2 # -*- mode: perl; perl-indent-level: 2; indent-tabs-mode: nil; -*-
6 crunch-job: Execute job steps, save snapshots as requested, collate output.
10 Obtain job details from Arvados, run tasks on compute nodes (typically
11 invoked by scheduler on controller):
13 crunch-job --job x-y-z --git-dir /path/to/repo/.git
15 Obtain job details from command line, run tasks on local machine
16 (typically invoked by application or developer on VM):
18 crunch-job --job '{"script_version":"/path/to/working/tree","script":"scriptname",...}'
20 crunch-job --job '{"repository":"https://github.com/curoverse/arvados.git","script_version":"master","script":"scriptname",...}'
28 If the job is already locked, steal the lock and run it anyway.
32 Path to a .git directory (or a git URL) where the commit given in the
33 job's C<script_version> attribute is to be found. If this is I<not>
34 given, the job's C<repository> attribute will be used.
38 Arvados API authorization token to use during the course of the job.
42 Do not clear per-job/task temporary directories during initial job
43 setup. This can speed up development and debugging when running jobs
48 UUID of the job to run, or a JSON-encoded job resource without a
49 UUID. If the latter is given, a new job object will be created.
53 =head1 RUNNING JOBS LOCALLY
55 crunch-job's log messages appear on stderr along with the job tasks'
56 stderr streams. The log is saved in Keep at each checkpoint and when
59 If the job succeeds, the job's output locator is printed on stdout.
61 While the job is running, the following signals are accepted:
65 =item control-C, SIGINT, SIGQUIT
67 Save a checkpoint, terminate any job tasks that are running, and stop.
71 Save a checkpoint and continue.
75 Refresh node allocation (i.e., check whether any nodes have been added
76 or unallocated) and attributes of the Job record that should affect
77 behavior (e.g., cancel job if cancelled_at becomes non-nil).
85 use POSIX ':sys_wait_h';
86 use POSIX qw(strftime);
87 use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK);
91 use Digest::MD5 qw(md5_hex);
97 use File::Path qw( make_path remove_tree );
99 use constant TASK_TEMPFAIL => 111;
100 use constant EX_TEMPFAIL => 75;
101 use constant EX_RETRY_UNLOCKED => 93;
103 $ENV{"TMPDIR"} ||= "/tmp";
104 unless (defined $ENV{"CRUNCH_TMP"}) {
105 $ENV{"CRUNCH_TMP"} = $ENV{"TMPDIR"} . "/crunch-job";
106 if ($ENV{"USER"} ne "crunch" && $< != 0) {
107 # use a tmp dir unique for my uid
108 $ENV{"CRUNCH_TMP"} .= "-$<";
112 # Create the tmp directory if it does not exist
113 if ( ! -d $ENV{"CRUNCH_TMP"} ) {
114 make_path $ENV{"CRUNCH_TMP"} or die "Failed to create temporary working directory: " . $ENV{"CRUNCH_TMP"};
117 $ENV{"JOB_WORK"} = $ENV{"CRUNCH_TMP"} . "/work";
118 $ENV{"CRUNCH_INSTALL"} = "$ENV{CRUNCH_TMP}/opt";
119 $ENV{"CRUNCH_WORK"} = $ENV{"JOB_WORK"}; # deprecated
120 mkdir ($ENV{"JOB_WORK"});
129 my $docker_bin = "docker.io";
130 my $docker_run_args = "";
131 GetOptions('force-unlock' => \$force_unlock,
132 'git-dir=s' => \$git_dir,
133 'job=s' => \$jobspec,
134 'job-api-token=s' => \$job_api_token,
135 'no-clear-tmp' => \$no_clear_tmp,
136 'resume-stash=s' => \$resume_stash,
137 'docker-bin=s' => \$docker_bin,
138 'docker-run-args=s' => \$docker_run_args,
141 if (defined $job_api_token) {
142 $ENV{ARVADOS_API_TOKEN} = $job_api_token;
145 my $have_slurm = exists $ENV{SLURM_JOB_ID} && exists $ENV{SLURM_NODELIST};
150 $main::ENV{CRUNCH_DEBUG} = 1;
154 $main::ENV{CRUNCH_DEBUG} = 0;
157 my $arv = Arvados->new('apiVersion' => 'v1');
166 if ($jobspec =~ /^[-a-z\d]+$/)
168 # $jobspec is an Arvados UUID, not a JSON job specification
169 $Job = api_call("jobs/get", uuid => $jobspec);
174 $local_job = JSON::decode_json($jobspec);
178 # Make sure our workers (our slurm nodes, localhost, or whatever) are
179 # at least able to run basic commands: they aren't down or severely
182 if (($Job || $local_job)->{docker_image_locator}) {
183 $cmd = [$docker_bin, 'ps', '-q'];
185 Log(undef, "Sanity check is `@$cmd`");
186 my ($exited, $stdout, $stderr) = srun_sync(
187 ["srun", "--nodes=\Q$ENV{SLURM_NNODES}\E", "--ntasks-per-node=1"],
189 {label => "sanity check"});
191 Log(undef, "Sanity check failed: ".exit_status_s($exited));
194 Log(undef, "Sanity check OK");
197 my $User = api_call("users/current");
200 if (!$force_unlock) {
201 # Claim this job, and make sure nobody else does
202 eval { api_call("jobs/lock", uuid => $Job->{uuid}); };
204 Log(undef, "Error while locking job, exiting ".EX_TEMPFAIL);
213 map { croak ("No $_ specified") unless $local_job->{$_} }
214 qw(script script_version script_parameters);
217 $local_job->{'is_locked_by_uuid'} = $User->{'uuid'};
218 $local_job->{'started_at'} = gmtime;
219 $local_job->{'state'} = 'Running';
221 $Job = api_call("jobs/create", job => $local_job);
223 $job_id = $Job->{'uuid'};
225 my $keep_logfile = $job_id . '.log.txt';
226 log_writer_start($keep_logfile);
228 $Job->{'runtime_constraints'} ||= {};
229 $Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0;
230 my $max_ncpus = $Job->{'runtime_constraints'}->{'max_tasks_per_node'};
232 my $gem_versions = `gem list --quiet arvados-cli 2>/dev/null`;
234 $gem_versions =~ s/^arvados-cli \(/ with arvados-cli Gem version(s) /;
235 chomp($gem_versions);
236 chop($gem_versions); # Closing parentheses
241 "running from " . ((-e $0) ? realpath($0) : "stdin") . $gem_versions);
243 Log (undef, "check slurm allocation");
246 # Should use $ENV{SLURM_TASKS_PER_NODE} instead of sinfo? (eg. "4(x3),2,4(x2)")
250 my $localcpus = 0 + `grep -cw ^processor /proc/cpuinfo` || 1;
251 push @sinfo, "$localcpus localhost";
253 if (exists $ENV{SLURM_NODELIST})
255 push @sinfo, `sinfo -h --format='%c %N' --nodes=\Q$ENV{SLURM_NODELIST}\E`;
259 my ($ncpus, $slurm_nodelist) = split;
260 $ncpus = $max_ncpus if $max_ncpus && $ncpus > $max_ncpus;
263 while ($slurm_nodelist =~ s/^([^\[,]+?(\[.*?\])?)(,|$)//)
266 if ($nodelist =~ /\[((\d+)(-(\d+))?(,(\d+)(-(\d+))?)*)\]/)
269 foreach (split (",", $ranges))
282 push @nodelist, map {
284 $n =~ s/\[[-,\d]+\]/$_/;
291 push @nodelist, $nodelist;
294 foreach my $nodename (@nodelist)
296 Log (undef, "node $nodename - $ncpus slots");
297 my $node = { name => $nodename,
299 # The number of consecutive times a task has been dispatched
300 # to this node and failed.
302 # The number of consecutive times that SLURM has reported
303 # a node failure since the last successful task.
305 # Don't dispatch work to this node until this time
306 # (in seconds since the epoch) has passed.
308 foreach my $cpu (1..$ncpus)
310 push @slot, { node => $node,
314 push @node, @nodelist;
319 # Ensure that we get one jobstep running on each allocated node before
320 # we start overloading nodes with concurrent steps
322 @slot = sort { $a->{cpu} <=> $b->{cpu} } @slot;
325 $Job->update_attributes(
326 'tasks_summary' => { 'failed' => 0,
331 Log (undef, "start");
332 $SIG{'INT'} = sub { $main::please_freeze = 1; };
333 $SIG{'QUIT'} = sub { $main::please_freeze = 1; };
334 $SIG{'TERM'} = \&croak;
335 $SIG{'TSTP'} = sub { $main::please_freeze = 1; };
336 $SIG{'ALRM'} = sub { $main::please_info = 1; };
337 $SIG{'CONT'} = sub { $main::please_continue = 1; };
338 $SIG{'HUP'} = sub { $main::please_refresh = 1; };
340 $main::please_freeze = 0;
341 $main::please_info = 0;
342 $main::please_continue = 0;
343 $main::please_refresh = 0;
344 my $jobsteps_must_output_keys = 0; # becomes 1 when any task outputs a key
346 grep { $ENV{$1} = $2 if /^(NOCACHE.*?)=(.*)/ } split ("\n", $$Job{knobs});
347 $ENV{"CRUNCH_JOB_UUID"} = $job_id;
348 $ENV{"JOB_UUID"} = $job_id;
351 my @jobstep_todo = ();
352 my @jobstep_done = ();
353 my @jobstep_tomerge = ();
354 my $jobstep_tomerge_level = 0;
355 my $squeue_checked = 0;
356 my $latest_refresh = scalar time;
360 if (defined $Job->{thawedfromkey})
362 thaw ($Job->{thawedfromkey});
366 my $first_task = api_call("job_tasks/create", job_task => {
367 'job_uuid' => $Job->{'uuid'},
372 push @jobstep, { 'level' => 0,
374 'arvados_task' => $first_task,
376 push @jobstep_todo, 0;
382 must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
385 my $build_script = handle_readall(\*DATA);
386 my $nodelist = join(",", @node);
387 my $git_tar_count = 0;
389 if (!defined $no_clear_tmp) {
390 # Find FUSE mounts under $CRUNCH_TMP and unmount them. Then clean
391 # up work directories crunch_tmp/work, crunch_tmp/opt,
394 # TODO: When #5036 is done and widely deployed, we can limit mount's
395 # -t option to simply fuse.keep.
