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 $cgroup_root = "/sys/fs/cgroup";
130 my $docker_bin = "docker.io";
131 my $docker_run_args = "";
132 GetOptions('force-unlock' => \$force_unlock,
133 'git-dir=s' => \$git_dir,
134 'job=s' => \$jobspec,
135 'job-api-token=s' => \$job_api_token,
136 'no-clear-tmp' => \$no_clear_tmp,
137 'resume-stash=s' => \$resume_stash,
138 'cgroup-root=s' => \$cgroup_root,
139 'docker-bin=s' => \$docker_bin,
140 'docker-run-args=s' => \$docker_run_args,
143 if (defined $job_api_token) {
144 $ENV{ARVADOS_API_TOKEN} = $job_api_token;
147 my $have_slurm = exists $ENV{SLURM_JOB_ID} && exists $ENV{SLURM_NODELIST};
152 $main::ENV{CRUNCH_DEBUG} = 1;
156 $main::ENV{CRUNCH_DEBUG} = 0;
159 my $arv = Arvados->new('apiVersion' => 'v1');
168 if ($jobspec =~ /^[-a-z\d]+$/)
170 # $jobspec is an Arvados UUID, not a JSON job specification
171 $Job = api_call("jobs/get", uuid => $jobspec);
176 $local_job = JSON::decode_json($jobspec);
180 # Make sure our workers (our slurm nodes, localhost, or whatever) are
181 # at least able to run basic commands: they aren't down or severely
184 if (($Job || $local_job)->{docker_image_locator}) {
185 $cmd = [$docker_bin, 'ps', '-q'];
187 Log(undef, "Sanity check is `@$cmd`");
188 srun(["srun", "--nodes=\Q$ENV{SLURM_NNODES}\E", "--ntasks-per-node=1"],
192 Log(undef, "Sanity check failed: ".exit_status_s($?));
195 Log(undef, "Sanity check OK");
198 my $User = api_call("users/current");
201 if (!$force_unlock) {
202 # Claim this job, and make sure nobody else does
203 eval { api_call("jobs/lock", uuid => $Job->{uuid}); };
205 Log(undef, "Error while locking job, exiting ".EX_TEMPFAIL);
214 map { croak ("No $_ specified") unless $local_job->{$_} }
215 qw(script script_version script_parameters);
218 $local_job->{'is_locked_by_uuid'} = $User->{'uuid'};
219 $local_job->{'started_at'} = gmtime;
220 $local_job->{'state'} = 'Running';
222 $Job = api_call("jobs/create", job => $local_job);
224 $job_id = $Job->{'uuid'};
226 my $keep_logfile = $job_id . '.log.txt';
227 log_writer_start($keep_logfile);
229 $Job->{'runtime_constraints'} ||= {};
230 $Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0;
231 my $max_ncpus = $Job->{'runtime_constraints'}->{'max_tasks_per_node'};
233 my $gem_versions = `gem list --quiet arvados-cli 2>/dev/null`;
235 $gem_versions =~ s/^arvados-cli \(/ with arvados-cli Gem version(s) /;
236 chomp($gem_versions);
237 chop($gem_versions); # Closing parentheses
242 "running from " . ((-e $0) ? realpath($0) : "stdin") . $gem_versions);
244 Log (undef, "check slurm allocation");
247 # Should use $ENV{SLURM_TASKS_PER_NODE} instead of sinfo? (eg. "4(x3),2,4(x2)")
251 my $localcpus = 0 + `grep -cw ^processor /proc/cpuinfo` || 1;
252 push @sinfo, "$localcpus localhost";
254 if (exists $ENV{SLURM_NODELIST})
256 push @sinfo, `sinfo -h --format='%c %N' --nodes=\Q$ENV{SLURM_NODELIST}\E`;
260 my ($ncpus, $slurm_nodelist) = split;
261 $ncpus = $max_ncpus if $max_ncpus && $ncpus > $max_ncpus;
264 while ($slurm_nodelist =~ s/^([^\[,]+?(\[.*?\])?)(,|$)//)
267 if ($nodelist =~ /\[((\d+)(-(\d+))?(,(\d+)(-(\d+))?)*)\]/)
270 foreach (split (",", $ranges))
283 push @nodelist, map {
285 $n =~ s/\[[-,\d]+\]/$_/;
292 push @nodelist, $nodelist;
295 foreach my $nodename (@nodelist)
297 Log (undef, "node $nodename - $ncpus slots");
298 my $node = { name => $nodename,
300 # The number of consecutive times a task has been dispatched
301 # to this node and failed.
303 # The number of consecutive times that SLURM has reported
304 # a node failure since the last successful task.
306 # Don't dispatch work to this node until this time
307 # (in seconds since the epoch) has passed.
