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 } else if (-f "/etc/pki/tls/certs/ca-bundle.crt") {
117 $ENV{"HOST_CERTS"} = "/etc/pki/tls/certs/ca-bundle.crt";
120 # Create the tmp directory if it does not exist
121 if ( ! -d $ENV{"CRUNCH_TMP"} ) {
122 make_path $ENV{"CRUNCH_TMP"} or die "Failed to create temporary working directory: " . $ENV{"CRUNCH_TMP"};
125 $ENV{"JOB_WORK"} = $ENV{"CRUNCH_TMP"} . "/work";
126 $ENV{"CRUNCH_INSTALL"} = "$ENV{CRUNCH_TMP}/opt";
127 $ENV{"CRUNCH_WORK"} = $ENV{"JOB_WORK"}; # deprecated
128 mkdir ($ENV{"JOB_WORK"});
137 my $cgroup_root = "/sys/fs/cgroup";
138 my $docker_bin = "docker.io";
139 my $docker_run_args = "";
140 GetOptions('force-unlock' => \$force_unlock,
141 'git-dir=s' => \$git_dir,
142 'job=s' => \$jobspec,
143 'job-api-token=s' => \$job_api_token,
144 'no-clear-tmp' => \$no_clear_tmp,
145 'resume-stash=s' => \$resume_stash,
146 'cgroup-root=s' => \$cgroup_root,
147 'docker-bin=s' => \$docker_bin,
148 'docker-run-args=s' => \$docker_run_args,
151 if (defined $job_api_token) {
152 $ENV{ARVADOS_API_TOKEN} = $job_api_token;
155 my $have_slurm = exists $ENV{SLURM_JOB_ID} && exists $ENV{SLURM_NODELIST};
160 $main::ENV{CRUNCH_DEBUG} = 1;
164 $main::ENV{CRUNCH_DEBUG} = 0;
167 my $arv = Arvados->new('apiVersion' => 'v1');
176 if ($jobspec =~ /^[-a-z\d]+$/)
178 # $jobspec is an Arvados UUID, not a JSON job specification
179 $Job = api_call("jobs/get", uuid => $jobspec);
184 $local_job = JSON::decode_json($jobspec);
188 # Make sure our workers (our slurm nodes, localhost, or whatever) are
189 # at least able to run basic commands: they aren't down or severely
192 if (($Job || $local_job)->{docker_image_locator}) {
193 $cmd = [$docker_bin, 'ps', '-q'];
195 Log(undef, "Sanity check is `@$cmd`");
196 my ($exited, $stdout, $stderr) = srun_sync(
197 ["srun", "--nodes=\Q$ENV{SLURM_NNODES}\E", "--ntasks-per-node=1"],
199 {label => "sanity check"});
201 Log(undef, "Sanity check failed: ".exit_status_s($exited));
204 Log(undef, "Sanity check OK");
207 my $User = api_call("users/current");
210 if (!$force_unlock) {
211 # Claim this job, and make sure nobody else does
212 eval { api_call("jobs/lock", uuid => $Job->{uuid}); };
214 Log(undef, "Error while locking job, exiting ".EX_TEMPFAIL);
223 map { croak ("No $_ specified") unless $local_job->{$_} }
224 qw(script script_version script_parameters);
227 $local_job->{'is_locked_by_uuid'} = $User->{'uuid'};
228 $local_job->{'started_at'} = gmtime;
229 $local_job->{'state'} = 'Running';
231 $Job = api_call("jobs/create", job => $local_job);
233 $job_id = $Job->{'uuid'};
235 my $keep_logfile = $job_id . '.log.txt';
236 log_writer_start($keep_logfile);
238 $Job->{'runtime_constraints'} ||= {};
239 $Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0;
240 my $max_ncpus = $Job->{'runtime_constraints'}->{'max_tasks_per_node'};
242 my $gem_versions = `gem list --quiet arvados-cli 2>/dev/null`;
244 $gem_versions =~ s/^arvados-cli \(/ with arvados-cli Gem version(s) /;
245 chomp($gem_versions);
246 chop($gem_versions); # Closing parentheses
251 "running from " . ((-e $0) ? realpath($0) : "stdin") . $gem_versions);
253 Log (undef, "check slurm allocation");
256 # Should use $ENV{SLURM_TASKS_PER_NODE} instead of sinfo? (eg. "4(x3),2,4(x2)")
260 my $localcpus = 0 + `grep -cw ^processor /proc/cpuinfo` || 1;
261 push @sinfo, "$localcpus localhost";
263 if (exists $ENV{SLURM_NODELIST})
265 push @sinfo, `sinfo -h --format='%c %N' --nodes=\Q$ENV{SLURM_NODELIST}\E`;
269 my ($ncpus, $slurm_nodelist) = split;
270 $ncpus = $max_ncpus if $max_ncpus && $ncpus > $max_ncpus;
273 while ($slurm_nodelist =~ s/^([^\[,]+?(\[.*?\])?)(,|$)//)
276 if ($nodelist =~ /\[((\d+)(-(\d+))?(,(\d+)(-(\d+))?)*)\]/)
279 foreach (split (",", $ranges))
292 push @nodelist, map {
294 $n =~ s/\[[-,\d]+\]/$_/;
301 push @nodelist, $nodelist;
304 foreach my $nodename (@nodelist)
306 Log (undef, "node $nodename - $ncpus slots");
307 my $node = { name => $nodename,
309 # The number of consecutive times a task has been dispatched
310 # to this node and failed.
312 # The number of consecutive times that SLURM has reported
313 # a node failure since the last successful task.
315 # Don't dispatch work to this node until this time
316 # (in seconds since the epoch) has passed.
318 foreach my $cpu (1..$ncpus)
320 push @slot, { node => $node,
324 push @node, @nodelist;
329 # Ensure that we get one jobstep running on each allocated node before
330 # we start overloading nodes with concurrent steps
332 @slot = sort { $a->{cpu} <=> $b->{cpu} } @slot;
335 $Job->update_attributes(
336 'tasks_summary' => { 'failed' => 0,
341 Log (undef, "start");
342 $SIG{'INT'} = sub { $main::please_freeze = 1; };
343 $SIG{'QUIT'} = sub { $main::please_freeze = 1; };
344 $SIG{'TERM'} = \&croak;
345 $SIG{'TSTP'} = sub { $main::please_freeze = 1; };
346 $SIG{'ALRM'} = sub { $main::please_info = 1; };
347 $SIG{'CONT'} = sub { $main::please_continue = 1; };
348 $SIG{'HUP'} = sub { $main::please_refresh = 1; };
350 $main::please_freeze = 0;
351 $main::please_info = 0;
352 $main::please_continue = 0;
353 $main::please_refresh = 0;
354 my $jobsteps_must_output_keys = 0; # becomes 1 when any task outputs a key
356 grep { $ENV{$1} = $2 if /^(NOCACHE.*?)=(.*)/ } split ("\n", $$Job{knobs});
357 $ENV{"CRUNCH_JOB_UUID"} = $job_id;
358 $ENV{"JOB_UUID"} = $job_id;
361 my @jobstep_todo = ();
362 my @jobstep_done = ();
363 my @jobstep_tomerge = ();
364 my $jobstep_tomerge_level = 0;
365 my $squeue_checked = 0;
366 my $latest_refresh = scalar time;
370 if (defined $Job->{thawedfromkey})
372 thaw ($Job->{thawedfromkey});
376 my $first_task = api_call("job_tasks/create", job_task => {
377 'job_uuid' => $Job->{'uuid'},
382 push @jobstep, { 'level' => 0,
384 'arvados_task' => $first_task,
386 push @jobstep_todo, 0;
392 must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
395 my $build_script = handle_readall(\*DATA);
396 my $nodelist = join(",", @node);
397 my $git_tar_count = 0;
399 if (!defined $no_clear_tmp) {
400 # Find FUSE mounts under $CRUNCH_TMP and unmount them. Then clean
401 # up work directories crunch_tmp/work, crunch_tmp/opt,
404 # TODO: When #5036 is done and widely deployed, we can limit mount's
405 # -t option to simply fuse.keep.
406 my ($exited, $stdout, $stderr) = srun_sync(
407 ["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
408 ['bash', '-ec', '-o', 'pipefail', 'mount -t fuse,fuse.keep | awk "(index(\$3, \"$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* $CRUNCH_TMP/*.cid'],
409 {label => "clean work dirs"});
411 exit(EX_RETRY_UNLOCKED);
415 # If this job requires a Docker image, install that.
416 my ($docker_locator, $docker_stream, $docker_hash, $docker_limitmem, $dockeruserarg);
417 if ($docker_locator = $Job->{docker_image_locator}) {
418 Log (undef, "Install docker image $docker_locator");
419 ($docker_stream, $docker_hash) = find_docker_image($docker_locator);
422 croak("No Docker image hash found from locator $docker_locator");
424 Log (undef, "docker image hash is $docker_hash");
425 $docker_stream =~ s/^\.//;
426 my $docker_install_script = qq{
427 if ! $docker_bin images -q --no-trunc --all | grep -qxF \Q$docker_hash\E; then
428 arv-get \Q$docker_locator$docker_stream/$docker_hash.tar\E | $docker_bin load
432 my ($exited, $stdout, $stderr) = srun_sync(
433 ["srun", "--nodelist=" . join(',', @node)],
434 ["/bin/bash", "-o", "pipefail", "-ec", $docker_install_script],
435 {label => "load docker image"});
438 exit(EX_RETRY_UNLOCKED);
441 # Determine whether this version of Docker supports memory+swap limits.