396 my ($exited, $stdout, $stderr) = srun_sync(
397 ["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
398 ['bash', '-ec', '-o', 'pipefail', 'mount -t fuse,fuse.keep | awk "(index(\$3, \"$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* $CRUNCH_TMP/*.cid'],
399 {label => "clean work dirs"});
401 exit(EX_RETRY_UNLOCKED);
405 # If this job requires a Docker image, install that.
406 my ($docker_locator, $docker_stream, $docker_hash, $docker_limitmem, $dockeruserarg);
407 if ($docker_locator = $Job->{docker_image_locator}) {
408 Log (undef, "Install docker image $docker_locator");
409 ($docker_stream, $docker_hash) = find_docker_image($docker_locator);
412 croak("No Docker image hash found from locator $docker_locator");
414 Log (undef, "docker image hash is $docker_hash");
415 $docker_stream =~ s/^\.//;
416 my $docker_install_script = qq{
417 if ! $docker_bin images -q --no-trunc --all | grep -qxF \Q$docker_hash\E; then
418 arv-get \Q$docker_locator$docker_stream/$docker_hash.tar\E | $docker_bin load
422 my ($exited, $stdout, $stderr) = srun_sync(
423 ["srun", "--nodelist=" . join(',', @node)],
424 ["/bin/bash", "-o", "pipefail", "-ec", $docker_install_script],
425 {label => "load docker image"});
428 exit(EX_RETRY_UNLOCKED);
431 # Determine whether this version of Docker supports memory+swap limits.
432 ($exited, $stdout, $stderr) = srun_sync(
433 ["srun", "--nodelist=" . $node[0]],
434 [$docker_bin, 'run', '--help'],
435 {label => "check --memory-swap feature"});
436 $docker_limitmem = ($stdout =~ /--memory-swap/);
438 # Find a non-root Docker user to use.
439 # Tries the default user for the container, then 'crunch', then 'nobody',
440 # testing for whether the actual user id is non-zero. This defends against
441 # mistakes but not malice, but we intend to harden the security in the future
442 # so we don't want anyone getting used to their jobs running as root in their
444 my @tryusers = ("", "crunch", "nobody");
445 foreach my $try_user (@tryusers) {
448 if ($try_user eq "") {
449 $label = "check whether default user is UID 0";
452 $label = "check whether user '$try_user' is UID 0";
453 $try_user_arg = "--user=$try_user";
455 my ($exited, $stdout, $stderr) = srun_sync(
456 ["srun", "--nodelist=" . $node[0]],
458 "$docker_bin run $docker_run_args $try_user_arg $docker_hash id --user"],
461 if ($exited == 0 && $stdout =~ /^\d+$/ && $stdout > 0) {
462 $dockeruserarg = $try_user_arg;
463 if ($try_user eq "") {
464 Log(undef, "Container will run with default user");
466 Log(undef, "Container will run with $dockeruserarg");
472 if (!defined $dockeruserarg) {
473 croak("Could not find a user in container that is not UID 0 (tried default user, @tryusers) or there was a problem running 'id' in the container.");
476 if ($Job->{arvados_sdk_version}) {
477 # The job also specifies an Arvados SDK version. Add the SDKs to the
478 # tar file for the build script to install.
479 Log(undef, sprintf("Packing Arvados SDK version %s for installation",
480 $Job->{arvados_sdk_version}));
481 add_git_archive("git", "--git-dir=$git_dir", "archive",
482 "--prefix=.arvados.sdk/",
483 $Job->{arvados_sdk_version}, "sdk");
487 if (!defined $git_dir && $Job->{'script_version'} =~ m{^/}) {
488 # If script_version looks like an absolute path, *and* the --git-dir
489 # argument was not given -- which implies we were not invoked by
490 # crunch-dispatch -- we will use the given path as a working
491 # directory instead of resolving script_version to a git commit (or
492 # doing anything else with git).
493 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{'script_version'};
494 $ENV{"CRUNCH_SRC"} = $Job->{'script_version'};
497 # Resolve the given script_version to a git commit sha1. Also, if
498 # the repository is remote, clone it into our local filesystem: this
499 # ensures "git archive" will work, and is necessary to reliably
500 # resolve a symbolic script_version like "master^".
501 $ENV{"CRUNCH_SRC"} = "$ENV{CRUNCH_TMP}/src";
503 Log (undef, "Looking for version ".$Job->{script_version}." from repository ".$Job->{repository});
505 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{script_version};
507 # If we're running under crunch-dispatch, it will have already
508 # pulled the appropriate source tree into its own repository, and
509 # given us that repo's path as $git_dir.
511 # If we're running a "local" job, we might have to fetch content
512 # from a remote repository.
514 # (Currently crunch-dispatch gives a local path with --git-dir, but
515 # we might as well accept URLs there too in case it changes its
517 my $repo = $git_dir || $Job->{'repository'};
519 # Repository can be remote or local. If remote, we'll need to fetch it
520 # to a local dir before doing `git log` et al.
523 if ($repo =~ m{://|^[^/]*:}) {
524 # $repo is a git url we can clone, like git:// or https:// or
525 # file:/// or [user@]host:repo.git. Note "user/name@host:foo" is
526 # not recognized here because distinguishing that from a local
527 # path is too fragile. If you really need something strange here,
528 # use the ssh:// form.
529 $repo_location = 'remote';
530 } elsif ($repo =~ m{^\.*/}) {
531 # $repo is a local path to a git index. We'll also resolve ../foo
532 # to ../foo/.git if the latter is a directory. To help
533 # disambiguate local paths from named hosted repositories, this
534 # form must be given as ./ or ../ if it's a relative path.
535 if (-d "$repo/.git") {
536 $repo = "$repo/.git";
538 $repo_location = 'local';
540 # $repo is none of the above. It must be the name of a hosted
542 my $arv_repo_list = api_call("repositories/list",
543 'filters' => [['name','=',$repo]]);
544 my @repos_found = @{$arv_repo_list->{'items'}};
545 my $n_found = $arv_repo_list->{'serverResponse'}->{'items_available'};
547 Log(undef, "Repository '$repo' -> "
548 . join(", ", map { $_->{'uuid'} } @repos_found));
551 croak("Error: Found $n_found repositories with name '$repo'.");
553 $repo = $repos_found[0]->{'fetch_url'};
554 $repo_location = 'remote';
556 Log(undef, "Using $repo_location repository '$repo'");
557 $ENV{"CRUNCH_SRC_URL"} = $repo;
559 # Resolve given script_version (we'll call that $treeish here) to a
560 # commit sha1 ($commit).
561 my $treeish = $Job->{'script_version'};
563 if ($repo_location eq 'remote') {
564 # We minimize excess object-fetching by re-using the same bare
565 # repository in CRUNCH_TMP/.git for multiple crunch-jobs -- we
566 # just keep adding remotes to it as needed.
567 my $local_repo = $ENV{'CRUNCH_TMP'}."/.git";
568 my $gitcmd = "git --git-dir=\Q$local_repo\E";
570 # Set up our local repo for caching remote objects, making
572 if (!-d $local_repo) {
573 make_path($local_repo) or croak("Error: could not create $local_repo");
575 # This works (exits 0 and doesn't delete fetched objects) even
576 # if $local_repo is already initialized:
577 `$gitcmd init --bare`;
579 croak("Error: $gitcmd init --bare exited ".exit_status_s($?));
582 # If $treeish looks like a hash (or abbrev hash) we look it up in
583 # our local cache first, since that's cheaper. (We don't want to
584 # do that with tags/branches though -- those change over time, so
585 # they should always be resolved by the remote repo.)
586 if ($treeish =~ /^[0-9a-f]{7,40}$/s) {
587 # Hide stderr because it's normal for this to fail:
588 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E 2>/dev/null`;
590 # Careful not to resolve a branch named abcdeff to commit 1234567:
591 $sha1 =~ /^$treeish/ &&
592 $sha1 =~ /^([0-9a-f]{40})$/s) {
594 Log(undef, "Commit $commit already present in $local_repo");
598 if (!defined $commit) {
599 # If $treeish isn't just a hash or abbrev hash, or isn't here
600 # yet, we need to fetch the remote to resolve it correctly.
602 # First, remove all local heads. This prevents a name that does
603 # not exist on the remote from resolving to (or colliding with)
604 # a previously fetched branch or tag (possibly from a different
606 remove_tree("$local_repo/refs/heads", {keep_root => 1});
608 Log(undef, "Fetching objects from $repo to $local_repo");
609 `$gitcmd fetch --no-progress --tags ''\Q$repo\E \Q+refs/heads/*:refs/heads/*\E`;
611 croak("Error: `$gitcmd fetch` exited ".exit_status_s($?));
615 # Now that the data is all here, we will use our local repo for
616 # the rest of our git activities.
620 my $gitcmd = "git --git-dir=\Q$repo\E";
621 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E`;
622 unless ($? == 0 && $sha1 =~ /^([0-9a-f]{40})$/) {
623 croak("`$gitcmd rev-list` exited "
625 .", '$treeish' not found, giving up");
628 Log(undef, "Version $treeish is commit $commit");
630 if ($commit ne $Job->{'script_version'}) {
631 # Record the real commit id in the database, frozentokey, logs,
632 # etc. -- instead of an abbreviation or a branch name which can
633 # become ambiguous or point to a different commit in the future.