309 foreach my $cpu (1..$ncpus)
311 push @slot, { node => $node,
315 push @node, @nodelist;
320 # Ensure that we get one jobstep running on each allocated node before
321 # we start overloading nodes with concurrent steps
323 @slot = sort { $a->{cpu} <=> $b->{cpu} } @slot;
326 $Job->update_attributes(
327 'tasks_summary' => { 'failed' => 0,
332 Log (undef, "start");
333 $SIG{'INT'} = sub { $main::please_freeze = 1; };
334 $SIG{'QUIT'} = sub { $main::please_freeze = 1; };
335 $SIG{'TERM'} = \&croak;
336 $SIG{'TSTP'} = sub { $main::please_freeze = 1; };
337 $SIG{'ALRM'} = sub { $main::please_info = 1; };
338 $SIG{'CONT'} = sub { $main::please_continue = 1; };
339 $SIG{'HUP'} = sub { $main::please_refresh = 1; };
341 $main::please_freeze = 0;
342 $main::please_info = 0;
343 $main::please_continue = 0;
344 $main::please_refresh = 0;
345 my $jobsteps_must_output_keys = 0; # becomes 1 when any task outputs a key
347 grep { $ENV{$1} = $2 if /^(NOCACHE.*?)=(.*)/ } split ("\n", $$Job{knobs});
348 $ENV{"CRUNCH_JOB_UUID"} = $job_id;
349 $ENV{"JOB_UUID"} = $job_id;
352 my @jobstep_todo = ();
353 my @jobstep_done = ();
354 my @jobstep_tomerge = ();
355 my $jobstep_tomerge_level = 0;
356 my $squeue_checked = 0;
357 my $latest_refresh = scalar time;
361 if (defined $Job->{thawedfromkey})
363 thaw ($Job->{thawedfromkey});
367 my $first_task = api_call("job_tasks/create", job_task => {
368 'job_uuid' => $Job->{'uuid'},
373 push @jobstep, { 'level' => 0,
375 'arvados_task' => $first_task,
377 push @jobstep_todo, 0;
383 must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
386 my $build_script = handle_readall(\*DATA);
387 my $nodelist = join(",", @node);
388 my $git_tar_count = 0;
390 if (!defined $no_clear_tmp) {
391 # Clean out crunch_tmp/work, crunch_tmp/opt, crunch_tmp/src*
392 Log (undef, "Clean work dirs");
394 my $cleanpid = fork();
397 # Find FUSE mounts under $CRUNCH_TMP and unmount them.
398 # Then clean up work directories.
399 # TODO: When #5036 is done and widely deployed, we can limit mount's
400 # -t option to simply fuse.keep.
401 srun (["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
402 ['bash', '-ec', '-o', 'pipefail', 'mount -t fuse,fuse.keep | awk "(index(\$3, \"$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']);
407 last if $cleanpid == waitpid (-1, WNOHANG);
408 freeze_if_want_freeze ($cleanpid);
409 select (undef, undef, undef, 0.1);
412 Log(undef, "Clean work dirs: exit ".exit_status_s($?));
413 exit(EX_RETRY_UNLOCKED);
417 # If this job requires a Docker image, install that.
418 my ($docker_locator, $docker_stream, $docker_hash, $docker_limitmem, $dockeruserarg);
419 if ($docker_locator = $Job->{docker_image_locator}) {
420 Log (undef, "Install docker image $docker_locator");
421 ($docker_stream, $docker_hash) = find_docker_image($docker_locator);
424 croak("No Docker image hash found from locator $docker_locator");
426 Log (undef, "docker image hash is $docker_hash");
427 $docker_stream =~ s/^\.//;
428 my $docker_install_script = qq{
429 if ! $docker_bin images -q --no-trunc --all | grep -qxF \Q$docker_hash\E; then
430 arv-get \Q$docker_locator$docker_stream/$docker_hash.tar\E | $docker_bin load
433 my $docker_pid = fork();
434 if ($docker_pid == 0)
436 srun (["srun", "--nodelist=" . join(',', @node)],
437 ["/bin/bash", "-o", "pipefail", "-ec", $docker_install_script]);
442 last if $docker_pid == waitpid (-1, WNOHANG);
443 freeze_if_want_freeze ($docker_pid);
444 select (undef, undef, undef, 0.1);
448 Log(undef, "Installing Docker image from $docker_locator exited " . exit_status_s($?));
449 exit(EX_RETRY_UNLOCKED);
452 # Determine whether this version of Docker supports memory+swap limits.
453 srun(["srun", "--nodelist=" . $node[0]],
454 ["/bin/sh", "-ec", "$docker_bin run --help | grep -qe --memory-swap="],
456 $docker_limitmem = ($? == 0);
458 # Find a non-root Docker user to use.
459 # Tries the default user for the container, then 'crunch', then 'nobody',
460 # testing for whether the actual user id is non-zero. This defends against
461 # mistakes but not malice, but we intend to harden the security in the future
462 # so we don't want anyone getting used to their jobs running as root in their
464 my @tryusers = ("", "crunch", "nobody");
465 foreach my $try_user (@tryusers) {
467 if ($try_user eq "") {
468 Log(undef, "Checking if container default user is not UID 0");
471 Log(undef, "Checking if user '$try_user' is not UID 0");
472 $try_user_arg = "--user=$try_user";
474 srun(["srun", "--nodelist=" . $node[0]],
476 "a=`$docker_bin run $docker_run_args $try_user_arg $docker_hash id --user` && " .
