arvados.venv";
2293 my $venv_built = -e "$venv_dir/bin/activate";
2294 if ((!$venv_built) and (-d $python_src) and can_run("virtualenv")) {
2295 shell_or_die(undef, "virtualenv", "--quiet", "--system-site-packages",
2296 "--python=python2.7", $venv_dir);
2297 shell_or_die(TASK_TEMPFAIL, "$venv_dir/bin/pip", "--quiet", "install", "-I", $python_src);
2299 $Log->("Built Python SDK virtualenv");
2302 my @pysdk_version_cmd = ("python", "-c",
2303 "from pkg_resources import get_distribution as get; print get('arvados-python-client').version");
2305 $Log->("Running in Python SDK virtualenv");
2306 @pysdk_version_cmd = ();
2307 my $orig_argv = join(" ", map { quotemeta($_); } @ARGV);
2308 @ARGV = ("/bin/sh", "-ec",
2309 ". \Q$venv_dir/bin/activate\E; exec $orig_argv");
2310 } elsif (-d $python_src) {
2311 $Log->("Warning: virtualenv not found inside Docker container default " .
2312 "\$PATH. Can't install Python SDK.");
2315 if (@pysdk_version_cmd) {
2316 open(my $pysdk_version_pipe, "-|", @pysdk_version_cmd);
2317 my $pysdk_version = <$pysdk_version_pipe>;
2318 close($pysdk_version_pipe);
2320 chomp($pysdk_version);
2321 $Log->("Using Arvados SDK version $pysdk_version");
2323 # A lot could've gone wrong here, but pretty much all of it means that
2324 # Python won't be able to load the Arvados SDK.
2325 $Log->("Warning: Arvados SDK not found");
2329 while (my ($sdk_dir, $sdk_envkey) = each(%SDK_ENVVARS)) {
2330 my $sdk_path = "$install_dir/$sdk_dir";
2332 if ($ENV{$sdk_envkey}) {
2333 $ENV{$sdk_envkey} = "$sdk_path:" . $ENV{$sdk_envkey};
2335 $ENV{$sdk_envkey} = $sdk_path;
2337 $Log->("Arvados SDK added to %s", $sdk_envkey);
2342 die "Cannot exec `@ARGV`: $!";
2345 ### Installation mode
2346 open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
2348 if (readlink ("$destdir.archive_hash") eq $archive_hash && -d $destdir) {
2349 # This exact git archive (source + arvados sdk) is already installed
2350 # here, so there's no need to reinstall it.
2352 # We must consume our DATA section, though: otherwise the process
2353 # feeding it to us will get SIGPIPE.
2355 while (read(DATA, $buf, 65536)) { }
2360 unlink "$destdir.archive_hash";
2364 # Ignore SIGPIPE: we check retval of close() instead. See perlipc(1).
2365 local $SIG{PIPE} = "IGNORE";
2366 warn "Extracting archive: $archive_hash\n";
2367 # --ignore-zeros is necessary sometimes: depending on how much NUL
2368 # padding tar -A put on our combined archive (which in turn depends
2369 # on the length of the component archives) tar without
2370 # --ignore-zeros will exit before consuming stdin and cause close()
2371 # to fail on the resulting SIGPIPE.
2372 if (!open(TARX, "|-", "tar", "--ignore-zeros", "-xC", $destdir)) {
2373 die "Error launching 'tar -xC $destdir': $!";
2375 # If we send too much data to tar in one write (> 4-5 MiB), it stops, and we
2376 # get SIGPIPE. We must feed it data incrementally.
2378 while (read(DATA, $tar_input, 65536)) {
2379 print TARX $tar_input;
2382 die "'tar -xC $destdir' exited $?: $!";
2388 my $sdk_root = "$destdir/.arvados.sdk/sdk";
2390 foreach my $sdk_lang (("python",
2391 map { (split /\//, $_, 2)[0]; } keys(%SDK_ENVVARS))) {
2392 if (-d "$sdk_root/$sdk_lang") {
2393 if (!rename("$sdk_root/$sdk_lang", "$install_dir/$sdk_lang")) {
2394 die "Failed to install $sdk_lang SDK: $!";
2400 my $python_dir = "$install_dir/python";
2401 if ((-d $python_dir) and can_run("python2.7")) {
2402 open(my $egg_info_pipe, "-|",
2403 "python2.7 \Q$python_dir/setup.py\E egg_info 2>&1 >/dev/null");
2404 my @egg_info_errors = <$egg_info_pipe>;
2405 close($egg_info_pipe);
2408 if (@egg_info_errors and (($egg_info_errors[-1] =~ /\bgit\b/) or ($egg_info_errors[-1] =~ /\[Errno 2\]/))) {
2409 # egg_info apparently failed because it couldn't ask git for a build tag.
2410 # Specify no build tag.
2411 open(my $pysdk_cfg, ">>", "$python_dir/setup.cfg");
2412 print $pysdk_cfg "\n[egg_info]\ntag_build =\n";
2415 my $egg_info_exit = $? >> 8;
2416 foreach my $errline (@egg_info_errors) {
2419 warn "python setup.py egg_info failed: exit $egg_info_exit";
2420 exit ($egg_info_exit || 1);
2425 # Hide messages from the install script (unless it fails: shell_or_die
2426 # will show $destdir.log in that case).
2427 open(STDOUT, ">>", "$destdir.log");
2428 open(STDERR, ">&", STDOUT);
2430 if (-e "$destdir/crunch_scripts/install") {
2431 shell_or_die (undef, "$destdir/crunch_scripts/install", $install_dir);
2432 } elsif (!-e "./install.sh" && -e "./tests/autotests.sh") {
2434 shell_or_die (undef, "./tests/autotests.sh", $install_dir);
2435 } elsif (-e "./install.sh") {
2436 shell_or_die (undef, "./install.sh", $install_dir);
2439 if ($archive_hash) {
2440 unlink "$destdir.archive_hash.new";
2441 symlink ($archive_hash, "$destdir.archive_hash.new") or die "$destdir.archive_hash.new: $!";
2442 rename ("$destdir.archive_hash.new", "$destdir.archive_hash") or die "$destdir.archive_hash: $!";
2448 my $command_name = shift;
2449 open(my $which, "-|", "which", $command_name);
2450 while (<$which>) { }
2457 my $exitcode = shift;
2459 if ($ENV{"DEBUG"}) {
2460 print STDERR "@_\n";
2462 if (system (@_) != 0) {
2465 my $exitstatus = sprintf("exit %d signal %d", $code >> 8, $code & 0x7f);
2466 open STDERR, ">&STDERR_ORIG";
2467 system ("cat $destdir.log >&2");
2468 warn "@_ failed ($err): $exitstatus";
2469 if (defined($exitcode)) {
2473 exit (($code >> 8) || 1);