634 if (!$Job->update_attributes('script_version' => $commit)) {
635 croak("Error: failed to update job's script_version attribute");
639 $ENV{"CRUNCH_SRC_COMMIT"} = $commit;
640 add_git_archive("$gitcmd archive ''\Q$commit\E");
643 my $git_archive = combined_git_archive();
644 if (!defined $git_archive) {
645 Log(undef, "Skip install phase (no git archive)");
647 Log(undef, "Warning: This probably means workers have no source tree!");
652 my $install_script_tries_left = 3;
653 for (my $attempts = 0; $attempts < 3; $attempts++) {
654 my @srunargs = ("srun",
655 "--nodelist=$nodelist",
656 "-D", $ENV{'TMPDIR'}, "--job-name=$job_id");
657 my @execargs = ("sh", "-c",
658 "mkdir -p $ENV{CRUNCH_INSTALL} && cd $ENV{CRUNCH_TMP} && perl -");
660 $ENV{"CRUNCH_GIT_ARCHIVE_HASH"} = md5_hex($git_archive);
661 my ($stdout, $stderr);
662 ($exited, $stdout, $stderr) = srun_sync(
663 \@srunargs, \@execargs,
664 {label => "run install script on all workers"},
665 $build_script . $git_archive);
667 my $stderr_anything_from_script = 0;
668 for my $line (split(/\n/, $stderr)) {
669 if ($line !~ /^(srun: error: |starting: \[)/) {
670 $stderr_anything_from_script = 1;
674 last if $exited == 0 || $main::please_freeze;
676 # If the install script fails but doesn't print an error message,
677 # the next thing anyone is likely to do is just run it again in
678 # case it was a transient problem like "slurm communication fails
679 # because the network isn't reliable enough". So we'll just do
680 # that ourselves (up to 3 attempts in total). OTOH, if there is an
681 # error message, the problem is more likely to have a real fix and
682 # we should fail the job so the fixing process can start, instead
683 # of doing 2 more attempts.
684 last if $stderr_anything_from_script;
687 foreach my $tar_filename (map { tar_filename_n($_); } (1..$git_tar_count)) {
688 unlink($tar_filename);
696 foreach (qw (script script_version script_parameters runtime_constraints))
700 (ref($Job->{$_}) ? JSON::encode_json($Job->{$_}) : $Job->{$_}));
702 foreach (split (/\n/, $Job->{knobs}))
704 Log (undef, "knob " . $_);
709 $main::success = undef;
715 my $thisround_succeeded = 0;
716 my $thisround_failed = 0;
717 my $thisround_failed_multiple = 0;
718 my $working_slot_count = scalar(@slot);
720 @jobstep_todo = sort { $jobstep[$a]->{level} <=> $jobstep[$b]->{level}
721 or $a <=> $b } @jobstep_todo;
722 my $level = $jobstep[$jobstep_todo[0]]->{level};
724 my $initial_tasks_this_level = 0;
725 foreach my $id (@jobstep_todo) {
726 $initial_tasks_this_level++ if ($jobstep[$id]->{level} == $level);
729 # If the number of tasks scheduled at this level #T is smaller than the number
730 # of slots available #S, only use the first #T slots, or the first slot on
731 # each node, whichever number is greater.
733 # When we dispatch tasks later, we'll allocate whole-node resources like RAM
734 # based on these numbers. Using fewer slots makes more resources available
735 # to each individual task, which should normally be a better strategy when
736 # there are fewer of them running with less parallelism.
738 # Note that this calculation is not redone if the initial tasks at
739 # this level queue more tasks at the same level. This may harm
740 # overall task throughput for that level.
742 if ($initial_tasks_this_level < @node) {
743 @freeslot = (0..$#node);
744 } elsif ($initial_tasks_this_level < @slot) {
745 @freeslot = (0..$initial_tasks_this_level - 1);
747 @freeslot = (0..$#slot);
749 my $round_num_freeslots = scalar(@freeslot);
750 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";
752 my %round_max_slots = ();
753 for (my $ii = $#freeslot; $ii >= 0; $ii--) {
754 my $this_slot = $slot[$freeslot[$ii]];
755 my $node_name = $this_slot->{node}->{name};
756 $round_max_slots{$node_name} ||= $this_slot->{cpu};
757 last if (scalar(keys(%round_max_slots)) >= @node);
760 Log(undef, "start level $level with $round_num_freeslots slots");
763 my $progress_is_dirty = 1;
764 my $progress_stats_updated = 0;
766 update_progress_stats();
770 for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
772 # Don't create new tasks if we already know the job's final result.
773 last if defined($main::success);
775 my $id = $jobstep_todo[$todo_ptr];
776 my $Jobstep = $jobstep[$id];
777 if ($Jobstep->{level} != $level)
782 pipe $reader{$id}, "writer" or croak("pipe() failed: $!");
783 set_nonblocking($reader{$id});
785 my $childslot = $freeslot[0];
786 my $childnode = $slot[$childslot]->{node};
787 my $childslotname = join (".",
788 $slot[$childslot]->{node}->{name},
789 $slot[$childslot]->{cpu});
791 my $childpid = fork();
794 $SIG{'INT'} = 'DEFAULT';
795 $SIG{'QUIT'} = 'DEFAULT';
796 $SIG{'TERM'} = 'DEFAULT';
798 foreach (values (%reader))
802 fcntl ("writer", F_SETFL, 0) or croak ($!); # no close-on-exec
803 open(STDOUT,">&writer");
804 open(STDERR,">&writer");
809 delete $ENV{"GNUPGHOME"};
810 $ENV{"TASK_UUID"} = $Jobstep->{'arvados_task'}->{'uuid'};
811 $ENV{"TASK_QSEQUENCE"} = $id;
812 $ENV{"TASK_SEQUENCE"} = $level;
813 $ENV{"JOB_SCRIPT"} = $Job->{script};
814 while (my ($param, $value) = each %{$Job->{script_parameters}}) {
815 $param =~ tr/a-z/A-Z/;
816 $ENV{"JOB_PARAMETER_$param"} = $value;
818 $ENV{"TASK_SLOT_NODE"} = $slot[$childslot]->{node}->{name};
819 $ENV{"TASK_SLOT_NUMBER"} = $slot[$childslot]->{cpu};
820 $ENV{"TASK_WORK"} = $ENV{"CRUNCH_TMP"}."/task/$childslotname";
821 $ENV{"HOME"} = $ENV{"TASK_WORK"};
822 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
823 $ENV{"CRUNCH_NODE_SLOTS"} = $round_max_slots{$ENV{TASK_SLOT_NODE}};
824 $ENV{"PATH"} = $ENV{"CRUNCH_INSTALL"} . "/bin:" . $ENV{"PATH"};
826 my $keep_mnt = $ENV{"TASK_WORK"}.".keep";
832 "--nodelist=".$childnode->{name},
833 qw(-n1 -c1 -N1 -D), $ENV{'TMPDIR'},
834 "--job-name=$job_id.$id.$$",
837 my $stdbuf = " stdbuf --output=0 --error=0 ";
839 my $arv_file_cache = "";
840 if (defined($Job->{'runtime_constraints'}->{'keep_cache_mb_per_task'})) {
841 $arv_file_cache = "--file-cache=" . ($Job->{'runtime_constraints'}->{'keep_cache_mb_per_task'} * 1024 * 1024);
845 "if [ -e \Q$ENV{TASK_WORK}\E ]; then rm -rf \Q$ENV{TASK_WORK}\E; fi; "
846 ."mkdir -p \Q$ENV{CRUNCH_TMP}\E \Q$ENV{JOB_WORK}\E \Q$ENV{TASK_WORK}\E \Q$keep_mnt\E "
847 ."&& cd \Q$ENV{CRUNCH_TMP}\E "
848 # These environment variables get used explicitly later in
849 # $command. No tool is expected to read these values directly.
850 .q{&& MEM=$(awk '($1 == "MemTotal:"){print $2}' </proc/meminfo) }
851 .q{&& SWAP=$(awk '($1 == "SwapTotal:"){print $2}' </proc/meminfo) }
852 ."&& MEMLIMIT=\$(( (\$MEM * 95) / ($ENV{CRUNCH_NODE_SLOTS} * 100) )) "
853 ."&& let SWAPLIMIT=\$MEMLIMIT+\$SWAP ";
855 $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 ";
856 $ENV{TASK_KEEPMOUNT} = "$keep_mnt/by_pdh";
857 $ENV{TASK_KEEPMOUNT_TMP} = "$keep_mnt/tmp";
861 my $containername = "$Jobstep->{arvados_task}->{uuid}-$Jobstep->{failures}";
862 my $cidfile = "$ENV{CRUNCH_TMP}/$containername.cid";
863 $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -cgroup-parent=docker -cgroup-cid=$cidfile -poll=10000 ";
864 $command .= "$docker_bin run $docker_run_args --name=$containername --attach=stdout --attach=stderr --attach=stdin -i \Q$dockeruserarg\E --cidfile=$cidfile --sig-proxy ";
865 # We only set memory limits if Docker lets us limit both memory and swap.
866 # Memory limits alone have been supported longer, but subprocesses tend
867 # to get SIGKILL if they exceed that without any swap limit set.
868 # See #5642 for additional background.
869 if ($docker_limitmem) {
870 $command .= "--memory=\${MEMLIMIT}k --memory-swap=\${SWAPLIMIT}k ";
873 # The source tree and $destdir directory (which we have
874 # installed on the worker host) are available in the container,
875 # under the same path.
876 $command .= "--volume=\Q$ENV{CRUNCH_SRC}:$ENV{CRUNCH_SRC}:ro\E ";
877 $command .= "--volume=\Q$ENV{CRUNCH_INSTALL}:$ENV{CRUNCH_INSTALL}:ro\E ";
879 # Currently, we make the "by_pdh" directory in arv-mount's mount
880 # point appear at /keep inside the container (instead of using
881 # the same path as the host like we do with CRUNCH_SRC and
882 # CRUNCH_INSTALL). However, crunch scripts and utilities must
883 # not rely on this. They must use $TASK_KEEPMOUNT.