442 ($exited, $stdout, $stderr) = srun_sync(
443 ["srun", "--nodes=1"],
444 [$docker_bin, 'run', '--help'],
445 {label => "check --memory-swap feature"});
446 $docker_limitmem = ($stdout =~ /--memory-swap/);
448 # Find a non-root Docker user to use.
449 # Tries the default user for the container, then 'crunch', then 'nobody',
450 # testing for whether the actual user id is non-zero. This defends against
451 # mistakes but not malice, but we intend to harden the security in the future
452 # so we don't want anyone getting used to their jobs running as root in their
454 my @tryusers = ("", "crunch", "nobody");
455 foreach my $try_user (@tryusers) {
458 if ($try_user eq "") {
459 $label = "check whether default user is UID 0";
462 $label = "check whether user '$try_user' is UID 0";
463 $try_user_arg = "--user=$try_user";
465 my ($exited, $stdout, $stderr) = srun_sync(
466 ["srun", "--nodes=1"],
468 "$docker_bin run $docker_run_args $try_user_arg $docker_hash id --user"],
471 if ($exited == 0 && $stdout =~ /^\d+$/ && $stdout > 0) {
472 $dockeruserarg = $try_user_arg;
473 if ($try_user eq "") {
474 Log(undef, "Container will run with default user");
476 Log(undef, "Container will run with $dockeruserarg");
482 if (!defined $dockeruserarg) {
483 croak("Could not find a user in container that is not UID 0 (tried default user, @tryusers) or there was a problem running 'id' in the container.");
486 if ($Job->{arvados_sdk_version}) {
487 # The job also specifies an Arvados SDK version. Add the SDKs to the
488 # tar file for the build script to install.
489 Log(undef, sprintf("Packing Arvados SDK version %s for installation",
490 $Job->{arvados_sdk_version}));
491 add_git_archive("git", "--git-dir=$git_dir", "archive",
492 "--prefix=.arvados.sdk/",
493 $Job->{arvados_sdk_version}, "sdk");
497 if (!defined $git_dir && $Job->{'script_version'} =~ m{^/}) {
498 # If script_version looks like an absolute path, *and* the --git-dir
499 # argument was not given -- which implies we were not invoked by
500 # crunch-dispatch -- we will use the given path as a working
501 # directory instead of resolving script_version to a git commit (or
502 # doing anything else with git).
503 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{'script_version'};
504 $ENV{"CRUNCH_SRC"} = $Job->{'script_version'};
507 # Resolve the given script_version to a git commit sha1. Also, if
508 # the repository is remote, clone it into our local filesystem: this
509 # ensures "git archive" will work, and is necessary to reliably
510 # resolve a symbolic script_version like "master^".
511 $ENV{"CRUNCH_SRC"} = "$ENV{CRUNCH_TMP}/src";
513 Log (undef, "Looking for version ".$Job->{script_version}." from repository ".$Job->{repository});
515 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{script_version};
517 # If we're running under crunch-dispatch, it will have already
518 # pulled the appropriate source tree into its own repository, and
519 # given us that repo's path as $git_dir.
521 # If we're running a "local" job, we might have to fetch content
522 # from a remote repository.
524 # (Currently crunch-dispatch gives a local path with --git-dir, but
525 # we might as well accept URLs there too in case it changes its
527 my $repo = $git_dir || $Job->{'repository'};
529 # Repository can be remote or local. If remote, we'll need to fetch it
530 # to a local dir before doing `git log` et al.
533 if ($repo =~ m{://|^[^/]*:}) {
534 # $repo is a git url we can clone, like git:// or https:// or
535 # file:/// or [user@]host:repo.git. Note "user/name@host:foo" is
536 # not recognized here because distinguishing that from a local
537 # path is too fragile. If you really need something strange here,
538 # use the ssh:// form.
539 $repo_location = 'remote';
540 } elsif ($repo =~ m{^\.*/}) {
541 # $repo is a local path to a git index. We'll also resolve ../foo
542 # to ../foo/.git if the latter is a directory. To help
543 # disambiguate local paths from named hosted repositories, this
544 # form must be given as ./ or ../ if it's a relative path.
545 if (-d "$repo/.git") {
546 $repo = "$repo/.git";
548 $repo_location = 'local';
550 # $repo is none of the above. It must be the name of a hosted
552 my $arv_repo_list = api_call("repositories/list",
553 'filters' => [['name','=',$repo]]);
554 my @repos_found = @{$arv_repo_list->{'items'}};
555 my $n_found = $arv_repo_list->{'serverResponse'}->{'items_available'};
557 Log(undef, "Repository '$repo' -> "
558 . join(", ", map { $_->{'uuid'} } @repos_found));
561 croak("Error: Found $n_found repositories with name '$repo'.");
563 $repo = $repos_found[0]->{'fetch_url'};
564 $repo_location = 'remote';
566 Log(undef, "Using $repo_location repository '$repo'");
567 $ENV{"CRUNCH_SRC_URL"} = $repo;
569 # Resolve given script_version (we'll call that $treeish here) to a
570 # commit sha1 ($commit).
571 my $treeish = $Job->{'script_version'};
573 if ($repo_location eq 'remote') {
574 # We minimize excess object-fetching by re-using the same bare
575 # repository in CRUNCH_TMP/.git for multiple crunch-jobs -- we
576 # just keep adding remotes to it as needed.
577 my $local_repo = $ENV{'CRUNCH_TMP'}."/.git";
578 my $gitcmd = "git --git-dir=\Q$local_repo\E";
580 # Set up our local repo for caching remote objects, making
582 if (!-d $local_repo) {
583 make_path($local_repo) or croak("Error: could not create $local_repo");
585 # This works (exits 0 and doesn't delete fetched objects) even
586 # if $local_repo is already initialized:
587 `$gitcmd init --bare`;
589 croak("Error: $gitcmd init --bare exited ".exit_status_s($?));
592 # If $treeish looks like a hash (or abbrev hash) we look it up in
593 # our local cache first, since that's cheaper. (We don't want to
594 # do that with tags/branches though -- those change over time, so
595 # they should always be resolved by the remote repo.)
596 if ($treeish =~ /^[0-9a-f]{7,40}$/s) {
597 # Hide stderr because it's normal for this to fail:
598 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E 2>/dev/null`;
600 # Careful not to resolve a branch named abcdeff to commit 1234567:
601 $sha1 =~ /^$treeish/ &&
602 $sha1 =~ /^([0-9a-f]{40})$/s) {
604 Log(undef, "Commit $commit already present in $local_repo");
608 if (!defined $commit) {
609 # If $treeish isn't just a hash or abbrev hash, or isn't here
610 # yet, we need to fetch the remote to resolve it correctly.
612 # First, remove all local heads. This prevents a name that does
613 # not exist on the remote from resolving to (or colliding with)
614 # a previously fetched branch or tag (possibly from a different
616 remove_tree("$local_repo/refs/heads", {keep_root => 1});
618 Log(undef, "Fetching objects from $repo to $local_repo");
619 `$gitcmd fetch --no-progress --tags ''\Q$repo\E \Q+refs/heads/*:refs/heads/*\E`;
621 croak("Error: `$gitcmd fetch` exited ".exit_status_s($?));
625 # Now that the data is all here, we will use our local repo for
626 # the rest of our git activities.
630 my $gitcmd = "git --git-dir=\Q$repo\E";
631 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E`;
632 unless ($? == 0 && $sha1 =~ /^([0-9a-f]{40})$/) {
633 croak("`$gitcmd rev-list` exited "
635 .", '$treeish' not found, giving up");
638 Log(undef, "Version $treeish is commit $commit");
640 if ($commit ne $Job->{'script_version'}) {
641 # Record the real commit id in the database, frozentokey, logs,
642 # etc. -- instead of an abbreviation or a branch name which can
643 # become ambiguous or point to a different commit in the future.
644 if (!$Job->update_attributes('script_version' => $commit)) {
645 croak("Error: failed to update job's script_version attribute");
649 $ENV{"CRUNCH_SRC_COMMIT"} = $commit;
650 add_git_archive("$gitcmd archive ''\Q$commit\E");
653 my $git_archive = combined_git_archive();
654 if (!defined $git_archive) {
655 Log(undef, "Skip install phase (no git archive)");
657 Log(undef, "Warning: This probably means workers have no source tree!");
662 my $install_script_tries_left = 3;
663 for (my $attempts = 0; $attempts < 3; $attempts++) {
664 my @srunargs = ("srun",
665 "--nodelist=$nodelist",
666 "-D", $ENV{'TMPDIR'}, "--job-name=$job_id");
667 my @execargs = ("sh", "-c",
668 "mkdir -p $ENV{CRUNCH_INSTALL} && cd $ENV{CRUNCH_TMP} && perl -");
670 $ENV{"CRUNCH_GIT_ARCHIVE_HASH"} = md5_hex($git_archive);
671 my ($stdout, $stderr);
672 ($exited, $stdout, $stderr) = srun_sync(
673 \@srunargs, \@execargs,
674 {label => "run install script on all workers"},
675 $build_script . $git_archive);
677 my $stderr_anything_from_script = 0;
678 for my $line (split(/\n/, $stderr)) {
679 if ($line !~ /^(srun: error: |starting: \[)/) {
680 $stderr_anything_from_script = 1;
684 last if $exited == 0 || $main::please_freeze;
686 # If the install script fails but doesn't print an error message,
687 # the next thing anyone is likely to do is just run it again in
688 # case it was a transient problem like "slurm communication fails
689 # because the network isn't reliable enough". So we'll just do
690 # that ourselves (up to 3 attempts in total). OTOH, if there is an
691 # error message, the problem is more likely to have a real fix and
692 # we should fail the job so the fixing process can start, instead
693 # of doing 2 more attempts.