480 $dockeruserarg = $try_user_arg;
481 if ($try_user eq "") {
482 Log(undef, "Container will run with default user");
484 Log(undef, "Container will run with $dockeruserarg");
490 if (!defined $dockeruserarg) {
491 croak("Could not find a user in container that is not UID 0 (tried default user, @tryusers) or there was a problem running 'id' in the container.");
494 if ($Job->{arvados_sdk_version}) {
495 # The job also specifies an Arvados SDK version. Add the SDKs to the
496 # tar file for the build script to install.
497 Log(undef, sprintf("Packing Arvados SDK version %s for installation",
498 $Job->{arvados_sdk_version}));
499 add_git_archive("git", "--git-dir=$git_dir", "archive",
500 "--prefix=.arvados.sdk/",
501 $Job->{arvados_sdk_version}, "sdk");
505 if (!defined $git_dir && $Job->{'script_version'} =~ m{^/}) {
506 # If script_version looks like an absolute path, *and* the --git-dir
507 # argument was not given -- which implies we were not invoked by
508 # crunch-dispatch -- we will use the given path as a working
509 # directory instead of resolving script_version to a git commit (or
510 # doing anything else with git).
511 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{'script_version'};
512 $ENV{"CRUNCH_SRC"} = $Job->{'script_version'};
515 # Resolve the given script_version to a git commit sha1. Also, if
516 # the repository is remote, clone it into our local filesystem: this
517 # ensures "git archive" will work, and is necessary to reliably
518 # resolve a symbolic script_version like "master^".
519 $ENV{"CRUNCH_SRC"} = "$ENV{CRUNCH_TMP}/src";
521 Log (undef, "Looking for version ".$Job->{script_version}." from repository ".$Job->{repository});
523 $ENV{"CRUNCH_SRC_COMMIT"} = $Job->{script_version};
525 # If we're running under crunch-dispatch, it will have already
526 # pulled the appropriate source tree into its own repository, and
527 # given us that repo's path as $git_dir.
529 # If we're running a "local" job, we might have to fetch content
530 # from a remote repository.
532 # (Currently crunch-dispatch gives a local path with --git-dir, but
533 # we might as well accept URLs there too in case it changes its
535 my $repo = $git_dir || $Job->{'repository'};
537 # Repository can be remote or local. If remote, we'll need to fetch it
538 # to a local dir before doing `git log` et al.
541 if ($repo =~ m{://|^[^/]*:}) {
542 # $repo is a git url we can clone, like git:// or https:// or
543 # file:/// or [user@]host:repo.git. Note "user/name@host:foo" is
544 # not recognized here because distinguishing that from a local
545 # path is too fragile. If you really need something strange here,
546 # use the ssh:// form.
547 $repo_location = 'remote';
548 } elsif ($repo =~ m{^\.*/}) {
549 # $repo is a local path to a git index. We'll also resolve ../foo
550 # to ../foo/.git if the latter is a directory. To help
551 # disambiguate local paths from named hosted repositories, this
552 # form must be given as ./ or ../ if it's a relative path.
553 if (-d "$repo/.git") {
554 $repo = "$repo/.git";
556 $repo_location = 'local';
558 # $repo is none of the above. It must be the name of a hosted
560 my $arv_repo_list = api_call("repositories/list",
561 'filters' => [['name','=',$repo]]);
562 my @repos_found = @{$arv_repo_list->{'items'}};
563 my $n_found = $arv_repo_list->{'serverResponse'}->{'items_available'};
565 Log(undef, "Repository '$repo' -> "
566 . join(", ", map { $_->{'uuid'} } @repos_found));
569 croak("Error: Found $n_found repositories with name '$repo'.");
571 $repo = $repos_found[0]->{'fetch_url'};
572 $repo_location = 'remote';
574 Log(undef, "Using $repo_location repository '$repo'");
575 $ENV{"CRUNCH_SRC_URL"} = $repo;
577 # Resolve given script_version (we'll call that $treeish here) to a
578 # commit sha1 ($commit).
579 my $treeish = $Job->{'script_version'};
581 if ($repo_location eq 'remote') {
582 # We minimize excess object-fetching by re-using the same bare
583 # repository in CRUNCH_TMP/.git for multiple crunch-jobs -- we
584 # just keep adding remotes to it as needed.
585 my $local_repo = $ENV{'CRUNCH_TMP'}."/.git";
586 my $gitcmd = "git --git-dir=\Q$local_repo\E";
588 # Set up our local repo for caching remote objects, making
590 if (!-d $local_repo) {
591 make_path($local_repo) or croak("Error: could not create $local_repo");
593 # This works (exits 0 and doesn't delete fetched objects) even
594 # if $local_repo is already initialized:
595 `$gitcmd init --bare`;
597 croak("Error: $gitcmd init --bare exited ".exit_status_s($?));
600 # If $treeish looks like a hash (or abbrev hash) we look it up in
601 # our local cache first, since that's cheaper. (We don't want to
602 # do that with tags/branches though -- those change over time, so
603 # they should always be resolved by the remote repo.)