884 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT}:/keep:ro\E ";
885 $ENV{TASK_KEEPMOUNT} = "/keep";
887 # Ditto TASK_KEEPMOUNT_TMP, as /keep_tmp.
888 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT_TMP}:/keep_tmp\E ";
889 $ENV{TASK_KEEPMOUNT_TMP} = "/keep_tmp";
891 # TASK_WORK is almost exactly like a docker data volume: it
892 # starts out empty, is writable, and persists until no
893 # containers use it any more. We don't use --volumes-from to
894 # share it with other containers: it is only accessible to this
895 # task, and it goes away when this task stops.
897 # However, a docker data volume is writable only by root unless
898 # the mount point already happens to exist in the container with
899 # different permissions. Therefore, we [1] assume /tmp already
900 # exists in the image and is writable by the crunch user; [2]
901 # avoid putting TASK_WORK inside CRUNCH_TMP (which won't be
902 # writable if they are created by docker while setting up the
903 # other --volumes); and [3] create $TASK_WORK inside the
904 # container using $build_script.
905 $command .= "--volume=/tmp ";
906 $ENV{"TASK_WORK"} = "/tmp/crunch-job-task-work/$childslotname";
907 $ENV{"HOME"} = $ENV{"TASK_WORK"};
908 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
910 # TODO: Share a single JOB_WORK volume across all task
911 # containers on a given worker node, and delete it when the job
912 # ends (and, in case that doesn't work, when the next job
915 # For now, use the same approach as TASK_WORK above.
916 $ENV{"JOB_WORK"} = "/tmp/crunch-job-work";
918 while (my ($env_key, $env_val) = each %ENV)
920 if ($env_key =~ /^(ARVADOS|CRUNCH|JOB|TASK)_/) {
921 $command .= "--env=\Q$env_key=$env_val\E ";
924 $command .= "--env=\QHOME=$ENV{HOME}\E ";
925 $command .= "\Q$docker_hash\E ";
927 if ($Job->{arvados_sdk_version}) {
929 $command .= "perl - \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E";
931 $command .= "/bin/sh -c \'python -c " .
932 '"from pkg_resources import get_distribution as get; print \"Using Arvados SDK version\", get(\"arvados-python-client\").version"' .
933 ">&2 2>/dev/null; " .
934 "mkdir -p \"$ENV{JOB_WORK}\" \"$ENV{TASK_WORK}\" && " .
935 "if which stdbuf >/dev/null ; then " .
936 " exec $stdbuf \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E ;" .
938 " exec \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E ;" .
943 $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -poll=10000 ";
945 $command .= "perl - $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
948 my @execargs = ('bash', '-c', $command);
949 srun (\@srunargs, \@execargs, undef, $build_script);
950 # exec() failed, we assume nothing happened.
951 die "srun() failed on build script\n";
954 if (!defined $childpid)
965 jobstepname => "$job_id.$id.$childpid",
967 croak ("assert failed: \$slot[$childslot]->{'pid'} exists") if exists $slot[$childslot]->{pid};
968 $slot[$childslot]->{pid} = $childpid;
970 Log ($id, "job_task ".$Jobstep->{'arvados_task'}->{'uuid'});
971 Log ($id, "child $childpid started on $childslotname");
972 $Jobstep->{starttime} = time;
973 $Jobstep->{node} = $childnode->{name};
974 $Jobstep->{slotindex} = $childslot;
975 delete $Jobstep->{stderr};
976 delete $Jobstep->{finishtime};
977 delete $Jobstep->{tempfail};
979 $Jobstep->{'arvados_task'}->{started_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{starttime});
980 $Jobstep->{'arvados_task'}->save;
982 splice @jobstep_todo, $todo_ptr, 1;
985 $progress_is_dirty = 1;
989 ($round_num_freeslots > @freeslot && $todo_ptr+1 > $#jobstep_todo))
991 last THISROUND if $main::please_freeze;
992 if ($main::please_info)
994 $main::please_info = 0;
996 create_output_collection();
998 update_progress_stats();
1003 if (!$gotsome || ($latest_refresh + 2 < scalar time))
1005 check_refresh_wanted();
1007 update_progress_stats();
1009 elsif (time - $progress_stats_updated >= 30 || $progress_is_dirty)
1011 update_progress_stats();
1014 select (undef, undef, undef, 0.1);
1016 $working_slot_count = scalar(grep { $_->{node}->{fail_count} == 0 &&
1017 $_->{node}->{hold_count} < 4 } @slot);
1018 if (($thisround_failed_multiple >= 8 && $thisround_succeeded == 0) ||
1019 ($thisround_failed_multiple >= 16 && $thisround_failed_multiple > $thisround_succeeded))
1021 my $message = "Repeated failure rate too high ($thisround_failed_multiple/"
1022 .($thisround_failed+$thisround_succeeded)
1023 .") -- giving up on this round";
1024 Log (undef, $message);
1028 # move slots from freeslot to holdslot (or back to freeslot) if necessary
1029 for (my $i=$#freeslot; $i>=0; $i--) {
1030 if ($slot[$freeslot[$i]]->{node}->{hold_until} > scalar time) {
1031 push @holdslot, (splice @freeslot, $i, 1);
1034 for (my $i=$#holdslot; $i>=0; $i--) {
1035 if ($slot[$holdslot[$i]]->{node}->{hold_until} <= scalar time) {
1036 push @freeslot, (splice @holdslot, $i, 1);
1040 # give up if no nodes are succeeding
1041 if ($working_slot_count < 1) {
1042 Log(undef, "Every node has failed -- giving up");
1049 push @freeslot, splice @holdslot;
1050 map { $slot[$freeslot[$_]]->{node}->{losing_streak} = 0 } (0..$#freeslot);
1053 Log (undef, "wait for last ".(scalar keys %proc)." children to finish");
1056 if ($main::please_continue) {
1057 $main::please_continue = 0;
1060 $main::please_info = 0, freeze(), create_output_collection(), save_meta(1) if $main::please_info;
1062 if (!reapchildren())
1064 check_refresh_wanted();
1066 update_progress_stats();
1067 select (undef, undef, undef, 0.1);
1068 killem (keys %proc) if $main::please_freeze;
1072 update_progress_stats();
1073 freeze_if_want_freeze();
1076 if (!defined $main::success)
1078 if (!@jobstep_todo) {
1080 } elsif ($working_slot_count < 1) {
1081 save_output_collection();
1083 exit(EX_RETRY_UNLOCKED);
1084 } elsif ($thisround_succeeded == 0 &&
1085 ($thisround_failed == 0 || $thisround_failed > 4)) {
1086 my $message = "stop because $thisround_failed tasks failed and none succeeded";
1087 Log (undef, $message);
1092 goto ONELEVEL if !defined $main::success;
1095 release_allocation();
1097 my $collated_output = save_output_collection();
1098 Log (undef, "finish");
1103 if ($collated_output && $main::success) {
1104 $final_state = 'Complete';
1106 $final_state = 'Failed';
1108 $Job->update_attributes('state' => $final_state);
1110 exit (($final_state eq 'Complete') ? 0 : 1);
1114 sub update_progress_stats
1116 $progress_stats_updated = time;
1117 return if !$progress_is_dirty;
1118 my ($todo, $done, $running) = (scalar @jobstep_todo,
1119 scalar @jobstep_done,
1120 scalar keys(%proc));
1121 $Job->{'tasks_summary'} ||= {};
1122 $Job->{'tasks_summary'}->{'todo'} = $todo;
1123 $Job->{'tasks_summary'}->{'done'} = $done;
1124 $Job->{'tasks_summary'}->{'running'} = $running;
1125 $Job->update_attributes('tasks_summary' => $Job->{'tasks_summary'});
1126 Log (undef, "status: $done done, $running running, $todo todo");
1127 $progress_is_dirty = 0;
1134 my $children_reaped = 0;
1135 while ((my $pid = waitpid (-1, WNOHANG)) > 0)
1137 my $childstatus = $?;
1139 my $whatslot = ($slot[$proc{$pid}->{slot}]->{node}->{name}
1141 . $slot[$proc{$pid}->{slot}]->{cpu});
1142 my $jobstepidx = $proc{$pid}->{jobstepidx};
1144 if (!WIFEXITED($childstatus))
1146 # child did not exit (may be temporarily stopped)
1147 Log ($jobstepidx, "child $pid did not actually exit in reapchildren, ignoring for now.");
1152 my $elapsed = time - $proc{$pid}->{time};
1153 my $Jobstep = $jobstep[$jobstepidx];
1155 my $exitvalue = $childstatus >> 8;
1156 my $exitinfo = "exit ".exit_status_s($childstatus);
1157 $Jobstep->{'arvados_task'}->reload;
1158 my $task_success = $Jobstep->{'arvados_task'}->{success};
1160 Log ($jobstepidx, "child $pid on $whatslot $exitinfo success=$task_success");
1162 if (!defined $task_success) {
1163 # task did not indicate one way or the other --> fail
1164 Log($jobstepidx, sprintf(
1165 "ERROR: Task process exited %s, but never updated its task record to indicate success and record its output.",
1166 exit_status_s($childstatus)));
1167 $Jobstep->{'arvados_task'}->{success} = 0;
1168 $Jobstep->{'arvados_task'}->save;
1175 $temporary_fail ||= $Jobstep->{tempfail};
1176 $temporary_fail ||= ($exitvalue == TASK_TEMPFAIL);
1178 ++$thisround_failed;
1179 ++$thisround_failed_multiple if $Jobstep->{'failures'} >= 1;