694 last if $stderr_anything_from_script;
697 foreach my $tar_filename (map { tar_filename_n($_); } (1..$git_tar_count)) {
698 unlink($tar_filename);
706 foreach (qw (script script_version script_parameters runtime_constraints))
710 (ref($Job->{$_}) ? JSON::encode_json($Job->{$_}) : $Job->{$_}));
712 foreach (split (/\n/, $Job->{knobs}))
714 Log (undef, "knob " . $_);
719 $main::success = undef;
725 my $thisround_succeeded = 0;
726 my $thisround_failed = 0;
727 my $thisround_failed_multiple = 0;
728 my $working_slot_count = scalar(@slot);
730 @jobstep_todo = sort { $jobstep[$a]->{level} <=> $jobstep[$b]->{level}
731 or $a <=> $b } @jobstep_todo;
732 my $level = $jobstep[$jobstep_todo[0]]->{level};
734 my $initial_tasks_this_level = 0;
735 foreach my $id (@jobstep_todo) {
736 $initial_tasks_this_level++ if ($jobstep[$id]->{level} == $level);
739 # If the number of tasks scheduled at this level #T is smaller than the number
740 # of slots available #S, only use the first #T slots, or the first slot on
741 # each node, whichever number is greater.
743 # When we dispatch tasks later, we'll allocate whole-node resources like RAM
744 # based on these numbers. Using fewer slots makes more resources available
745 # to each individual task, which should normally be a better strategy when
746 # there are fewer of them running with less parallelism.
748 # Note that this calculation is not redone if the initial tasks at
749 # this level queue more tasks at the same level. This may harm
750 # overall task throughput for that level.
752 if ($initial_tasks_this_level < @node) {
753 @freeslot = (0..$#node);
754 } elsif ($initial_tasks_this_level < @slot) {
755 @freeslot = (0..$initial_tasks_this_level - 1);
757 @freeslot = (0..$#slot);
759 my $round_num_freeslots = scalar(@freeslot);
760 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";
762 my %round_max_slots = ();
763 for (my $ii = $#freeslot; $ii >= 0; $ii--) {
764 my $this_slot = $slot[$freeslot[$ii]];
765 my $node_name = $this_slot->{node}->{name};
766 $round_max_slots{$node_name} ||= $this_slot->{cpu};
767 last if (scalar(keys(%round_max_slots)) >= @node);
770 Log(undef, "start level $level with $round_num_freeslots slots");
773 my $progress_is_dirty = 1;
774 my $progress_stats_updated = 0;
776 update_progress_stats();
780 for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
782 # Don't create new tasks if we already know the job's final result.
783 last if defined($main::success);
785 my $id = $jobstep_todo[$todo_ptr];
786 my $Jobstep = $jobstep[$id];
787 if ($Jobstep->{level} != $level)
792 pipe $reader{$id}, "writer" or croak("pipe() failed: $!");
793 set_nonblocking($reader{$id});
795 my $childslot = $freeslot[0];
796 my $childnode = $slot[$childslot]->{node};
797 my $childslotname = join (".",
798 $slot[$childslot]->{node}->{name},
799 $slot[$childslot]->{cpu});
801 my $childpid = fork();
804 $SIG{'INT'} = 'DEFAULT';
805 $SIG{'QUIT'} = 'DEFAULT';
806 $SIG{'TERM'} = 'DEFAULT';
808 foreach (values (%reader))
812 fcntl ("writer", F_SETFL, 0) or croak ($!); # no close-on-exec
813 open(STDOUT,">&writer");
814 open(STDERR,">&writer");
819 delete $ENV{"GNUPGHOME"};
820 $ENV{"TASK_UUID"} = $Jobstep->{'arvados_task'}->{'uuid'};
821 $ENV{"TASK_QSEQUENCE"} = $id;
822 $ENV{"TASK_SEQUENCE"} = $level;
823 $ENV{"JOB_SCRIPT"} = $Job->{script};
824 while (my ($param, $value) = each %{$Job->{script_parameters}}) {
825 $param =~ tr/a-z/A-Z/;
826 $ENV{"JOB_PARAMETER_$param"} = $value;
828 $ENV{"TASK_SLOT_NODE"} = $slot[$childslot]->{node}->{name};
829 $ENV{"TASK_SLOT_NUMBER"} = $slot[$childslot]->{cpu};
830 $ENV{"TASK_WORK"} = $ENV{"CRUNCH_TMP"}."/task/$childslotname";
831 $ENV{"HOME"} = $ENV{"TASK_WORK"};
832 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
833 $ENV{"CRUNCH_NODE_SLOTS"} = $round_max_slots{$ENV{TASK_SLOT_NODE}};
834 $ENV{"PATH"} = $ENV{"CRUNCH_INSTALL"} . "/bin:" . $ENV{"PATH"};
836 my $keep_mnt = $ENV{"TASK_WORK"}.".keep";
842 "--nodelist=".$childnode->{name},
843 qw(-n1 -c1 -N1 -D), $ENV{'TMPDIR'},
844 "--job-name=$job_id.$id.$$",
847 my $stdbuf = " stdbuf --output=0 --error=0 ";
849 my $arv_file_cache = "";
850 if (defined($Job->{'runtime_constraints'}->{'keep_cache_mb_per_task'})) {
851 $arv_file_cache = "--file-cache=" . ($Job->{'runtime_constraints'}->{'keep_cache_mb_per_task'} * 1024 * 1024);
855 "if [ -e \Q$ENV{TASK_WORK}\E ]; then rm -rf \Q$ENV{TASK_WORK}\E; fi; "
856 ."mkdir -p \Q$ENV{CRUNCH_TMP}\E \Q$ENV{JOB_WORK}\E \Q$ENV{TASK_WORK}\E \Q$keep_mnt\E "
857 ."&& cd \Q$ENV{CRUNCH_TMP}\E "
858 # These environment variables get used explicitly later in
859 # $command. No tool is expected to read these values directly.
860 .q{&& MEM=$(awk '($1 == "MemTotal:"){print $2}' </proc/meminfo) }
861 .q{&& SWAP=$(awk '($1 == "SwapTotal:"){print $2}' </proc/meminfo) }
862 ."&& MEMLIMIT=\$(( (\$MEM * 95) / ($ENV{CRUNCH_NODE_SLOTS} * 100) )) "
863 ."&& let SWAPLIMIT=\$MEMLIMIT+\$SWAP ";
865 $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 ";
866 $ENV{TASK_KEEPMOUNT} = "$keep_mnt/by_pdh";
867 $ENV{TASK_KEEPMOUNT_TMP} = "$keep_mnt/tmp";
871 my $containername = "$Jobstep->{arvados_task}->{uuid}-$Jobstep->{failures}";
872 my $cidfile = "$ENV{CRUNCH_TMP}/$containername.cid";
873 $command .= "crunchstat -cgroup-root=\Q$cgroup_root\E -cgroup-parent=docker -cgroup-cid=$cidfile -poll=10000 ";
874 $command .= "$docker_bin run $docker_run_args --name=$containername --attach=stdout --attach=stderr --attach=stdin -i \Q$dockeruserarg\E --cidfile=$cidfile --sig-proxy ";
875 # We only set memory limits if Docker lets us limit both memory and swap.
876 # Memory limits alone have been supported longer, but subprocesses tend
877 # to get SIGKILL if they exceed that without any swap limit set.
878 # See #5642 for additional background.
879 if ($docker_limitmem) {
880 $command .= "--memory=\${MEMLIMIT}k --memory-swap=\${SWAPLIMIT}k ";
883 # The source tree and $destdir directory (which we have
884 # installed on the worker host) are available in the container,
885 # under the same path.
886 $command .= "--volume=\Q$ENV{CRUNCH_SRC}:$ENV{CRUNCH_SRC}:ro\E ";
887 $command .= "--volume=\Q$ENV{CRUNCH_INSTALL}:$ENV{CRUNCH_INSTALL}:ro\E ";
889 # Currently, we make the "by_pdh" directory in arv-mount's mount
890 # point appear at /keep inside the container (instead of using
891 # the same path as the host like we do with CRUNCH_SRC and
892 # CRUNCH_INSTALL). However, crunch scripts and utilities must
893 # not rely on this. They must use $TASK_KEEPMOUNT.
894 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT}:/keep:ro\E ";
895 $ENV{TASK_KEEPMOUNT} = "/keep";
897 # Ditto TASK_KEEPMOUNT_TMP, as /keep_tmp.
898 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT_TMP}:/keep_tmp\E ";
899 $ENV{TASK_KEEPMOUNT_TMP} = "/keep_tmp";
901 # TASK_WORK is almost exactly like a docker data volume: it
902 # starts out empty, is writable, and persists until no
903 # containers use it any more. We don't use --volumes-from to
904 # share it with other containers: it is only accessible to this
905 # task, and it goes away when this task stops.
907 # However, a docker data volume is writable only by root unless
908 # the mount point already happens to exist in the container with
909 # different permissions. Therefore, we [1] assume /tmp already
910 # exists in the image and is writable by the crunch user; [2]
911 # avoid putting TASK_WORK inside CRUNCH_TMP (which won't be
912 # writable if they are created by docker while setting up the
913 # other --volumes); and [3] create $TASK_WORK inside the
914 # container using $build_script.
915 $command .= "--volume=/tmp ";
916 $ENV{"TASK_WORK"} = "/tmp/crunch-job-task-work/$childslotname";
917 $ENV{"HOME"} = $ENV{"TASK_WORK"};
918 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
920 # TODO: Share a single JOB_WORK volume across all task
921 # containers on a given worker node, and delete it when the job
922 # ends (and, in case that doesn't work, when the next job
925 # For now, use the same approach as TASK_WORK above.