604 if ($treeish =~ /^[0-9a-f]{7,40}$/s) {
605 # Hide stderr because it's normal for this to fail:
606 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E 2>/dev/null`;
608 # Careful not to resolve a branch named abcdeff to commit 1234567:
609 $sha1 =~ /^$treeish/ &&
610 $sha1 =~ /^([0-9a-f]{40})$/s) {
612 Log(undef, "Commit $commit already present in $local_repo");
616 if (!defined $commit) {
617 # If $treeish isn't just a hash or abbrev hash, or isn't here
618 # yet, we need to fetch the remote to resolve it correctly.
620 # First, remove all local heads. This prevents a name that does
621 # not exist on the remote from resolving to (or colliding with)
622 # a previously fetched branch or tag (possibly from a different
624 remove_tree("$local_repo/refs/heads", {keep_root => 1});
626 Log(undef, "Fetching objects from $repo to $local_repo");
627 `$gitcmd fetch --no-progress --tags ''\Q$repo\E \Q+refs/heads/*:refs/heads/*\E`;
629 croak("Error: `$gitcmd fetch` exited ".exit_status_s($?));
633 # Now that the data is all here, we will use our local repo for
634 # the rest of our git activities.
638 my $gitcmd = "git --git-dir=\Q$repo\E";
639 my $sha1 = `$gitcmd rev-list -n1 ''\Q$treeish\E`;
640 unless ($? == 0 && $sha1 =~ /^([0-9a-f]{40})$/) {
641 croak("`$gitcmd rev-list` exited "
643 .", '$treeish' not found, giving up");
646 Log(undef, "Version $treeish is commit $commit");
648 if ($commit ne $Job->{'script_version'}) {
649 # Record the real commit id in the database, frozentokey, logs,
650 # etc. -- instead of an abbreviation or a branch name which can
651 # become ambiguous or point to a different commit in the future.
652 if (!$Job->update_attributes('script_version' => $commit)) {
653 croak("Error: failed to update job's script_version attribute");
657 $ENV{"CRUNCH_SRC_COMMIT"} = $commit;
658 add_git_archive("$gitcmd archive ''\Q$commit\E");
661 my $git_archive = combined_git_archive();
662 if (!defined $git_archive) {
663 Log(undef, "Skip install phase (no git archive)");
665 Log(undef, "Warning: This probably means workers have no source tree!");
670 my $install_script_tries_left = 3;
671 for (my $attempts = 0; $attempts < 3; $attempts++) {
672 Log(undef, "Run install script on all workers");
674 my @srunargs = ("srun",
675 "--nodelist=$nodelist",
676 "-D", $ENV{'TMPDIR'}, "--job-name=$job_id");
677 my @execargs = ("sh", "-c",
678 "mkdir -p $ENV{CRUNCH_INSTALL} && cd $ENV{CRUNCH_TMP} && perl -");
680 $ENV{"CRUNCH_GIT_ARCHIVE_HASH"} = md5_hex($git_archive);
681 my ($install_stderr_r, $install_stderr_w);
682 pipe $install_stderr_r, $install_stderr_w or croak("pipe() failed: $!");
683 set_nonblocking($install_stderr_r);
684 my $installpid = fork();
685 if ($installpid == 0)
687 close($install_stderr_r);
688 fcntl($install_stderr_w, F_SETFL, 0) or croak($!); # no close-on-exec
689 open(STDOUT, ">&", $install_stderr_w);
690 open(STDERR, ">&", $install_stderr_w);
691 srun (\@srunargs, \@execargs, {}, $build_script . $git_archive);
694 close($install_stderr_w);
695 # Tell freeze_if_want_freeze how to kill the child, otherwise the
696 # "waitpid(installpid)" loop won't get interrupted by a freeze:
697 $proc{$installpid} = {};
699 # Track whether anything appears on stderr other than slurm errors
700 # ("srun: ...") and the "starting: ..." message printed by the
701 # srun subroutine itself:
702 my $stderr_anything_from_script = 0;
703 my $match_our_own_errors = '^(srun: error: |starting: \[)';
704 while ($installpid != waitpid(-1, WNOHANG)) {
705 freeze_if_want_freeze ($installpid);
706 # Wait up to 0.1 seconds for something to appear on stderr, then
707 # do a non-blocking read.
708 my $bits = fhbits($install_stderr_r);
709 select ($bits, undef, $bits, 0.1);
710 if (0 < sysread ($install_stderr_r, $stderr_buf, 8192, length($stderr_buf)))
712 while ($stderr_buf =~ /^(.*?)\n/) {
714 substr $stderr_buf, 0, 1+length($line), "";
715 Log(undef, "stderr $line");
716 if ($line !~ /$match_our_own_errors/) {
717 $stderr_anything_from_script = 1;
722 delete $proc{$installpid};
723 $install_exited = $?;
724 close($install_stderr_r);
725 if (length($stderr_buf) > 0) {
726 if ($stderr_buf !~ /$match_our_own_errors/) {
727 $stderr_anything_from_script = 1;
729 Log(undef, "stderr $stderr_buf")
732 Log (undef, "Install script exited ".exit_status_s($install_exited));
733 last if $install_exited == 0 || $main::please_freeze;
734 # If the install script fails but doesn't print an error message,
735 # the next thing anyone is likely to do is just run it again in
736 # case it was a transient problem like "slurm communication fails
737 # because the network isn't reliable enough". So we'll just do
738 # that ourselves (up to 3 attempts in total). OTOH, if there is an
739 # error message, the problem is more likely to have a real fix and
740 # we should fail the job so the fixing process can start, instead
741 # of doing 2 more attempts.