1181 # Check for signs of a failed or misconfigured node
1182 if (++$slot[$proc{$pid}->{slot}]->{node}->{losing_streak} >=
1183 2+$slot[$proc{$pid}->{slot}]->{node}->{ncpus}) {
1184 # Don't count this against jobstep failure thresholds if this
1185 # node is already suspected faulty and srun exited quickly
1186 if ($slot[$proc{$pid}->{slot}]->{node}->{hold_until} &&
1188 Log ($jobstepidx, "blaming failure on suspect node " .
1189 $slot[$proc{$pid}->{slot}]->{node}->{name});
1190 $temporary_fail ||= 1;
1192 ban_node_by_slot($proc{$pid}->{slot});
1195 Log ($jobstepidx, sprintf('failure (#%d, %s) after %d seconds',
1196 ++$Jobstep->{'failures'},
1197 $temporary_fail ? 'temporary' : 'permanent',
1200 if (!$temporary_fail || $Jobstep->{'failures'} >= 3) {
1201 # Give up on this task, and the whole job
1204 # Put this task back on the todo queue
1205 push @jobstep_todo, $jobstepidx;
1206 $Job->{'tasks_summary'}->{'failed'}++;
1210 ++$thisround_succeeded;
1211 $slot[$proc{$pid}->{slot}]->{node}->{losing_streak} = 0;
1212 $slot[$proc{$pid}->{slot}]->{node}->{hold_until} = 0;
1213 $slot[$proc{$pid}->{slot}]->{node}->{fail_count} = 0;
1214 push @jobstep_done, $jobstepidx;
1215 Log ($jobstepidx, "success in $elapsed seconds");
1217 $Jobstep->{exitcode} = $childstatus;
1218 $Jobstep->{finishtime} = time;
1219 $Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
1220 $Jobstep->{'arvados_task'}->save;
1221 process_stderr_final ($jobstepidx);
1222 Log ($jobstepidx, sprintf("task output (%d bytes): %s",
1223 length($Jobstep->{'arvados_task'}->{output}),
1224 $Jobstep->{'arvados_task'}->{output}));
1226 close $reader{$jobstepidx};
1227 delete $reader{$jobstepidx};
1228 delete $slot[$proc{$pid}->{slot}]->{pid};
1229 push @freeslot, $proc{$pid}->{slot};
1232 if ($task_success) {
1234 my $newtask_list = [];
1235 my $newtask_results;
1237 $newtask_results = api_call(
1240 'created_by_job_task_uuid' => $Jobstep->{'arvados_task'}->{uuid}
1242 'order' => 'qsequence',
1243 'offset' => scalar(@$newtask_list),
1245 push(@$newtask_list, @{$newtask_results->{items}});
1246 } while (@{$newtask_results->{items}});
1247 foreach my $arvados_task (@$newtask_list) {
1249 'level' => $arvados_task->{'sequence'},
1251 'arvados_task' => $arvados_task
1253 push @jobstep, $jobstep;
1254 push @jobstep_todo, $#jobstep;
1257 $progress_is_dirty = 1;
1260 return $children_reaped;
1263 sub check_refresh_wanted
1265 my @stat = stat $ENV{"CRUNCH_REFRESH_TRIGGER"};
1267 $stat[9] > $latest_refresh &&
1268 # ...and we have actually locked the job record...
1269 $job_id eq $Job->{'uuid'}) {
1270 $latest_refresh = scalar time;
1271 my $Job2 = api_call("jobs/get", uuid => $jobspec);
1272 for my $attr ('cancelled_at',
1273 'cancelled_by_user_uuid',
1274 'cancelled_by_client_uuid',
1276 $Job->{$attr} = $Job2->{$attr};
1278 if ($Job->{'state'} ne "Running") {
1279 if ($Job->{'state'} eq "Cancelled") {
1280 Log (undef, "Job cancelled at " . $Job->{'cancelled_at'} . " by user " . $Job->{'cancelled_by_user_uuid'});
1282 Log (undef, "Job state unexpectedly changed to " . $Job->{'state'});
1285 $main::please_freeze = 1;
1292 my $last_squeue_check = $squeue_checked;
1294 # Do not call `squeue` or check the kill list more than once every
1296 return if $last_squeue_check > time - 15;
1297 $squeue_checked = time;
1299 # Look for children from which we haven't received stderr data since
1300 # the last squeue check. If no such children exist, all procs are
1301 # alive and there's no need to even look at squeue.
1303 # As long as the crunchstat poll interval (10s) is shorter than the
1304 # squeue check interval (15s) this should make the squeue check an
1306 my $silent_procs = 0;
1307 for my $js (map {$jobstep[$_->{jobstepidx}]} values %proc)
1309 if (!exists($js->{stderr_at}))
1311 $js->{stderr_at} = 0;
1313 if ($js->{stderr_at} < $last_squeue_check)
1318 return if $silent_procs == 0;
1320 # use killem() on procs whose killtime is reached
1321 while (my ($pid, $procinfo) = each %proc)
1323 my $js = $jobstep[$procinfo->{jobstepidx}];
1324 if (exists $procinfo->{killtime}
1325 && $procinfo->{killtime} <= time
1326 && $js->{stderr_at} < $last_squeue_check)
1329 if ($js->{stderr_at}) {
1330 $sincewhen = " in last " . (time - $js->{stderr_at}) . "s";
1332 Log($procinfo->{jobstepidx}, "killing orphaned srun process $pid (task not in slurm queue, no stderr received$sincewhen)");
1339 # here is an opportunity to check for mysterious problems with local procs
1343 # Get a list of steps still running. Note: squeue(1) says --steps
1344 # selects a format (which we override anyway) and allows us to
1345 # specify which steps we're interested in (which we don't).
1346 # Importantly, it also changes the meaning of %j from "job name" to
1347 # "step name" and (although this isn't mentioned explicitly in the
1348 # docs) switches from "one line per job" mode to "one line per step"
1349 # mode. Without it, we'd just get a list of one job, instead of a
1351 my @squeue = `squeue --jobs=\Q$ENV{SLURM_JOB_ID}\E --steps --format='%j' --noheader`;
1354 Log(undef, "warning: squeue exit status $? ($!)");
1359 # which of my jobsteps are running, according to squeue?
1361 for my $jobstepname (@squeue)
1363 $ok{$jobstepname} = 1;
1366 # Check for child procs >60s old and not mentioned by squeue.
1367 while (my ($pid, $procinfo) = each %proc)
1369 if ($procinfo->{time} < time - 60
1370 && $procinfo->{jobstepname}
1371 && !exists $ok{$procinfo->{jobstepname}}
1372 && !exists $procinfo->{killtime})
1374 # According to slurm, this task has ended (successfully or not)
1375 # -- but our srun child hasn't exited. First we must wait (30
1376 # seconds) in case this is just a race between communication
1377 # channels. Then, if our srun child process still hasn't
1378 # terminated, we'll conclude some slurm communication
1379 # error/delay has caused the task to die without notifying srun,
1380 # and we'll kill srun ourselves.
1381 $procinfo->{killtime} = time + 30;
1382 Log($procinfo->{jobstepidx}, "notice: task is not in slurm queue but srun process $pid has not exited");
1388 sub release_allocation
1392 Log (undef, "release job allocation");
1393 system "scancel $ENV{SLURM_JOB_ID}";
1402 my $sel = IO::Select->new();
1403 foreach my $jobstepidx (keys %reader)
1405 my $fd = $reader{$jobstepidx};
1407 $fd_job{$fd} = $jobstepidx;
1409 if (my $stdout_fd = $jobstep[$jobstepidx]->{stdout_r}) {
1410 $sel->add($stdout_fd);
1411 $fd_job{$stdout_fd} = $jobstepidx;
1414 # select on all reader fds with 0.1s timeout
1415 my @ready_fds = $sel->can_read(0.1);
1416 foreach my $fd (@ready_fds)
1419 if (0 < sysread ($fd, $buf, 65536))
1422 print STDERR $buf if $ENV{CRUNCH_DEBUG};
1424 my $jobstepidx = $fd_job{$fd};
1425 if ($jobstep[$jobstepidx]->{stdout_r} == $fd) {
1426 $jobstep[$jobstepidx]->{stdout_captured} .= $buf;
1430 $jobstep[$jobstepidx]->{stderr_at} = time;
1431 $jobstep[$jobstepidx]->{stderr} .= $buf;
1433 # Consume everything up to the last \n
1434 preprocess_stderr ($jobstepidx);
1436 if (length ($jobstep[$jobstepidx]->{stderr}) > 16384)
1438 # If we get a lot of stderr without a newline, chop off the
1439 # front to avoid letting our buffer grow indefinitely.
1440 substr ($jobstep[$jobstepidx]->{stderr},
1441 0, length($jobstep[$jobstepidx]->{stderr}) - 8192) = "";
1449 # Consume all full lines of stderr for a jobstep. Everything after the
1450 # last newline will remain in $jobstep[$jobstepidx]->{stderr} after
1452 sub preprocess_stderr
1454 my $jobstepidx = shift;
1456 while ($jobstep[$jobstepidx]->{stderr} =~ /^(.*?)\n/) {
1458 substr $jobstep[$jobstepidx]->{stderr}, 0, 1+length($line), "";
1459 Log ($jobstepidx, "stderr $line");
1460 if ($line =~ /srun: error: (SLURM job $ENV{SLURM_JOB_ID} has expired|Unable to confirm allocation for job $ENV{SLURM_JOB_ID})/) {
1462 $main::please_freeze = 1;
1464 elsif (!exists $jobstep[$jobstepidx]->{slotindex}) {
1465 # Skip the following tempfail checks if this srun proc isn't
1466 # attached to a particular worker slot.