926 $ENV{"JOB_WORK"} = "/tmp/crunch-job-work";
928 # Bind mount the crunchrunner binary and host TLS certificates file into
930 $command .= "--volume=\Q$ENV{HOST_CRUNCHRUNNER_BIN}:/usr/local/bin/crunchrunner\E ";
931 $command .= "--volume=\Q$ENV{HOST_CERTS}:/etc/arvados/ca-certificates.crt\E ";
933 while (my ($env_key, $env_val) = each %ENV)
935 if ($env_key =~ /^(ARVADOS|CRUNCH|JOB|TASK)_/) {
936 $command .= "--env=\Q$env_key=$env_val\E ";
939 $command .= "--env=\QHOME=$ENV{HOME}\E ";
940 $command .= "\Q$docker_hash\E ";
942 if ($Job->{arvados_sdk_version}) {
944 $command .= "perl - \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E";
946 $command .= "/bin/sh -c \'python -c " .
947 '"from pkg_resources import get_distribution as get; print \"Using Arvados SDK version\", get(\"arvados-python-client\").version"' .
948 ">&2 2>/dev/null; " .
949 "mkdir -p \"$ENV{JOB_WORK}\" \"$ENV{TASK_WORK}\" && " .
950 "if which stdbuf >/dev/null ; then " .
951 " exec $stdbuf \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E ;" .
953 " exec \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E ;" .
958 $command .= "crunchstat -cgroup-root=\Q$cgroup_root\E -poll=10000 ";
960 $command .= "perl - $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
963 my @execargs = ('bash', '-c', $command);
964 srun (\@srunargs, \@execargs, undef, $build_script);
965 # exec() failed, we assume nothing happened.
966 die "srun() failed on build script\n";
969 if (!defined $childpid)
980 jobstepname => "$job_id.$id.$childpid",
982 croak ("assert failed: \$slot[$childslot]->{'pid'} exists") if exists $slot[$childslot]->{pid};
983 $slot[$childslot]->{pid} = $childpid;
985 Log ($id, "job_task ".$Jobstep->{'arvados_task'}->{'uuid'});
986 Log ($id, "child $childpid started on $childslotname");
987 $Jobstep->{starttime} = time;
988 $Jobstep->{node} = $childnode->{name};
989 $Jobstep->{slotindex} = $childslot;
990 delete $Jobstep->{stderr};
991 delete $Jobstep->{finishtime};
992 delete $Jobstep->{tempfail};
994 $Jobstep->{'arvados_task'}->{started_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{starttime});
995 $Jobstep->{'arvados_task'}->save;
997 splice @jobstep_todo, $todo_ptr, 1;
1000 $progress_is_dirty = 1;
1004 ($round_num_freeslots > @freeslot && $todo_ptr+1 > $#jobstep_todo))
1006 last THISROUND if $main::please_freeze;
1007 if ($main::please_info)
1009 $main::please_info = 0;
1011 create_output_collection();
1013 update_progress_stats();
1018 if (!$gotsome || ($latest_refresh + 2 < scalar time))
1020 check_refresh_wanted();
1022 update_progress_stats();
1024 elsif (time - $progress_stats_updated >= 30 || $progress_is_dirty)
1026 update_progress_stats();
1029 select (undef, undef, undef, 0.1);
1031 $working_slot_count = scalar(grep { $_->{node}->{fail_count} == 0 &&
1032 $_->{node}->{hold_count} < 4 } @slot);
1033 if (($thisround_failed_multiple >= 8 && $thisround_succeeded == 0) ||
1034 ($thisround_failed_multiple >= 16 && $thisround_failed_multiple > $thisround_succeeded))
1036 my $message = "Repeated failure rate too high ($thisround_failed_multiple/"
1037 .($thisround_failed+$thisround_succeeded)
1038 .") -- giving up on this round";
1039 Log (undef, $message);
1043 # move slots from freeslot to holdslot (or back to freeslot) if necessary
1044 for (my $i=$#freeslot; $i>=0; $i--) {
1045 if ($slot[$freeslot[$i]]->{node}->{hold_until} > scalar time) {
1046 push @holdslot, (splice @freeslot, $i, 1);
1049 for (my $i=$#holdslot; $i>=0; $i--) {
1050 if ($slot[$holdslot[$i]]->{node}->{hold_until} <= scalar time) {
1051 push @freeslot, (splice @holdslot, $i, 1);
1055 # give up if no nodes are succeeding
1056 if ($working_slot_count < 1) {
1057 Log(undef, "Every node has failed -- giving up");
1064 push @freeslot, splice @holdslot;
1065 map { $slot[$freeslot[$_]]->{node}->{losing_streak} = 0 } (0..$#freeslot);
1068 Log (undef, "wait for last ".(scalar keys %proc)." children to finish");
1071 if ($main::please_continue) {
1072 $main::please_continue = 0;
1075 $main::please_info = 0, freeze(), create_output_collection(), save_meta(1) if $main::please_info;
1077 if (!reapchildren())
1079 check_refresh_wanted();
1081 update_progress_stats();
1082 select (undef, undef, undef, 0.1);
1083 killem (keys %proc) if $main::please_freeze;
1087 update_progress_stats();
1088 freeze_if_want_freeze();
1091 if (!defined $main::success)
1093 if (!@jobstep_todo) {
1095 } elsif ($working_slot_count < 1) {
1096 save_output_collection();
1098 exit(EX_RETRY_UNLOCKED);
1099 } elsif ($thisround_succeeded == 0 &&
1100 ($thisround_failed == 0 || $thisround_failed > 4)) {
1101 my $message = "stop because $thisround_failed tasks failed and none succeeded";
1102 Log (undef, $message);
1107 goto ONELEVEL if !defined $main::success;
1110 release_allocation();
1112 my $collated_output = save_output_collection();
1113 Log (undef, "finish");
1118 if ($collated_output && $main::success) {
1119 $final_state = 'Complete';
1121 $final_state = 'Failed';
1123 $Job->update_attributes('state' => $final_state);
1125 exit (($final_state eq 'Complete') ? 0 : 1);
1129 sub update_progress_stats
1131 $progress_stats_updated = time;
1132 return if !$progress_is_dirty;
1133 my ($todo, $done, $running) = (scalar @jobstep_todo,
1134 scalar @jobstep_done,
1135 scalar keys(%proc));
1136 $Job->{'tasks_summary'} ||= {};
1137 $Job->{'tasks_summary'}->{'todo'} = $todo;
1138 $Job->{'tasks_summary'}->{'done'} = $done;
1139 $Job->{'tasks_summary'}->{'running'} = $running;
1140 $Job->update_attributes('tasks_summary' => $Job->{'tasks_summary'});
1141 Log (undef, "status: $done done, $running running, $todo todo");
1142 $progress_is_dirty = 0;
1149 my $children_reaped = 0;
1150 my @successful_task_uuids = ();
1152 while((my $pid = waitpid (-1, WNOHANG)) > 0)
1154 my $childstatus = $?;
1156 my $whatslot = ($slot[$proc{$pid}->{slot}]->{node}->{name}
1158 . $slot[$proc{$pid}->{slot}]->{cpu});
1159 my $jobstepidx = $proc{$pid}->{jobstepidx};
1161 if (!WIFEXITED($childstatus))
1163 # child did not exit (may be temporarily stopped)
1164 Log ($jobstepidx, "child $pid did not actually exit in reapchildren, ignoring for now.");
1169 my $elapsed = time - $proc{$pid}->{time};
1170 my $Jobstep = $jobstep[$jobstepidx];
1172 my $exitvalue = $childstatus >> 8;
1173 my $exitinfo = "exit ".exit_status_s($childstatus);
1174 $Jobstep->{'arvados_task'}->reload;
1175 my $task_success = $Jobstep->{'arvados_task'}->{success};
1177 Log ($jobstepidx, "child $pid on $whatslot $exitinfo success=$task_success");
1179 if (!defined $task_success) {
1180 # task did not indicate one way or the other --> fail
1181 Log($jobstepidx, sprintf(
1182 "ERROR: Task process exited %s, but never updated its task record to indicate success and record its output.",
1183 exit_status_s($childstatus)));
1184 $Jobstep->{'arvados_task'}->{success} = 0;
1185 $Jobstep->{'arvados_task'}->save;
1192 $temporary_fail ||= $Jobstep->{tempfail};
1193 $temporary_fail ||= ($exitvalue == TASK_TEMPFAIL);
1195 ++$thisround_failed;
1196 ++$thisround_failed_multiple if $Jobstep->{'failures'} >= 1;