742 last if $stderr_anything_from_script;
745 foreach my $tar_filename (map { tar_filename_n($_); } (1..$git_tar_count)) {
746 unlink($tar_filename);
749 if ($install_exited != 0) {
754 foreach (qw (script script_version script_parameters runtime_constraints))
758 (ref($Job->{$_}) ? JSON::encode_json($Job->{$_}) : $Job->{$_}));
760 foreach (split (/\n/, $Job->{knobs}))
762 Log (undef, "knob " . $_);
767 $main::success = undef;
773 my $thisround_succeeded = 0;
774 my $thisround_failed = 0;
775 my $thisround_failed_multiple = 0;
776 my $working_slot_count = scalar(@slot);
778 @jobstep_todo = sort { $jobstep[$a]->{level} <=> $jobstep[$b]->{level}
779 or $a <=> $b } @jobstep_todo;
780 my $level = $jobstep[$jobstep_todo[0]]->{level};
782 my $initial_tasks_this_level = 0;
783 foreach my $id (@jobstep_todo) {
784 $initial_tasks_this_level++ if ($jobstep[$id]->{level} == $level);
787 # If the number of tasks scheduled at this level #T is smaller than the number
788 # of slots available #S, only use the first #T slots, or the first slot on
789 # each node, whichever number is greater.
791 # When we dispatch tasks later, we'll allocate whole-node resources like RAM
792 # based on these numbers. Using fewer slots makes more resources available
793 # to each individual task, which should normally be a better strategy when
794 # there are fewer of them running with less parallelism.
796 # Note that this calculation is not redone if the initial tasks at
797 # this level queue more tasks at the same level. This may harm
798 # overall task throughput for that level.
800 if ($initial_tasks_this_level < @node) {
801 @freeslot = (0..$#node);
802 } elsif ($initial_tasks_this_level < @slot) {
803 @freeslot = (0..$initial_tasks_this_level - 1);
805 @freeslot = (0..$#slot);
807 my $round_num_freeslots = scalar(@freeslot);
809 my %round_max_slots = ();
810 for (my $ii = $#freeslot; $ii >= 0; $ii--) {
811 my $this_slot = $slot[$freeslot[$ii]];
812 my $node_name = $this_slot->{node}->{name};
813 $round_max_slots{$node_name} ||= $this_slot->{cpu};
814 last if (scalar(keys(%round_max_slots)) >= @node);
817 Log(undef, "start level $level with $round_num_freeslots slots");
820 my $progress_is_dirty = 1;
821 my $progress_stats_updated = 0;
823 update_progress_stats();
827 for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
829 # Don't create new tasks if we already know the job's final result.
830 last if defined($main::success);
832 my $id = $jobstep_todo[$todo_ptr];
833 my $Jobstep = $jobstep[$id];
834 if ($Jobstep->{level} != $level)
839 pipe $reader{$id}, "writer" or croak("pipe() failed: $!");
840 set_nonblocking($reader{$id});
842 my $childslot = $freeslot[0];
843 my $childnode = $slot[$childslot]->{node};
844 my $childslotname = join (".",
845 $slot[$childslot]->{node}->{name},
846 $slot[$childslot]->{cpu});
848 my $childpid = fork();
851 $SIG{'INT'} = 'DEFAULT';
852 $SIG{'QUIT'} = 'DEFAULT';
853 $SIG{'TERM'} = 'DEFAULT';
855 foreach (values (%reader))
859 fcntl ("writer", F_SETFL, 0) or croak ($!); # no close-on-exec
860 open(STDOUT,">&writer");
861 open(STDERR,">&writer");
866 delete $ENV{"GNUPGHOME"};
867 $ENV{"TASK_UUID"} = $Jobstep->{'arvados_task'}->{'uuid'};
868 $ENV{"TASK_QSEQUENCE"} = $id;
869 $ENV{"TASK_SEQUENCE"} = $level;
870 $ENV{"JOB_SCRIPT"} = $Job->{script};
871 while (my ($param, $value) = each %{$Job->{script_parameters}}) {
872 $param =~ tr/a-z/A-Z/;
873 $ENV{"JOB_PARAMETER_$param"} = $value;
875 $ENV{"TASK_SLOT_NODE"} = $slot[$childslot]->{node}->{name};
876 $ENV{"TASK_SLOT_NUMBER"} = $slot[$childslot]->{cpu};
877 $ENV{"TASK_WORK"} = $ENV{"CRUNCH_TMP"}."/task/$childslotname";
878 $ENV{"HOME"} = $ENV{"TASK_WORK"};
879 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
880 $ENV{"CRUNCH_NODE_SLOTS"} = $round_max_slots{$ENV{TASK_SLOT_NODE}};