1468 elsif ($line =~ /srun: error: (Node failure on|Aborting, .*\bio error\b)/) {
1469 my $job_slot_index = $jobstep[$jobstepidx]->{slotindex};
1470 my $job_slot_index = $jobstep[$jobstepidx]->{slotindex};
1471 $slot[$job_slot_index]->{node}->{fail_count}++;
1472 $jobstep[$jobstepidx]->{tempfail} = 1;
1473 ban_node_by_slot($job_slot_index);
1475 elsif ($line =~ /srun: error: (Unable to create job step|.*: Communication connection failure)/) {
1476 $jobstep[$jobstepidx]->{tempfail} = 1;
1477 ban_node_by_slot($jobstep[$jobstepidx]->{slotindex});
1479 elsif ($line =~ /arvados\.errors\.Keep/) {
1480 $jobstep[$jobstepidx]->{tempfail} = 1;
1486 sub process_stderr_final
1488 my $jobstepidx = shift;
1489 preprocess_stderr ($jobstepidx);
1492 Log ($jobstepidx, "stderr $_");
1493 } split ("\n", $jobstep[$jobstepidx]->{stderr});
1494 $jobstep[$jobstepidx]->{stderr} = '';
1501 if (!open($keep, "-|", "arv-get", "--retries", retry_count(), $hash)) {
1502 Log(undef, "fetch_block run error from arv-get $hash: $!");
1505 my $output_block = "";
1508 my $bytes = sysread($keep, $buf, 1024 * 1024);
1509 if (!defined $bytes) {
1510 Log(undef, "fetch_block read error from arv-get: $!");
1511 $output_block = undef;
1513 } elsif ($bytes == 0) {
1514 # sysread returns 0 at the end of the pipe.
1517 # some bytes were read into buf.
1518 $output_block .= $buf;
1523 Log(undef, "fetch_block arv-get exited " . exit_status_s($?));
1524 $output_block = undef;
1526 return $output_block;
1529 # Create a collection by concatenating the output of all tasks (each
1530 # task's output is either a manifest fragment, a locator for a
1531 # manifest fragment stored in Keep, or nothing at all). Return the
1532 # portable_data_hash of the new collection.
1533 sub create_output_collection
1535 Log (undef, "collate");
1537 my ($child_out, $child_in);
1538 my $pid = open2($child_out, $child_in, 'python', '-c', q{
1541 print (arvados.api("v1").collections().
1542 create(body={"manifest_text": sys.stdin.read()}).
1543 execute(num_retries=int(sys.argv[1]))["portable_data_hash"])
1547 my $manifest_size = 0;
1551 my $output = $_->{'arvados_task'}->{output};
1552 next if (!defined($output));
1554 if ($output =~ /^[0-9a-f]{32}(\+\S+)*$/) {
1555 $next_write = fetch_block($output);
1557 $next_write = $output;
1559 if (defined($next_write)) {
1560 if (!defined(syswrite($child_in, $next_write))) {
1561 # There's been an error writing. Stop the loop.
1562 # We'll log details about the exit code later.
1565 $manifest_size += length($next_write);
1568 my $uuid = $_->{'arvados_task'}->{'uuid'};
1569 Log (undef, "Error retrieving '$output' output by task $task_idx ($uuid)");
1574 Log(undef, "collated output manifest text to send to API server is $manifest_size bytes with access tokens");
1577 my $s = IO::Select->new($child_out);
1578 if ($s->can_read(120)) {
1579 sysread($child_out, $joboutput, 1024 * 1024);
1582 Log(undef, "output collection creation exited " . exit_status_s($?));
1588 Log (undef, "timed out while creating output collection");
1589 foreach my $signal (2, 2, 2, 15, 15, 9) {
1590 kill($signal, $pid);
1591 last if waitpid($pid, WNOHANG) == -1;
1600 # Calls create_output_collection, logs the result, and returns it.
1601 # If that was successful, save that as the output in the job record.
1602 sub save_output_collection {
1603 my $collated_output = create_output_collection();
1605 if (!$collated_output) {
1606 Log(undef, "Failed to write output collection");
1609 Log(undef, "job output $collated_output");
1610 $Job->update_attributes('output' => $collated_output);
1612 return $collated_output;
1619 my $sig = 2; # SIGINT first
1620 if (exists $proc{$_}->{"sent_$sig"} &&
1621 time - $proc{$_}->{"sent_$sig"} > 4)
1623 $sig = 15; # SIGTERM if SIGINT doesn't work
1625 if (exists $proc{$_}->{"sent_$sig"} &&
1626 time - $proc{$_}->{"sent_$sig"} > 4)
1628 $sig = 9; # SIGKILL if SIGTERM doesn't work
1630 if (!exists $proc{$_}->{"sent_$sig"})
1632 Log ($proc{$_}->{jobstepidx}, "sending 2x signal $sig to pid $_");
1634 select (undef, undef, undef, 0.1);
1637 kill $sig, $_; # srun wants two SIGINT to really interrupt
1639 $proc{$_}->{"sent_$sig"} = time;
1640 $proc{$_}->{"killedafter"} = time - $proc{$_}->{"time"};
1650 vec($bits,fileno($_),1) = 1;
1656 # Send log output to Keep via arv-put.
1658 # $log_pipe_in and $log_pipe_out are the input and output filehandles to the arv-put pipe.
1659 # $log_pipe_out_buf is a string containing all output read from arv-put so far.
1660 # $log_pipe_out_select is an IO::Select object around $log_pipe_out.
1661 # $log_pipe_pid is the pid of the arv-put subprocess.
1663 # The only functions that should access these variables directly are:
1665 # log_writer_start($logfilename)
1666 # Starts an arv-put pipe, reading data on stdin and writing it to
1667 # a $logfilename file in an output collection.
1669 # log_writer_read_output([$timeout])
1670 # Read output from $log_pipe_out and append it to $log_pipe_out_buf.
1671 # Passes $timeout to the select() call, with a default of 0.01.
1672 # Returns the result of the last read() call on $log_pipe_out, or
1673 # -1 if read() wasn't called because select() timed out.
1674 # Only other log_writer_* functions should need to call this.
1676 # log_writer_send($txt)
1677 # Writes $txt to the output log collection.
1679 # log_writer_finish()
1680 # Closes the arv-put pipe and returns the output that it produces.
1682 # log_writer_is_active()
1683 # Returns a true value if there is currently a live arv-put
1684 # process, false otherwise.
1686 my ($log_pipe_in, $log_pipe_out, $log_pipe_out_buf, $log_pipe_out_select,
1689 sub log_writer_start($)
1691 my $logfilename = shift;
1692 $log_pipe_pid = open2($log_pipe_out, $log_pipe_in,
1696 '--filename', $logfilename,
1698 $log_pipe_out_buf = "";
1699 $log_pipe_out_select = IO::Select->new($log_pipe_out);
1702 sub log_writer_read_output {
1703 my $timeout = shift || 0.01;
1705 while ($read && $log_pipe_out_select->can_read($timeout)) {
1706 $read = read($log_pipe_out, $log_pipe_out_buf, 65536,
1707 length($log_pipe_out_buf));
1709 if (!defined($read)) {
1710 Log(undef, "error reading log manifest from arv-put: $!");
1715 sub log_writer_send($)
1718 print $log_pipe_in $txt;
1719 log_writer_read_output();
1722 sub log_writer_finish()
1724 return unless $log_pipe_pid;
1726 close($log_pipe_in);
1728 my $logger_failed = 0;
1729 my $read_result = log_writer_read_output(120);
1730 if ($read_result == -1) {
1731 $logger_failed = -1;
1732 Log (undef, "timed out reading from 'arv-put'");
1733 } elsif ($read_result != 0) {
1734 $logger_failed = -2;
1735 Log(undef, "failed to read arv-put log manifest to EOF");
1738 waitpid($log_pipe_pid, 0);
1740 $logger_failed ||= $?;