1198 # Check for signs of a failed or misconfigured node
1199 if (++$slot[$proc{$pid}->{slot}]->{node}->{losing_streak} >=
1200 2+$slot[$proc{$pid}->{slot}]->{node}->{ncpus}) {
1201 # Don't count this against jobstep failure thresholds if this
1202 # node is already suspected faulty and srun exited quickly
1203 if ($slot[$proc{$pid}->{slot}]->{node}->{hold_until} &&
1205 Log ($jobstepidx, "blaming failure on suspect node " .
1206 $slot[$proc{$pid}->{slot}]->{node}->{name});
1207 $temporary_fail ||= 1;
1209 ban_node_by_slot($proc{$pid}->{slot});
1212 Log ($jobstepidx, sprintf('failure (#%d, %s) after %d seconds',
1213 ++$Jobstep->{'failures'},
1214 $temporary_fail ? 'temporary' : 'permanent',
1217 if (!$temporary_fail || $Jobstep->{'failures'} >= 3) {
1218 # Give up on this task, and the whole job
1221 # Put this task back on the todo queue
1222 push @jobstep_todo, $jobstepidx;
1223 $Job->{'tasks_summary'}->{'failed'}++;
1227 push @successful_task_uuids, $Jobstep->{'arvados_task'}->{uuid};
1228 ++$thisround_succeeded;
1229 $slot[$proc{$pid}->{slot}]->{node}->{losing_streak} = 0;
1230 $slot[$proc{$pid}->{slot}]->{node}->{hold_until} = 0;
1231 $slot[$proc{$pid}->{slot}]->{node}->{fail_count} = 0;
1232 push @jobstep_done, $jobstepidx;
1233 Log ($jobstepidx, "success in $elapsed seconds");
1235 $Jobstep->{exitcode} = $childstatus;
1236 $Jobstep->{finishtime} = time;
1237 $Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
1238 $Jobstep->{'arvados_task'}->save;
1239 process_stderr_final ($jobstepidx);
1240 Log ($jobstepidx, sprintf("task output (%d bytes): %s",
1241 length($Jobstep->{'arvados_task'}->{output}),
1242 $Jobstep->{'arvados_task'}->{output}));
1244 close $reader{$jobstepidx};
1245 delete $reader{$jobstepidx};
1246 delete $slot[$proc{$pid}->{slot}]->{pid};
1247 push @freeslot, $proc{$pid}->{slot};
1250 $progress_is_dirty = 1;
1253 if (scalar(@successful_task_uuids) > 0)
1255 Log (undef, sprintf("%d tasks exited (%d succeeded), checking for new tasks from API server.", $children_reaped, scalar(@successful_task_uuids)));
1257 my $newtask_list = [];
1258 my $newtask_results;
1260 $newtask_results = api_call(
1262 'filters' => [["created_by_job_task_uuid","in",\@successful_task_uuids]],
1263 'order' => 'qsequence',
1264 'offset' => scalar(@$newtask_list),
1266 push(@$newtask_list, @{$newtask_results->{items}});
1267 } while (@{$newtask_results->{items}});
1268 Log (undef, sprintf("Got %d new tasks from API server.", scalar(@$newtask_list)));
1269 foreach my $arvados_task (@$newtask_list) {
1271 'level' => $arvados_task->{'sequence'},
1273 'arvados_task' => $arvados_task
1275 push @jobstep, $jobstep;
1276 push @jobstep_todo, $#jobstep;
1280 return $children_reaped;
1283 sub check_refresh_wanted
1285 my @stat = stat $ENV{"CRUNCH_REFRESH_TRIGGER"};
1287 $stat[9] > $latest_refresh &&
1288 # ...and we have actually locked the job record...
1289 $job_id eq $Job->{'uuid'}) {
1290 $latest_refresh = scalar time;
1291 my $Job2 = api_call("jobs/get", uuid => $jobspec);
1292 for my $attr ('cancelled_at',
1293 'cancelled_by_user_uuid',
1294 'cancelled_by_client_uuid',
1296 $Job->{$attr} = $Job2->{$attr};
1298 if ($Job->{'state'} ne "Running") {
1299 if ($Job->{'state'} eq "Cancelled") {
1300 Log (undef, "Job cancelled at " . $Job->{'cancelled_at'} . " by user " . $Job->{'cancelled_by_user_uuid'});
1302 Log (undef, "Job state unexpectedly changed to " . $Job->{'state'});
1305 $main::please_freeze = 1;
1312 my $last_squeue_check = $squeue_checked;
1314 # Do not call `squeue` or check the kill list more than once every
1316 return if $last_squeue_check > time - 15;
1317 $squeue_checked = time;
1319 # Look for children from which we haven't received stderr data since
1320 # the last squeue check. If no such children exist, all procs are
1321 # alive and there's no need to even look at squeue.
1323 # As long as the crunchstat poll interval (10s) is shorter than the
1324 # squeue check interval (15s) this should make the squeue check an
1326 my $silent_procs = 0;
1327 for my $js (map {$jobstep[$_->{jobstepidx}]} values %proc)
1329 if (!exists($js->{stderr_at}))
1331 $js->{stderr_at} = 0;
1333 if ($js->{stderr_at} < $last_squeue_check)
1338 return if $silent_procs == 0;
1340 # use killem() on procs whose killtime is reached
1341 while (my ($pid, $procinfo) = each %proc)
1343 my $js = $jobstep[$procinfo->{jobstepidx}];
1344 if (exists $procinfo->{killtime}
1345 && $procinfo->{killtime} <= time
1346 && $js->{stderr_at} < $last_squeue_check)
1349 if ($js->{stderr_at}) {
1350 $sincewhen = " in last " . (time - $js->{stderr_at}) . "s";
1352 Log($procinfo->{jobstepidx}, "killing orphaned srun process $pid (task not in slurm queue, no stderr received$sincewhen)");
1359 # here is an opportunity to check for mysterious problems with local procs
1363 # Get a list of steps still running. Note: squeue(1) says --steps
1364 # selects a format (which we override anyway) and allows us to
1365 # specify which steps we're interested in (which we don't).
1366 # Importantly, it also changes the meaning of %j from "job name" to
1367 # "step name" and (although this isn't mentioned explicitly in the
1368 # docs) switches from "one line per job" mode to "one line per step"
1369 # mode. Without it, we'd just get a list of one job, instead of a
1371 my @squeue = `squeue --jobs=\Q$ENV{SLURM_JOB_ID}\E --steps --format='%j' --noheader`;
1374 Log(undef, "warning: squeue exit status $? ($!)");
1379 # which of my jobsteps are running, according to squeue?
1381 for my $jobstepname (@squeue)
1383 $ok{$jobstepname} = 1;
1386 # Check for child procs >60s old and not mentioned by squeue.
1387 while (my ($pid, $procinfo) = each %proc)
1389 if ($procinfo->{time} < time - 60
1390 && $procinfo->{jobstepname}
1391 && !exists $ok{$procinfo->{jobstepname}}
1392 && !exists $procinfo->{killtime})
1394 # According to slurm, this task has ended (successfully or not)
1395 # -- but our srun child hasn't exited. First we must wait (30
1396 # seconds) in case this is just a race between communication
1397 # channels. Then, if our srun child process still hasn't
1398 # terminated, we'll conclude some slurm communication
1399 # error/delay has caused the task to die without notifying srun,
1400 # and we'll kill srun ourselves.
1401 $procinfo->{killtime} = time + 30;
1402 Log($procinfo->{jobstepidx}, "notice: task is not in slurm queue but srun process $pid has not exited");
1408 sub release_allocation
1412 Log (undef, "release job allocation");
1413 system "scancel $ENV{SLURM_JOB_ID}";
1422 my $sel = IO::Select->new();
1423 foreach my $jobstepidx (keys %reader)
1425 my $fd = $reader{$jobstepidx};
1427 $fd_job{$fd} = $jobstepidx;
1429 if (my $stdout_fd = $jobstep[$jobstepidx]->{stdout_r}) {
1430 $sel->add($stdout_fd);
1431 $fd_job{$stdout_fd} = $jobstepidx;
1434 # select on all reader fds with 0.1s timeout
1435 my @ready_fds = $sel->can_read(0.1);
1436 foreach my $fd (@ready_fds)
1439 if (0 < sysread ($fd, $buf, 65536))
1442 print STDERR $buf if $ENV{CRUNCH_DEBUG};
1444 my $jobstepidx = $fd_job{$fd};
1445 if ($jobstep[$jobstepidx]->{stdout_r} == $fd) {
1446 $jobstep[$jobstepidx]->{stdout_captured} .= $buf;
1450 $jobstep[$jobstepidx]->{stderr_at} = time;
1451 $jobstep[$jobstepidx]->{stderr} .= $buf;
1453 # Consume everything up to the last \n
1454 preprocess_stderr ($jobstepidx);
1456 if (length ($jobstep[$jobstepidx]->{stderr}) > 16384)
1458 # If we get a lot of stderr without a newline, chop off the
1459 # front to avoid letting our buffer grow indefinitely.
1460 substr ($jobstep[$jobstepidx]->{stderr},
1461 0, length($jobstep[$jobstepidx]->{stderr}) - 8192) = "";
1469 # Consume all full lines of stderr for a jobstep. Everything after the
1470 # last newline will remain in $jobstep[$jobstepidx]->{stderr} after
1472 sub preprocess_stderr
1474 my $jobstepidx = shift;
1476 while ($jobstep[$jobstepidx]->{stderr} =~ /^(.*?)\n/) {
1478 substr $jobstep[$jobstepidx]->{stderr}, 0, 1+length($line), "";
1479 Log ($jobstepidx, "stderr $line");
1480 if ($line =~ /srun: error: (SLURM job $ENV{SLURM_JOB_ID} has expired|Unable to confirm allocation for job $ENV{SLURM_JOB_ID})/) {
1482 $main::please_freeze = 1;
1484 elsif (!exists $jobstep[$jobstepidx]->{slotindex}) {
1485 # Skip the following tempfail checks if this srun proc isn't
1486 # attached to a particular worker slot.