881 $ENV{"PATH"} = $ENV{"CRUNCH_INSTALL"} . "/bin:" . $ENV{"PATH"};
883 my $keep_mnt = $ENV{"TASK_WORK"}.".keep";
889 "--nodelist=".$childnode->{name},
890 qw(-n1 -c1 -N1 -D), $ENV{'TMPDIR'},
891 "--job-name=$job_id.$id.$$",
894 my $stdbuf = " stdbuf --output=0 --error=0 ";
896 my $arv_file_cache = "";
897 if (defined($Job->{'runtime_constraints'}->{'keep_cache_mb_per_task'})) {
898 $arv_file_cache = "--file-cache=" . ($Job->{'runtime_constraints'}->{'keep_cache_mb_per_task'} * 1024 * 1024);
902 "if [ -e \Q$ENV{TASK_WORK}\E ]; then rm -rf \Q$ENV{TASK_WORK}\E; fi; "
903 ."mkdir -p \Q$ENV{CRUNCH_TMP}\E \Q$ENV{JOB_WORK}\E \Q$ENV{TASK_WORK}\E \Q$keep_mnt\E "
904 ."&& cd \Q$ENV{CRUNCH_TMP}\E "
905 # These environment variables get used explicitly later in
906 # $command. No tool is expected to read these values directly.
907 .q{&& MEM=$(awk '($1 == "MemTotal:"){print $2}' </proc/meminfo) }
908 .q{&& SWAP=$(awk '($1 == "SwapTotal:"){print $2}' </proc/meminfo) }
909 ."&& MEMLIMIT=\$(( (\$MEM * 95) / ($ENV{CRUNCH_NODE_SLOTS} * 100) )) "
910 ."&& let SWAPLIMIT=\$MEMLIMIT+\$SWAP ";
912 $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 ";
913 $ENV{TASK_KEEPMOUNT} = "$keep_mnt/by_pdh";
914 $ENV{TASK_KEEPMOUNT_TMP} = "$keep_mnt/tmp";
918 my $containername = "$Jobstep->{arvados_task}->{uuid}-$Jobstep->{failures}";
919 my $cidfile = "$ENV{CRUNCH_TMP}/$containername.cid";
920 $command .= "crunchstat -cgroup-root=\Q$cgroup_root\E -cgroup-parent=docker -cgroup-cid=$cidfile -poll=10000 ";
921 $command .= "$docker_bin run $docker_run_args --name=$containername --attach=stdout --attach=stderr --attach=stdin -i \Q$dockeruserarg\E --cidfile=$cidfile --sig-proxy ";
922 # We only set memory limits if Docker lets us limit both memory and swap.
923 # Memory limits alone have been supported longer, but subprocesses tend
924 # to get SIGKILL if they exceed that without any swap limit set.
925 # See #5642 for additional background.
926 if ($docker_limitmem) {
927 $command .= "--memory=\${MEMLIMIT}k --memory-swap=\${SWAPLIMIT}k ";
930 # The source tree and $destdir directory (which we have
931 # installed on the worker host) are available in the container,
932 # under the same path.
933 $command .= "--volume=\Q$ENV{CRUNCH_SRC}:$ENV{CRUNCH_SRC}:ro\E ";
934 $command .= "--volume=\Q$ENV{CRUNCH_INSTALL}:$ENV{CRUNCH_INSTALL}:ro\E ";
936 # Currently, we make the "by_pdh" directory in arv-mount's mount
937 # point appear at /keep inside the container (instead of using
938 # the same path as the host like we do with CRUNCH_SRC and
939 # CRUNCH_INSTALL). However, crunch scripts and utilities must
940 # not rely on this. They must use $TASK_KEEPMOUNT.
941 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT}:/keep:ro\E ";
942 $ENV{TASK_KEEPMOUNT} = "/keep";
944 # Ditto TASK_KEEPMOUNT_TMP, as /keep_tmp.
945 $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT_TMP}:/keep_tmp\E ";
946 $ENV{TASK_KEEPMOUNT_TMP} = "/keep_tmp";
948 # TASK_WORK is almost exactly like a docker data volume: it
949 # starts out empty, is writable, and persists until no
950 # containers use it any more. We don't use --volumes-from to
951 # share it with other containers: it is only accessible to this
952 # task, and it goes away when this task stops.
954 # However, a docker data volume is writable only by root unless
955 # the mount point already happens to exist in the container with
956 # different permissions. Therefore, we [1] assume /tmp already
957 # exists in the image and is writable by the crunch user; [2]
958 # avoid putting TASK_WORK inside CRUNCH_TMP (which won't be
959 # writable if they are created by docker while setting up the
960 # other --volumes); and [3] create $TASK_WORK inside the
961 # container using $build_script.