1741 Log(undef, "log_writer_finish: arv-put exited " . exit_status_s($?))
1744 close($log_pipe_out);
1745 my $arv_put_output = $logger_failed ? undef : $log_pipe_out_buf;
1746 $log_pipe_pid = $log_pipe_in = $log_pipe_out = $log_pipe_out_buf =
1747 $log_pipe_out_select = undef;
1749 return $arv_put_output;
1752 sub log_writer_is_active() {
1753 return $log_pipe_pid;
1756 sub Log # ($jobstepidx, $logmessage)
1758 my ($jobstepidx, $logmessage) = @_;
1759 if ($logmessage =~ /\n/) {
1760 for my $line (split (/\n/, $_[1])) {
1761 Log ($jobstepidx, $line);
1765 my $fh = select STDERR; $|=1; select $fh;
1767 if (defined($jobstepidx) && exists($jobstep[$jobstepidx]->{arvados_task})) {
1768 $task_qseq = $jobstepidx;
1770 my $message = sprintf ("%s %d %s %s", $job_id, $$, $task_qseq, $logmessage);
1771 $message =~ s{([^ -\176])}{"\\" . sprintf ("%03o", ord($1))}ge;
1774 if (log_writer_is_active() || -t STDERR) {
1775 my @gmtime = gmtime;
1776 $datetime = sprintf ("%04d-%02d-%02d_%02d:%02d:%02d",
1777 $gmtime[5]+1900, $gmtime[4]+1, @gmtime[3,2,1,0]);
1779 print STDERR ((-t STDERR) ? ($datetime." ".$message) : $message);
1781 if (log_writer_is_active()) {
1782 log_writer_send($datetime . " " . $message);
1789 my ($package, $file, $line) = caller;
1790 my $message = "@_ at $file line $line\n";
1791 Log (undef, $message);
1792 freeze() if @jobstep_todo;
1793 create_output_collection() if @jobstep_todo;
1803 if ($Job->{'state'} eq 'Cancelled') {
1804 $Job->update_attributes('finished_at' => scalar gmtime);
1806 $Job->update_attributes('state' => 'Failed');
1813 my $justcheckpoint = shift; # false if this will be the last meta saved
1814 return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm
1815 return unless log_writer_is_active();
1816 my $log_manifest = log_writer_finish();
1817 return unless defined($log_manifest);
1820 my $prev_log_coll = api_call("collections/get", uuid => $Job->{log});
1821 $log_manifest = $prev_log_coll->{manifest_text} . $log_manifest;
1824 my $log_coll = api_call(
1825 "collections/create", ensure_unique_name => 1, collection => {
1826 manifest_text => $log_manifest,
1827 owner_uuid => $Job->{owner_uuid},
1828 name => sprintf("Log from %s job %s", $Job->{script}, $Job->{uuid}),
1830 Log(undef, "log collection is " . $log_coll->{portable_data_hash});
1831 $Job->update_attributes('log' => $log_coll->{portable_data_hash});
1835 sub freeze_if_want_freeze
1837 if ($main::please_freeze)
1839 release_allocation();
1842 # kill some srun procs before freeze+stop
1843 map { $proc{$_} = {} } @_;
1846 killem (keys %proc);
1847 select (undef, undef, undef, 0.1);
1849 while (($died = waitpid (-1, WNOHANG)) > 0)
1851 delete $proc{$died};
1856 create_output_collection();
1866 Log (undef, "Freeze not implemented");
1873 croak ("Thaw not implemented");
1889 $s =~ s{\\(.)}{$1 eq "n" ? "\n" : $1}ge;
1896 my $srunargs = shift;
1897 my $execargs = shift;
1898 my $opts = shift || {};
1901 my $label = exists $opts->{label} ? $opts->{label} : "@$execargs";
1902 Log (undef, "$label: start");
1904 my ($stderr_r, $stderr_w);
1905 pipe $stderr_r, $stderr_w or croak("pipe() failed: $!");
1907 my ($stdout_r, $stdout_w);
1908 pipe $stdout_r, $stdout_w or croak("pipe() failed: $!");
1910 my $srunpid = fork();
1915 fcntl($stderr_w, F_SETFL, 0) or croak($!); # no close-on-exec
1916 fcntl($stdout_w, F_SETFL, 0) or croak($!);
1917 open(STDERR, ">&", $stderr_w);
1918 open(STDOUT, ">&", $stdout_w);
1919 srun ($srunargs, $execargs, $opts, $stdin);
1925 set_nonblocking($stderr_r);
1926 set_nonblocking($stdout_r);
1928 # Add entries to @jobstep and %proc so check_squeue() and
1929 # freeze_if_want_freeze() can treat it like a job task process.
1933 stderr_captured => '',
1934 stdout_r => $stdout_r,
1935 stdout_captured => '',
1937 my $jobstepidx = $#jobstep;
1939 jobstepidx => $jobstepidx,
1941 $reader{$jobstepidx} = $stderr_r;
1943 while ($srunpid != waitpid ($srunpid, WNOHANG)) {
1944 my $busy = readfrompipes();
1945 if (!$busy || ($latest_refresh + 2 < scalar time)) {
1946 check_refresh_wanted();
1950 select(undef, undef, undef, 0.1);
1952 killem(keys %proc) if $main::please_freeze;
1956 1 while readfrompipes();
1957 process_stderr_final ($jobstepidx);
1959 Log (undef, "$label: exit ".exit_status_s($exited));
1963 delete $proc{$srunpid};
1964 delete $reader{$jobstepidx};
1966 my $j = pop @jobstep;
1967 return ($exited, $j->{stdout_captured}, $j->{stderr_captured});
1973 my $srunargs = shift;
1974 my $execargs = shift;
1975 my $opts = shift || {};
1977 my $args = $have_slurm ? [@$srunargs, @$execargs] : $execargs;
1979 $Data::Dumper::Terse = 1;
1980 $Data::Dumper::Indent = 0;
1981 my $show_cmd = Dumper($args);
1982 $show_cmd =~ s/(TOKEN\\*=)[^\s\']+/${1}[...]/g;
1983 $show_cmd =~ s/\n/ /g;
1984 if ($opts->{fork}) {
1985 Log(undef, "starting: $show_cmd");
1987 # This is a child process: parent is in charge of reading our
1988 # stderr and copying it to Log() if needed.
1989 warn "starting: $show_cmd\n";
1992 if (defined $stdin) {
1993 my $child = open STDIN, "-|";
1994 defined $child or die "no fork: $!";
1996 print $stdin or die $!;
1997 close STDOUT or die $!;
2002 return system (@$args) if $opts->{fork};
2005 warn "ENV size is ".length(join(" ",%ENV));
2006 die "exec failed: $!: @$args";
2010 sub ban_node_by_slot {
2011 # Don't start any new jobsteps on this node for 60 seconds
2013 $slot[$slotid]->{node}->{hold_until} = 60 + scalar time;
2014 $slot[$slotid]->{node}->{hold_count}++;
2015 Log (undef, "backing off node " . $slot[$slotid]->{node}->{name} . " for 60 seconds");
2020 my ($lockfile, $error_message) = @_;
2021 open L, ">", $lockfile or croak("$lockfile: $!");
2022 if (!flock L, LOCK_EX|LOCK_NB) {
2023 croak("Can't lock $lockfile: $error_message\n");
2027 sub find_docker_image {
2028 # Given a Keep locator, check to see if it contains a Docker image.
2029 # If so, return its stream name and Docker hash.
2030 # If not, return undef for both values.
2031 my $locator = shift;
2032 my ($streamname, $filename);
2033 my $image = api_call("collections/get", uuid => $locator);
2035 foreach my $line (split(/\n/, $image->{manifest_text})) {
2036 my @tokens = split(/\s+/, $line);
2038 $streamname = shift(@tokens);
2039 foreach my $filedata (grep(/^\d+:\d+:/, @tokens)) {
2040 if (defined($filename)) {
2041 return (undef, undef); # More than one file in the Collection.
2043 $filename = (split(/:/, $filedata, 3))[2];
2048 if (defined($filename) and ($filename =~ /^([0-9A-Fa-f]{64})\.tar$/)) {
2049 return ($streamname, $1);
2051 return (undef, undef);
2056 # Calculate the number of times an operation should be retried,
2057 # assuming exponential backoff, and that we're willing to retry as
2058 # long as tasks have been running. Enforce a minimum of 3 retries.
2059 my ($starttime, $endtime, $timediff, $retries);
2061 $starttime = $jobstep[0]->{starttime};
2062 $endtime = $jobstep[-1]->{finishtime};
2064 if (!defined($starttime)) {
2066 } elsif (!defined($endtime)) {
2067 $timediff = time - $starttime;
2069 $timediff = ($endtime - $starttime) - (time - $endtime);
2071 if ($timediff > 0) {
2072 $retries = int(log($timediff) / log(2));
2074 $retries = 1; # Use the minimum.
2076 return ($retries > 3) ? $retries : 3;
2080 # Pass in two function references.
2081 # This method will be called with the remaining arguments.
2082 # If it dies, retry it with exponential backoff until it succeeds,
2083 # or until the current retry_count is exhausted. After each failure
2084 # that can be retried, the second function will be called with
2085 # the current try count (0-based), next try time, and error message.
2086 my $operation = shift;
2087 my $retry_callback = shift;
2088 my $retries = retry_count();
2089 foreach my $try_count (0..$retries) {
2090 my $next_try = time + (2 ** $try_count);
2091 my $result = eval { $operation->(@_); };
2094 } elsif ($try_count < $retries) {
2095 $retry_callback->($try_count, $next_try, $@);
2096 my $sleep_time = $next_try - time;
2097 sleep($sleep_time) if ($sleep_time > 0);
2100 # Ensure the error message ends in a newline, so Perl doesn't add
2101 # retry_op's line number to it.
2107 # Pass in a /-separated API method name, and arguments for it.
2108 # This function will call that method, retrying as needed until
2109 # the current retry_count is exhausted, with a log on the first failure.
2110 my $method_name = shift;
2111 my $log_api_retry = sub {
2112 my ($try_count, $next_try_at, $errmsg) = @_;
2113 $errmsg =~ s/\s*\bat \Q$0\E line \d+\.?\s*//;
2114 $errmsg =~ s/\s/ /g;
2115 $errmsg =~ s/\s+$//;
2117 if ($next_try_at < time) {
2118 $retry_msg = "Retrying.";
2120 my $next_try_fmt = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($next_try_at);
2121 $retry_msg = "Retrying at $next_try_fmt.";
2123 Log(undef, "API method $method_name failed: $errmsg. $retry_msg");
2126 foreach my $key (split(/\//, $method_name)) {
2127 $method = $method->{$key};
2129 return retry_op(sub { $method->execute(@_); }, $log_api_retry, @_);
2133 # Given a $?, return a human-readable exit code string like "0" or
2134 # "1" or "0 with signal 1" or "1 with signal 11".
2135 my $exitcode = shift;
2136 my $s = $exitcode >> 8;
2137 if ($exitcode & 0x7f) {
2138 $s .= " with signal " . ($exitcode & 0x7f);
2140 if ($exitcode & 0x80) {
2141 $s .= " with core dump";
2146 sub handle_readall {
2147 # Pass in a glob reference to a file handle.
2148 # Read all its contents and return them as a string.
2149 my $fh_glob_ref = shift;
2151 return <$fh_glob_ref>;
2154 sub tar_filename_n {
2156 return sprintf("%s/git.%s.%d.tar", $ENV{CRUNCH_TMP}, $job_id, $n);
2159 sub add_git_archive {
2160 # Pass in a git archive command as a string or list, a la system().
2161 # This method will save its output to be included in the archive sent to the
2165 if (!open(GIT_ARCHIVE, ">", tar_filename_n($git_tar_count))) {
2166 croak("Failed to save git archive: $!");
2168 my $git_pid = open2(">&GIT_ARCHIVE", $git_input, @_);
2170 waitpid($git_pid, 0);
2173 croak("Failed to save git archive: git exited " . exit_status_s($?));
2177 sub combined_git_archive {
2178 # Combine all saved tar archives into a single archive, then return its
2179 # contents in a string. Return undef if no archives have been saved.