1488 elsif ($line =~ /srun: error: (Node failure on|Aborting, .*\bio error\b)/) {
1489 my $job_slot_index = $jobstep[$jobstepidx]->{slotindex};
1490 $slot[$job_slot_index]->{node}->{fail_count}++;
1491 $jobstep[$jobstepidx]->{tempfail} = 1;
1492 ban_node_by_slot($job_slot_index);
1494 elsif ($line =~ /srun: error: (Unable to create job step|.*: Communication connection failure)/) {
1495 $jobstep[$jobstepidx]->{tempfail} = 1;
1496 ban_node_by_slot($jobstep[$jobstepidx]->{slotindex});
1498 elsif ($line =~ /\bKeep(Read|Write|Request)Error:/) {
1499 $jobstep[$jobstepidx]->{tempfail} = 1;
1505 sub process_stderr_final
1507 my $jobstepidx = shift;
1508 preprocess_stderr ($jobstepidx);
1511 Log ($jobstepidx, "stderr $_");
1512 } split ("\n", $jobstep[$jobstepidx]->{stderr});
1513 $jobstep[$jobstepidx]->{stderr} = '';
1520 if (!open($keep, "-|", "arv-get", "--retries", retry_count(), $hash)) {
1521 Log(undef, "fetch_block run error from arv-get $hash: $!");
1524 my $output_block = "";
1527 my $bytes = sysread($keep, $buf, 1024 * 1024);
1528 if (!defined $bytes) {
1529 Log(undef, "fetch_block read error from arv-get: $!");
1530 $output_block = undef;
1532 } elsif ($bytes == 0) {
1533 # sysread returns 0 at the end of the pipe.
1536 # some bytes were read into buf.
1537 $output_block .= $buf;
1542 Log(undef, "fetch_block arv-get exited " . exit_status_s($?));
1543 $output_block = undef;
1545 return $output_block;
1548 # Create a collection by concatenating the output of all tasks (each
1549 # task's output is either a manifest fragment, a locator for a
1550 # manifest fragment stored in Keep, or nothing at all). Return the
1551 # portable_data_hash of the new collection.
1552 sub create_output_collection
1554 Log (undef, "collate");
1556 my ($child_out, $child_in);
1557 my $pid = open2($child_out, $child_in, 'python', '-c', q{
1560 print (arvados.api("v1").collections().
1561 create(body={"manifest_text": sys.stdin.read()}).
1562 execute(num_retries=int(sys.argv[1]))["portable_data_hash"])
1566 my $manifest_size = 0;
1570 my $output = $_->{'arvados_task'}->{output};
1571 next if (!defined($output));
1573 if ($output =~ /^[0-9a-f]{32}(\+\S+)*$/) {
1574 $next_write = fetch_block($output);
1576 $next_write = $output;
1578 if (defined($next_write)) {
1579 if (!defined(syswrite($child_in, $next_write))) {
1580 # There's been an error writing. Stop the loop.
1581 # We'll log details about the exit code later.
1584 $manifest_size += length($next_write);
1587 my $uuid = $_->{'arvados_task'}->{'uuid'};
1588 Log (undef, "Error retrieving '$output' output by task $task_idx ($uuid)");
1593 Log(undef, "collated output manifest text to send to API server is $manifest_size bytes with access tokens");
1596 my $s = IO::Select->new($child_out);
1597 if ($s->can_read(120)) {
1598 sysread($child_out, $joboutput, 1024 * 1024);
1601 Log(undef, "output collection creation exited " . exit_status_s($?));
1607 Log (undef, "timed out while creating output collection");
1608 foreach my $signal (2, 2, 2, 15, 15, 9) {
1609 kill($signal, $pid);
1610 last if waitpid($pid, WNOHANG) == -1;
1619 # Calls create_output_collection, logs the result, and returns it.
1620 # If that was successful, save that as the output in the job record.
1621 sub save_output_collection {
1622 my $collated_output = create_output_collection();
1624 if (!$collated_output) {
1625 Log(undef, "Failed to write output collection");
1628 Log(undef, "job output $collated_output");
1629 $Job->update_attributes('output' => $collated_output);
1631 return $collated_output;
1638 my $sig = 2; # SIGINT first
1639 if (exists $proc{$_}->{"sent_$sig"} &&
1640 time - $proc{$_}->{"sent_$sig"} > 4)
1642 $sig = 15; # SIGTERM if SIGINT doesn't work
1644 if (exists $proc{$_}->{"sent_$sig"} &&
1645 time - $proc{$_}->{"sent_$sig"} > 4)
1647 $sig = 9; # SIGKILL if SIGTERM doesn't work
1649 if (!exists $proc{$_}->{"sent_$sig"})
1651 Log ($proc{$_}->{jobstepidx}, "sending 2x signal $sig to pid $_");
1653 select (undef, undef, undef, 0.1);
1656 kill $sig, $_; # srun wants two SIGINT to really interrupt
1658 $proc{$_}->{"sent_$sig"} = time;
1659 $proc{$_}->{"killedafter"} = time - $proc{$_}->{"time"};
1669 vec($bits,fileno($_),1) = 1;
1675 # Send log output to Keep via arv-put.
1677 # $log_pipe_in and $log_pipe_out are the input and output filehandles to the arv-put pipe.
1678 # $log_pipe_out_buf is a string containing all output read from arv-put so far.
1679 # $log_pipe_out_select is an IO::Select object around $log_pipe_out.
1680 # $log_pipe_pid is the pid of the arv-put subprocess.
1682 # The only functions that should access these variables directly are:
1684 # log_writer_start($logfilename)
1685 # Starts an arv-put pipe, reading data on stdin and writing it to
1686 # a $logfilename file in an output collection.
1688 # log_writer_read_output([$timeout])
1689 # Read output from $log_pipe_out and append it to $log_pipe_out_buf.
1690 # Passes $timeout to the select() call, with a default of 0.01.
1691 # Returns the result of the last read() call on $log_pipe_out, or
1692 # -1 if read() wasn't called because select() timed out.
1693 # Only other log_writer_* functions should need to call this.
1695 # log_writer_send($txt)
1696 # Writes $txt to the output log collection.
1698 # log_writer_finish()
1699 # Closes the arv-put pipe and returns the output that it produces.
1701 # log_writer_is_active()
1702 # Returns a true value if there is currently a live arv-put
1703 # process, false otherwise.
1705 my ($log_pipe_in, $log_pipe_out, $log_pipe_out_buf, $log_pipe_out_select,
1708 sub log_writer_start($)
1710 my $logfilename = shift;
1711 $log_pipe_pid = open2($log_pipe_out, $log_pipe_in,
1715 '--filename', $logfilename,
1717 $log_pipe_out_buf = "";
1718 $log_pipe_out_select = IO::Select->new($log_pipe_out);
1721 sub log_writer_read_output {
1722 my $timeout = shift || 0.01;
1724 while ($read && $log_pipe_out_select->can_read($timeout)) {
1725 $read = read($log_pipe_out, $log_pipe_out_buf, 65536,
1726 length($log_pipe_out_buf));
1728 if (!defined($read)) {
1729 Log(undef, "error reading log manifest from arv-put: $!");
1734 sub log_writer_send($)
1737 print $log_pipe_in $txt;
1738 log_writer_read_output();
1741 sub log_writer_finish()
1743 return unless $log_pipe_pid;
1745 close($log_pipe_in);
1747 my $logger_failed = 0;
1748 my $read_result = log_writer_read_output(120);
1749 if ($read_result == -1) {
1750 $logger_failed = -1;
1751 Log (undef, "timed out reading from 'arv-put'");
1752 } elsif ($read_result != 0) {
1753 $logger_failed = -2;
1754 Log(undef, "failed to read arv-put log manifest to EOF");
1757 waitpid($log_pipe_pid, 0);
1759 $logger_failed ||= $?;
1760 Log(undef, "log_writer_finish: arv-put exited " . exit_status_s($?))
1763 close($log_pipe_out);
1764 my $arv_put_output = $logger_failed ? undef : $log_pipe_out_buf;
1765 $log_pipe_pid = $log_pipe_in = $log_pipe_out = $log_pipe_out_buf =
1766 $log_pipe_out_select = undef;
1768 return $arv_put_output;
1771 sub log_writer_is_active() {
1772 return $log_pipe_pid;
1775 sub Log # ($jobstepidx, $logmessage)
1777 my ($jobstepidx, $logmessage) = @_;
1778 if ($logmessage =~ /\n/) {
1779 for my $line (split (/\n/, $_[1])) {
1780 Log ($jobstepidx, $line);
1784 my $fh = select STDERR; $|=1; select $fh;
1786 if (defined($jobstepidx) && exists($jobstep[$jobstepidx]->{arvados_task})) {
1787 $task_qseq = $jobstepidx;
1789 my $message = sprintf ("%s %d %s %s", $job_id, $$, $task_qseq, $logmessage);
1790 $message =~ s{([^ -\176])}{"\\" . sprintf ("%03o", ord($1))}ge;
1793 if (log_writer_is_active() || -t STDERR) {
1794 my @gmtime = gmtime;
1795 $datetime = sprintf ("%04d-%02d-%02d_%02d:%02d:%02d",
1796 $gmtime[5]+1900, $gmtime[4]+1, @gmtime[3,2,1,0]);
1798 print STDERR ((-t STDERR) ? ($datetime." ".$message) : $message);
1800 if (log_writer_is_active()) {
1801 log_writer_send($datetime . " " . $message);
1808 my ($package, $file, $line) = caller;
1809 my $message = "@_ at $file line $line\n";
1810 Log (undef, $message);
1811 freeze() if @jobstep_todo;
1812 create_output_collection() if @jobstep_todo;
1822 if ($Job->{'state'} eq 'Cancelled') {
1823 $Job->update_attributes('finished_at' => scalar gmtime);
1825 $Job->update_attributes('state' => 'Failed');
1832 my $justcheckpoint = shift; # false if this will be the last meta saved
1833 return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm
1834 return unless log_writer_is_active();
1835 my $log_manifest = log_writer_finish();
1836 return unless defined($log_manifest);
1839 my $prev_log_coll = api_call("collections/get", uuid => $Job->{log});
1840 $log_manifest = $prev_log_coll->{manifest_text} . $log_manifest;
1843 my $log_coll = api_call(
1844 "collections/create", ensure_unique_name => 1, collection => {
1845 manifest_text => $log_manifest,
1846 owner_uuid => $Job->{owner_uuid},
1847 name => sprintf("Log from %s job %s", $Job->{script}, $Job->{uuid}),
1849 Log(undef, "log collection is " . $log_coll->{portable_data_hash});
1850 $Job->update_attributes('log' => $log_coll->{portable_data_hash});
1854 sub freeze_if_want_freeze
1856 if ($main::please_freeze)
1858 release_allocation();
1861 # kill some srun procs before freeze+stop
1862 map { $proc{$_} = {} } @_;
1865 killem (keys %proc);
1866 select (undef, undef, undef, 0.1);
1868 while (($died = waitpid (-1, WNOHANG)) > 0)
1870 delete $proc{$died};
1875 create_output_collection();
1885 Log (undef, "Freeze not implemented");
1892 croak ("Thaw not implemented");
1908 $s =~ s{\\(.)}{$1 eq "n" ? "\n" : $1}ge;
1915 my $srunargs = shift;
1916 my $execargs = shift;
1917 my $opts = shift || {};
1920 my $label = exists $opts->{label} ? $opts->{label} : "@$execargs";
1921 Log (undef, "$label: start");
1923 my ($stderr_r, $stderr_w);
1924 pipe $stderr_r, $stderr_w or croak("pipe() failed: $!");
1926 my ($stdout_r, $stdout_w);
1927 pipe $stdout_r, $stdout_w or croak("pipe() failed: $!");
1929 my $srunpid = fork();
1934 fcntl($stderr_w, F_SETFL, 0) or croak($!); # no close-on-exec
1935 fcntl($stdout_w, F_SETFL, 0) or croak($!);
1936 open(STDERR, ">&", $stderr_w);
1937 open(STDOUT, ">&", $stdout_w);
1938 srun ($srunargs, $execargs, $opts, $stdin);
1944 set_nonblocking($stderr_r);
1945 set_nonblocking($stdout_r);
1947 # Add entries to @jobstep and %proc so check_squeue() and
1948 # freeze_if_want_freeze() can treat it like a job task process.