962 $command .= "--volume=/tmp ";
963 $ENV{"TASK_WORK"} = "/tmp/crunch-job-task-work/$childslotname";
964 $ENV{"HOME"} = $ENV{"TASK_WORK"};
965 $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated
967 # TODO: Share a single JOB_WORK volume across all task
968 # containers on a given worker node, and delete it when the job
969 # ends (and, in case that doesn't work, when the next job
972 # For now, use the same approach as TASK_WORK above.
973 $ENV{"JOB_WORK"} = "/tmp/crunch-job-work";
975 while (my ($env_key, $env_val) = each %ENV)
977 if ($env_key =~ /^(ARVADOS|CRUNCH|JOB|TASK)_/) {
978 $command .= "--env=\Q$env_key=$env_val\E ";
981 $command .= "--env=\QHOME=$ENV{HOME}\E ";
982 $command .= "\Q$docker_hash\E ";
984 if ($Job->{arvados_sdk_version}) {
986 $command .= "perl - \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E";
988 $command .= "/bin/sh -c \'python -c " .
989 '"from pkg_resources import get_distribution as get; print \"Using Arvados SDK version\", get(\"arvados-python-client\").version"' .
990 ">&2 2>/dev/null; " .
991 "mkdir -p \"$ENV{JOB_WORK}\" \"$ENV{TASK_WORK}\" && " .
992 "if which stdbuf >/dev/null ; then " .
993 " exec $stdbuf \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E ;" .
995 " exec \Q$ENV{CRUNCH_SRC}/crunch_scripts/$Job->{script}\E ;" .
1000 $command .= "crunchstat -cgroup-root=\Q$cgroup_root\E -poll=10000 ";
1001 $command .= $stdbuf;
1002 $command .= "perl - $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"};
1005 my @execargs = ('bash', '-c', $command);
1006 srun (\@srunargs, \@execargs, undef, $build_script);
1007 # exec() failed, we assume nothing happened.
1008 die "srun() failed on build script\n";
1011 if (!defined $childpid)
1014 delete $reader{$id};
1018 $proc{$childpid} = { jobstep => $id,
1021 jobstepname => "$job_id.$id.$childpid",
1023 croak ("assert failed: \$slot[$childslot]->{'pid'} exists") if exists $slot[$childslot]->{pid};
1024 $slot[$childslot]->{pid} = $childpid;
1026 Log ($id, "job_task ".$Jobstep->{'arvados_task'}->{'uuid'});
1027 Log ($id, "child $childpid started on $childslotname");
1028 $Jobstep->{starttime} = time;
1029 $Jobstep->{node} = $childnode->{name};
1030 $Jobstep->{slotindex} = $childslot;
1031 delete $Jobstep->{stderr};
1032 delete $Jobstep->{finishtime};
1033 delete $Jobstep->{tempfail};
1035 $Jobstep->{'arvados_task'}->{started_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{starttime});
1036 $Jobstep->{'arvados_task'}->save;
1038 splice @jobstep_todo, $todo_ptr, 1;
1041 $progress_is_dirty = 1;
1045 ($round_num_freeslots > @freeslot && $todo_ptr+1 > $#jobstep_todo))
1047 last THISROUND if $main::please_freeze;
1048 if ($main::please_info)
1050 $main::please_info = 0;
1052 create_output_collection();
1054 update_progress_stats();
1059 if (!$gotsome || ($latest_refresh + 2 < scalar time))
1061 check_refresh_wanted();
1063 update_progress_stats();
1065 elsif (time - $progress_stats_updated >= 30 || $progress_is_dirty)
1067 update_progress_stats();
1070 select (undef, undef, undef, 0.1);
1072 $working_slot_count = scalar(grep { $_->{node}->{fail_count} == 0 &&
1073 $_->{node}->{hold_count} < 4 } @slot);
1074 if (($thisround_failed_multiple >= 8 && $thisround_succeeded == 0) ||
1075 ($thisround_failed_multiple >= 16 && $thisround_failed_multiple > $thisround_succeeded))
1077 my $message = "Repeated failure rate too high ($thisround_failed_multiple/"
1078 .($thisround_failed+$thisround_succeeded)
1079 .") -- giving up on this round";
1080 Log (undef, $message);
1084 # move slots from freeslot to holdslot (or back to freeslot) if necessary
1085 for (my $i=$#freeslot; $i>=0; $i--) {
1086 if ($slot[$freeslot[$i]]->{node}->{hold_until} > scalar time) {
1087 push @holdslot, (splice @freeslot, $i, 1);
1090 for (my $i=$#holdslot; $i>=0; $i--) {
1091 if ($slot[$holdslot[$i]]->{node}->{hold_until} <= scalar time) {
1092 push @freeslot, (splice @holdslot, $i, 1);
1096 # give up if no nodes are succeeding
1097 if ($working_slot_count < 1) {
1098 Log(undef, "Every node has failed -- giving up");
1105 push @freeslot, splice @holdslot;