2180 if ($git_tar_count < 1) {
2183 my $base_tar_name = tar_filename_n(1);
2184 foreach my $tar_to_append (map { tar_filename_n($_); } (2..$git_tar_count)) {
2185 my $tar_exit = system("tar", "-Af", $base_tar_name, $tar_to_append);
2186 if ($tar_exit != 0) {
2187 croak("Error preparing build archive: tar -A exited " .
2188 exit_status_s($tar_exit));
2191 if (!open(GIT_TAR, "<", $base_tar_name)) {
2192 croak("Could not open build archive: $!");
2194 my $tar_contents = handle_readall(\*GIT_TAR);
2196 return $tar_contents;
2199 sub set_nonblocking {
2201 my $flags = fcntl ($fh, F_GETFL, 0) or croak ($!);
2202 fcntl ($fh, F_SETFL, $flags | O_NONBLOCK) or croak ($!);
2208 # This is crunch-job's internal dispatch script. crunch-job running on the API
2209 # server invokes this script on individual compute nodes, or localhost if we're
2210 # running a job locally. It gets called in two modes:
2212 # * No arguments: Installation mode. Read a tar archive from the DATA
2213 # file handle; it includes the Crunch script's source code, and
2214 # maybe SDKs as well. Those should be installed in the proper
2215 # locations. This runs outside of any Docker container, so don't try to
2216 # introspect Crunch's runtime environment.
2218 # * With arguments: Crunch script run mode. This script should set up the
2219 # environment, then run the command specified in the arguments. This runs
2220 # inside any Docker container.
2223 use File::Path qw( make_path remove_tree );
2224 use POSIX qw(getcwd);
2226 use constant TASK_TEMPFAIL => 111;
2228 # Map SDK subdirectories to the path environments they belong to.
2229 my %SDK_ENVVARS = ("perl/lib" => "PERLLIB", "ruby/lib" => "RUBYLIB");
2231 my $destdir = $ENV{"CRUNCH_SRC"};
2232 my $archive_hash = $ENV{"CRUNCH_GIT_ARCHIVE_HASH"};
2233 my $repo = $ENV{"CRUNCH_SRC_URL"};
2234 my $install_dir = $ENV{"CRUNCH_INSTALL"} || (getcwd() . "/opt");
2235 my $job_work = $ENV{"JOB_WORK"};
2236 my $task_work = $ENV{"TASK_WORK"};
2238 open(STDOUT_ORIG, ">&", STDOUT);
2239 open(STDERR_ORIG, ">&", STDERR);
2241 for my $dir ($destdir, $job_work, $task_work) {
2244 -e $dir or die "Failed to create temporary directory ($dir): $!";
2249 remove_tree($task_work, {keep_root => 1});
2252 ### Crunch script run mode
2254 # We want to do routine logging during task 0 only. This gives the user
2255 # the information they need, but avoids repeating the information for every
2258 if ($ENV{TASK_SEQUENCE} eq "0") {
2261 printf STDERR_ORIG "[Crunch] $msg\n", @_;
2267 my $python_src = "$install_dir/python";
2268 my $venv_dir = "$job_work/.arvados.venv";
2269 my $venv_built = -e "$venv_dir/bin/activate";
2270 if ((!$venv_built) and (-d $python_src) and can_run("virtualenv")) {
2271 shell_or_die(undef, "virtualenv", "--quiet", "--system-site-packages",
2272 "--python=python2.7", $venv_dir);
2273 shell_or_die(TASK_TEMPFAIL, "$venv_dir/bin/pip", "--quiet", "install", "-I", $python_src);
2275 $Log->("Built Python SDK virtualenv");
2278 my @pysdk_version_cmd = ("python", "-c",
2279 "from pkg_resources import get_distribution as get; print get('arvados-python-client').version");
2281 $Log->("Running in Python SDK virtualenv");
2282 @pysdk_version_cmd = ();
2283 my $orig_argv = join(" ", map { quotemeta($_); } @ARGV);
2284 @ARGV = ("/bin/sh", "-ec",
2285 ". \Q$venv_dir/bin/activate\E; exec $orig_argv");
2286 } elsif (-d $python_src) {
2287 $Log->("Warning: virtualenv not found inside Docker container default " .
2288 "\$PATH. Can't install Python SDK.");
2291 if (@pysdk_version_cmd) {
2292 open(my $pysdk_version_pipe, "-|", @pysdk_version_cmd);
2293 my $pysdk_version = <$pysdk_version_pipe>;
2294 close($pysdk_version_pipe);
2296 chomp($pysdk_version);
2297 $Log->("Using Arvados SDK version $pysdk_version");
2299 # A lot could've gone wrong here, but pretty much all of it means that
2300 # Python won't be able to load the Arvados SDK.
2301 $Log->("Warning: Arvados SDK not found");
2305 while (my ($sdk_dir, $sdk_envkey) = each(%SDK_ENVVARS)) {
2306 my $sdk_path = "$install_dir/$sdk_dir";
2308 if ($ENV{$sdk_envkey}) {
2309 $ENV{$sdk_envkey} = "$sdk_path:" . $ENV{$sdk_envkey};
2311 $ENV{$sdk_envkey} = $sdk_path;
2313 $Log->("Arvados SDK added to %s", $sdk_envkey);
2318 die "Cannot exec `@ARGV`: $!";
2321 ### Installation mode
2322 open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
2324 if (readlink ("$destdir.archive_hash") eq $archive_hash && -d $destdir) {
2325 # This exact git archive (source + arvados sdk) is already installed
2326 # here, so there's no need to reinstall it.
2328 # We must consume our DATA section, though: otherwise the process
2329 # feeding it to us will get SIGPIPE.
2331 while (read(DATA, $buf, 65536)) { }
2336 unlink "$destdir.archive_hash";
2340 # Ignore SIGPIPE: we check retval of close() instead. See perlipc(1).
2341 local $SIG{PIPE} = "IGNORE";
2342 warn "Extracting archive: $archive_hash\n";
2343 # --ignore-zeros is necessary sometimes: depending on how much NUL
2344 # padding tar -A put on our combined archive (which in turn depends
2345 # on the length of the component archives) tar without
2346 # --ignore-zeros will exit before consuming stdin and cause close()
2347 # to fail on the resulting SIGPIPE.
2348 if (!open(TARX, "|-", "tar", "--ignore-zeros", "-xC", $destdir)) {
2349 die "Error launching 'tar -xC $destdir': $!";
2351 # If we send too much data to tar in one write (> 4-5 MiB), it stops, and we
2352 # get SIGPIPE. We must feed it data incrementally.
2354 while (read(DATA, $tar_input, 65536)) {
2355 print TARX $tar_input;
2358 die "'tar -xC $destdir' exited $?: $!";
2364 my $sdk_root = "$destdir/.arvados.sdk/sdk";
2366 foreach my $sdk_lang (("python",
2367 map { (split /\//, $_, 2)[0]; } keys(%SDK_ENVVARS))) {
2368 if (-d "$sdk_root/$sdk_lang") {
2369 if (!rename("$sdk_root/$sdk_lang", "$install_dir/$sdk_lang")) {
2370 die "Failed to install $sdk_lang SDK: $!";
2376 my $python_dir = "$install_dir/python";
2377 if ((-d $python_dir) and can_run("python2.7")) {
2378 open(my $egg_info_pipe, "-|",
2379 "python2.7 \Q$python_dir/setup.py\E egg_info 2>&1 >/dev/null");
2380 my @egg_info_errors = <$egg_info_pipe>;
2381 close($egg_info_pipe);
2384 if (@egg_info_errors and (($egg_info_errors[-1] =~ /\bgit\b/) or ($egg_info_errors[-1] =~ /\[Errno 2\]/))) {
2385 # egg_info apparently failed because it couldn't ask git for a build tag.
2386 # Specify no build tag.
2387 open(my $pysdk_cfg, ">>", "$python_dir/setup.cfg");
2388 print $pysdk_cfg "\n[egg_info]\ntag_build =\n";
2391 my $egg_info_exit = $? >> 8;
2392 foreach my $errline (@egg_info_errors) {
2395 warn "python setup.py egg_info failed: exit $egg_info_exit";
2396 exit ($egg_info_exit || 1);
2401 # Hide messages from the install script (unless it fails: shell_or_die
2402 # will show $destdir.log in that case).
2403 open(STDOUT, ">>", "$destdir.log");
2404 open(STDERR, ">&", STDOUT);
2406 if (-e "$destdir/crunch_scripts/install") {
2407 shell_or_die (undef, "$destdir/crunch_scripts/install", $install_dir);
2408 } elsif (!-e "./install.sh" && -e "./tests/autotests.sh") {
2410 shell_or_die (undef, "./tests/autotests.sh", $install_dir);
2411 } elsif (-e "./install.sh") {
2412 shell_or_die (undef, "./install.sh", $install_dir);
2415 if ($archive_hash) {
2416 unlink "$destdir.archive_hash.new";
2417 symlink ($archive_hash, "$destdir.archive_hash.new") or die "$destdir.archive_hash.new: $!";
2418 rename ("$destdir.archive_hash.new", "$destdir.archive_hash") or die "$destdir.archive_hash: $!";
2424 my $command_name = shift;
2425 open(my $which, "-|", "which", $command_name);
2426 while (<$which>) { }
2433 my $exitcode = shift;
2435 if ($ENV{"DEBUG"}) {
2436 print STDERR "@_\n";
2438 if (system (@_) != 0) {
2441 my $exitstatus = sprintf("exit %d signal %d", $code >> 8, $code & 0x7f);
2442 open STDERR, ">&STDERR_ORIG";
2443 system ("cat $destdir.log >&2");
2444 warn "@_ failed ($err): $exitstatus";
2445 if (defined($exitcode)) {
2449 exit (($code >> 8) || 1);