1952 stderr_captured => '',
1953 stdout_r => $stdout_r,
1954 stdout_captured => '',
1956 my $jobstepidx = $#jobstep;
1958 jobstepidx => $jobstepidx,
1960 $reader{$jobstepidx} = $stderr_r;
1962 while ($srunpid != waitpid ($srunpid, WNOHANG)) {
1963 my $busy = readfrompipes();
1964 if (!$busy || ($latest_refresh + 2 < scalar time)) {
1965 check_refresh_wanted();
1969 select(undef, undef, undef, 0.1);
1971 killem(keys %proc) if $main::please_freeze;
1975 1 while readfrompipes();
1976 process_stderr_final ($jobstepidx);
1978 Log (undef, "$label: exit ".exit_status_s($exited));
1982 delete $proc{$srunpid};
1983 delete $reader{$jobstepidx};
1985 my $j = pop @jobstep;
1986 return ($exited, $j->{stdout_captured}, $j->{stderr_captured});
1992 my $srunargs = shift;
1993 my $execargs = shift;
1994 my $opts = shift || {};
1996 my $args = $have_slurm ? [@$srunargs, @$execargs] : $execargs;
1998 $Data::Dumper::Terse = 1;
1999 $Data::Dumper::Indent = 0;
2000 my $show_cmd = Dumper($args);
2001 $show_cmd =~ s/(TOKEN\\*=)[^\s\']+/${1}[...]/g;
2002 $show_cmd =~ s/\n/ /g;
2003 if ($opts->{fork}) {
2004 Log(undef, "starting: $show_cmd");
2006 # This is a child process: parent is in charge of reading our
2007 # stderr and copying it to Log() if needed.
2008 warn "starting: $show_cmd\n";
2011 if (defined $stdin) {
2012 my $child = open STDIN, "-|";
2013 defined $child or die "no fork: $!";
2015 print $stdin or die $!;
2016 close STDOUT or die $!;
2021 return system (@$args) if $opts->{fork};
2024 warn "ENV size is ".length(join(" ",%ENV));
2025 die "exec failed: $!: @$args";
2029 sub ban_node_by_slot {
2030 # Don't start any new jobsteps on this node for 60 seconds
2032 $slot[$slotid]->{node}->{hold_until} = 60 + scalar time;
2033 $slot[$slotid]->{node}->{hold_count}++;
2034 Log (undef, "backing off node " . $slot[$slotid]->{node}->{name} . " for 60 seconds");
2039 my ($lockfile, $error_message) = @_;
2040 open L, ">", $lockfile or croak("$lockfile: $!");
2041 if (!flock L, LOCK_EX|LOCK_NB) {
2042 croak("Can't lock $lockfile: $error_message\n");
2046 sub find_docker_image {
2047 # Given a Keep locator, check to see if it contains a Docker image.
2048 # If so, return its stream name and Docker hash.
2049 # If not, return undef for both values.
2050 my $locator = shift;
2051 my ($streamname, $filename);
2052 my $image = api_call("collections/get", uuid => $locator);
2054 foreach my $line (split(/\n/, $image->{manifest_text})) {
2055 my @tokens = split(/\s+/, $line);
2057 $streamname = shift(@tokens);
2058 foreach my $filedata (grep(/^\d+:\d+:/, @tokens)) {
2059 if (defined($filename)) {
2060 return (undef, undef); # More than one file in the Collection.
2062 $filename = (split(/:/, $filedata, 3))[2];
2067 if (defined($filename) and ($filename =~ /^([0-9A-Fa-f]{64})\.tar$/)) {
2068 return ($streamname, $1);
2070 return (undef, undef);
2075 # Calculate the number of times an operation should be retried,
2076 # assuming exponential backoff, and that we're willing to retry as
2077 # long as tasks have been running. Enforce a minimum of 3 retries.
2078 my ($starttime, $endtime, $timediff, $retries);
2080 $starttime = $jobstep[0]->{starttime};
2081 $endtime = $jobstep[-1]->{finishtime};
2083 if (!defined($starttime)) {
2085 } elsif (!defined($endtime)) {
2086 $timediff = time - $starttime;
2088 $timediff = ($endtime - $starttime) - (time - $endtime);
2090 if ($timediff > 0) {
2091 $retries = int(log($timediff) / log(2));
2093 $retries = 1; # Use the minimum.
2095 return ($retries > 3) ? $retries : 3;
2099 # Pass in two function references.
2100 # This method will be called with the remaining arguments.
2101 # If it dies, retry it with exponential backoff until it succeeds,
2102 # or until the current retry_count is exhausted. After each failure
2103 # that can be retried, the second function will be called with
2104 # the current try count (0-based), next try time, and error message.
2105 my $operation = shift;
2106 my $retry_callback = shift;
2107 my $retries = retry_count();
2108 foreach my $try_count (0..$retries) {
2109 my $next_try = time + (2 ** $try_count);
2110 my $result = eval { $operation->(@_); };
2113 } elsif ($try_count < $retries) {
2114 $retry_callback->($try_count, $next_try, $@);
2115 my $sleep_time = $next_try - time;
2116 sleep($sleep_time) if ($sleep_time > 0);
2119 # Ensure the error message ends in a newline, so Perl doesn't add
2120 # retry_op's line number to it.
2126 # Pass in a /-separated API method name, and arguments for it.
2127 # This function will call that method, retrying as needed until
2128 # the current retry_count is exhausted, with a log on the first failure.
2129 my $method_name = shift;
2130 my $log_api_retry = sub {
2131 my ($try_count, $next_try_at, $errmsg) = @_;
2132 $errmsg =~ s/\s*\bat \Q$0\E line \d+\.?\s*//;
2133 $errmsg =~ s/\s/ /g;
2134 $errmsg =~ s/\s+$//;
2136 if ($next_try_at < time) {
2137 $retry_msg = "Retrying.";
2139 my $next_try_fmt = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($next_try_at);
2140 $retry_msg = "Retrying at $next_try_fmt.";
2142 Log(undef, "API method $method_name failed: $errmsg. $retry_msg");
2145 foreach my $key (split(/\//, $method_name)) {
2146 $method = $method->{$key};
2148 return retry_op(sub { $method->execute(@_); }, $log_api_retry, @_);
2152 # Given a $?, return a human-readable exit code string like "0" or
2153 # "1" or "0 with signal 1" or "1 with signal 11".
2154 my $exitcode = shift;
2155 my $s = $exitcode >> 8;
2156 if ($exitcode & 0x7f) {
2157 $s .= " with signal " . ($exitcode & 0x7f);
2159 if ($exitcode & 0x80) {
2160 $s .= " with core dump";
2165 sub handle_readall {
2166 # Pass in a glob reference to a file handle.
2167 # Read all its contents and return them as a string.
2168 my $fh_glob_ref = shift;
2170 return <$fh_glob_ref>;
2173 sub tar_filename_n {
2175 return sprintf("%s/git.%s.%d.tar", $ENV{CRUNCH_TMP}, $job_id, $n);
2178 sub add_git_archive {
2179 # Pass in a git archive command as a string or list, a la system().
2180 # This method will save its output to be included in the archive sent to the
2184 if (!open(GIT_ARCHIVE, ">", tar_filename_n($git_tar_count))) {
2185 croak("Failed to save git archive: $!");
2187 my $git_pid = open2(">&GIT_ARCHIVE", $git_input, @_);
2189 waitpid($git_pid, 0);
2192 croak("Failed to save git archive: git exited " . exit_status_s($?));
2196 sub combined_git_archive {
2197 # Combine all saved tar archives into a single archive, then return its
2198 # contents in a string. Return undef if no archives have been saved.
2199 if ($git_tar_count < 1) {
2202 my $base_tar_name = tar_filename_n(1);
2203 foreach my $tar_to_append (map { tar_filename_n($_); } (2..$git_tar_count)) {
2204 my $tar_exit = system("tar", "-Af", $base_tar_name, $tar_to_append);
2205 if ($tar_exit != 0) {
2206 croak("Error preparing build archive: tar -A exited " .
2207 exit_status_s($tar_exit));
2210 if (!open(GIT_TAR, "<", $base_tar_name)) {
2211 croak("Could not open build archive: $!");
2213 my $tar_contents = handle_readall(\*GIT_TAR);
2215 return $tar_contents;
2218 sub set_nonblocking {
2220 my $flags = fcntl ($fh, F_GETFL, 0) or croak ($!);
2221 fcntl ($fh, F_SETFL, $flags | O_NONBLOCK) or croak ($!);
2227 # This is crunch-job's internal dispatch script. crunch-job running on the API
2228 # server invokes this script on individual compute nodes, or localhost if we're
2229 # running a job locally. It gets called in two modes:
2231 # * No arguments: Installation mode. Read a tar archive from the DATA
2232 # file handle; it includes the Crunch script's source code, and
2233 # maybe SDKs as well. Those should be installed in the proper
2234 # locations. This runs outside of any Docker container, so don't try to
2235 # introspect Crunch's runtime environment.
2237 # * With arguments: Crunch script run mode. This script should set up the
2238 # environment, then run the command specified in the arguments. This runs
2239 # inside any Docker container.
2242 use File::Path qw( make_path remove_tree );
2243 use POSIX qw(getcwd);
2245 use constant TASK_TEMPFAIL => 111;
2247 # Map SDK subdirectories to the path environments they belong to.
2248 my %SDK_ENVVARS = ("perl/lib" => "PERLLIB", "ruby/lib" => "RUBYLIB");
2250 my $destdir = $ENV{"CRUNCH_SRC"};
2251 my $archive_hash = $ENV{"CRUNCH_GIT_ARCHIVE_HASH"};
2252 my $repo = $ENV{"CRUNCH_SRC_URL"};
2253 my $install_dir = $ENV{"CRUNCH_INSTALL"} || (getcwd() . "/opt");
2254 my $job_work = $ENV{"JOB_WORK"};
2255 my $task_work = $ENV{"TASK_WORK"};
2257 open(STDOUT_ORIG, ">&", STDOUT);
2258 open(STDERR_ORIG, ">&", STDERR);
2260 for my $dir ($destdir, $job_work, $task_work) {
2263 -e $dir or die "Failed to create temporary directory ($dir): $!";
2268 remove_tree($task_work, {keep_root => 1});
2271 ### Crunch script run mode
2273 # We want to do routine logging during task 0 only. This gives the user
2274 # the information they need, but avoids repeating the information for every
2277 if ($ENV{TASK_SEQUENCE} eq "0") {
2280 printf STDERR_ORIG "[Crunch] $msg\n", @_;
2286 my $python_src = "$install_dir/python";
2287 my $venv_dir = "$job_work/.arvados.venv";
2288 my $venv_built = -e "$venv_dir/bin/activate";
2289 if ((!$venv_built) and (-d $python_src) and can_run("virtualenv")) {
2290 shell_or_die(undef, "virtualenv", "--quiet", "--system-site-packages",
2291 "--python=python2.7", $venv_dir);
2292 shell_or_die(TASK_TEMPFAIL, "$venv_dir/bin/pip", "--quiet", "install", "-I", $python_src);
2294 $Log->("Built Python SDK virtualenv");
2297 my @pysdk_version_cmd = ("python", "-c",
2298 "from pkg_resources import get_distribution as get; print get('arvados-python-client').version");
2300 $Log->("Running in Python SDK virtualenv");
2301 @pysdk_version_cmd = ();
2302 my $orig_argv = join(" ", map { quotemeta($_); } @ARGV);
2303 @ARGV = ("/bin/sh", "-ec",
2304 ". \Q$venv_dir/bin/activate\E; exec $orig_argv");
2305 } elsif (-d $python_src) {
2306 $Log->("Warning: virtualenv not found inside Docker container default " .
2307 "\$PATH. Can't install Python SDK.");
2310 if (@pysdk_version_cmd) {
2311 open(my $pysdk_version_pipe, "-|", @pysdk_version_cmd);
2312 my $pysdk_version = <$pysdk_version_pipe>;
2313 close($pysdk_version_pipe);
2315 chomp($pysdk_version);
2316 $Log->("Using Arvados SDK version $pysdk_version");
2318 # A lot could've gone wrong here, but pretty much all of it means that
2319 # Python won't be able to load the Arvados SDK.
2320 $Log->("Warning: Arvados SDK not found");
2324 while (my ($sdk_dir, $sdk_envkey) = each(%SDK_ENVVARS)) {
2325 my $sdk_path = "$install_dir/$sdk_dir";
2327 if ($ENV{$sdk_envkey}) {
2328 $ENV{$sdk_envkey} = "$sdk_path:" . $ENV{$sdk_envkey};
2330 $ENV{$sdk_envkey} = $sdk_path;
2332 $Log->("Arvados SDK added to %s", $sdk_envkey);
2337 die "Cannot exec `@ARGV`: $!";
2340 ### Installation mode
2341 open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
2343 if (readlink ("$destdir.archive_hash") eq $archive_hash && -d $destdir) {
2344 # This exact git archive (source + arvados sdk) is already installed
2345 # here, so there's no need to reinstall it.
2347 # We must consume our DATA section, though: otherwise the process
2348 # feeding it to us will get SIGPIPE.
2350 while (read(DATA, $buf, 65536)) { }
2355 unlink "$destdir.archive_hash";
2359 # Ignore SIGPIPE: we check retval of close() instead. See perlipc(1).
2360 local $SIG{PIPE} = "IGNORE";
2361 warn "Extracting archive: $archive_hash\n";
2362 # --ignore-zeros is necessary sometimes: depending on how much NUL
2363 # padding tar -A put on our combined archive (which in turn depends
2364 # on the length of the component archives) tar without
2365 # --ignore-zeros will exit before consuming stdin and cause close()
2366 # to fail on the resulting SIGPIPE.
2367 if (!open(TARX, "|-", "tar", "--ignore-zeros", "-xC", $destdir)) {
2368 die "Error launching 'tar -xC $destdir': $!";
2370 # If we send too much data to tar in one write (> 4-5 MiB), it stops, and we
2371 # get SIGPIPE. We must feed it data incrementally.
2373 while (read(DATA, $tar_input, 65536)) {
2374 print TARX $tar_input;
2377 die "'tar -xC $destdir' exited $?: $!";
2383 my $sdk_root = "$destdir/.arvados.sdk/sdk";
2385 foreach my $sdk_lang (("python",
2386 map { (split /\//, $_, 2)[0]; } keys(%SDK_ENVVARS))) {
2387 if (-d "$sdk_root/$sdk_lang") {
2388 if (!rename("$sdk_root/$sdk_lang", "$install_dir/$sdk_lang")) {
2389 die "Failed to install $sdk_lang SDK: $!";
2395 my $python_dir = "$install_dir/python";
2396 if ((-d $python_dir) and can_run("python2.7")) {
2397 open(my $egg_info_pipe, "-|",
2398 "python2.7 \Q$python_dir/setup.py\E egg_info 2>&1 >/dev/null");
2399 my @egg_info_errors = <$egg_info_pipe>;
2400 close($egg_info_pipe);
2403 if (@egg_info_errors and (($egg_info_errors[-1] =~ /\bgit\b/) or ($egg_info_errors[-1] =~ /\[Errno 2\]/))) {
2404 # egg_info apparently failed because it couldn't ask git for a build tag.
2405 # Specify no build tag.
2406 open(my $pysdk_cfg, ">>", "$python_dir/setup.cfg");
2407 print $pysdk_cfg "\n[egg_info]\ntag_build =\n";
2410 my $egg_info_exit = $? >> 8;
2411 foreach my $errline (@egg_info_errors) {
2414 warn "python setup.py egg_info failed: exit $egg_info_exit";
2415 exit ($egg_info_exit || 1);
2420 # Hide messages from the install script (unless it fails: shell_or_die
2421 # will show $destdir.log in that case).
2422 open(STDOUT, ">>", "$destdir.log");
2423 open(STDERR, ">&", STDOUT);
2425 if (-e "$destdir/crunch_scripts/install") {
2426 shell_or_die (undef, "$destdir/crunch_scripts/install", $install_dir);
2427 } elsif (!-e "./install.sh" && -e "./tests/autotests.sh") {
2429 shell_or_die (undef, "./tests/autotests.sh", $install_dir);
2430 } elsif (-e "./install.sh") {
2431 shell_or_die (undef, "./install.sh", $install_dir);
2434 if ($archive_hash) {
2435 unlink "$destdir.archive_hash.new";
2436 symlink ($archive_hash, "$destdir.archive_hash.new") or die "$destdir.archive_hash.new: $!";
2437 rename ("$destdir.archive_hash.new", "$destdir.archive_hash") or die "$destdir.archive_hash: $!";
2443 my $command_name = shift;
2444 open(my $which, "-|", "which", $command_name);
2445 while (<$which>) { }
2452 my $exitcode = shift;
2454 if ($ENV{"DEBUG"}) {
2455 print STDERR "@_\n";
2457 if (system (@_) != 0) {
2460 my $exitstatus = sprintf("exit %d signal %d", $code >> 8, $code & 0x7f);
2461 open STDERR, ">&STDERR_ORIG";
2462 system ("cat $destdir.log >&2");
2463 warn "@_ failed ($err): $exitstatus";
2464 if (defined($exitcode)) {
2468 exit (($code >> 8) || 1);