1106 map { $slot[$freeslot[$_]]->{node}->{losing_streak} = 0 } (0..$#freeslot);
1109 Log (undef, "wait for last ".(scalar keys %proc)." children to finish");
1112 if ($main::please_continue) {
1113 $main::please_continue = 0;
1116 $main::please_info = 0, freeze(), create_output_collection(), save_meta(1) if $main::please_info;
1118 if (!reapchildren())
1120 check_refresh_wanted();
1122 update_progress_stats();
1123 select (undef, undef, undef, 0.1);
1124 killem (keys %proc) if $main::please_freeze;
1128 update_progress_stats();
1129 freeze_if_want_freeze();
1132 if (!defined $main::success)
1134 if (!@jobstep_todo) {
1136 } elsif ($working_slot_count < 1) {
1137 save_output_collection();
1139 exit(EX_RETRY_UNLOCKED);
1140 } elsif ($thisround_succeeded == 0 &&
1141 ($thisround_failed == 0 || $thisround_failed > 4)) {
1142 my $message = "stop because $thisround_failed tasks failed and none succeeded";
1143 Log (undef, $message);
1148 goto ONELEVEL if !defined $main::success;
1151 release_allocation();
1153 my $collated_output = save_output_collection();
1154 Log (undef, "finish");
1159 if ($collated_output && $main::success) {
1160 $final_state = 'Complete';
1162 $final_state = 'Failed';
1164 $Job->update_attributes('state' => $final_state);
1166 exit (($final_state eq 'Complete') ? 0 : 1);
1170 sub update_progress_stats
1172 $progress_stats_updated = time;
1173 return if !$progress_is_dirty;
1174 my ($todo, $done, $running) = (scalar @jobstep_todo,
1175 scalar @jobstep_done,
1176 scalar keys(%proc));
1177 $Job->{'tasks_summary'} ||= {};
1178 $Job->{'tasks_summary'}->{'todo'} = $todo;
1179 $Job->{'tasks_summary'}->{'done'} = $done;
1180 $Job->{'tasks_summary'}->{'running'} = $running;
1181 $Job->update_attributes('tasks_summary' => $Job->{'tasks_summary'});
1182 Log (undef, "status: $done done, $running running, $todo todo");
1183 $progress_is_dirty = 0;
1190 my $pid = waitpid (-1, WNOHANG);
1191 return 0 if $pid <= 0;
1193 my $whatslot = ($slot[$proc{$pid}->{slot}]->{node}->{name}
1195 . $slot[$proc{$pid}->{slot}]->{cpu});
1196 my $jobstepid = $proc{$pid}->{jobstep};
1197 my $elapsed = time - $proc{$pid}->{time};
1198 my $Jobstep = $jobstep[$jobstepid];
1200 my $childstatus = $?;
1201 my $exitvalue = $childstatus >> 8;
1202 my $exitinfo = "exit ".exit_status_s($childstatus);
1203 $Jobstep->{'arvados_task'}->reload;
1204 my $task_success = $Jobstep->{'arvados_task'}->{success};
1206 Log ($jobstepid, "child $pid on $whatslot $exitinfo success=$task_success");
1208 if (!defined $task_success) {
1209 # task did not indicate one way or the other --> fail
1210 Log($jobstepid, sprintf(
1211 "ERROR: Task process exited %s, but never updated its task record to indicate success and record its output.",
1212 exit_status_s($childstatus)));
1213 $Jobstep->{'arvados_task'}->{success} = 0;
1214 $Jobstep->{'arvados_task'}->save;
1221 $temporary_fail ||= $Jobstep->{tempfail};
1222 $temporary_fail ||= ($exitvalue == TASK_TEMPFAIL);
1224 ++$thisround_failed;
1225 ++$thisround_failed_multiple if $Jobstep->{'failures'} >= 1;
1227 # Check for signs of a failed or misconfigured node
1228 if (++$slot[$proc{$pid}->{slot}]->{node}->{losing_streak} >=
1229 2+$slot[$proc{$pid}->{slot}]->{node}->{ncpus}) {
1230 # Don't count this against jobstep failure thresholds if this
1231 # node is already suspected faulty and srun exited quickly
1232 if ($slot[$proc{$pid}->{slot}]->{node}->{hold_until} &&
1234 Log ($jobstepid, "blaming failure on suspect node " .
1235 $slot[$proc{$pid}->{slot}]->{node}->{name});
1236 $temporary_fail ||= 1;
1238 ban_node_by_slot($proc{$pid}->{slot});
1241 Log ($jobstepid, sprintf('failure (#%d, %s) after %d seconds',
1242 ++$Jobstep->{'failures'},
1243 $temporary_fail ? 'temporary' : 'permanent',
1246 if (!$temporary_fail || $Jobstep->{'failures'} >= 3) {
1247 # Give up on this task, and the whole job
1250 # Put this task back on the todo queue
1251 push @jobstep_todo, $jobstepid;
1252 $Job->{'tasks_summary'}->{'failed'}++;
1256 ++$thisround_succeeded;
1257 $slot[$proc{$pid}->{slot}]->{node}->{losing_streak} = 0;
1258 $slot[$proc{$pid}->{slot}]->{node}->{hold_until} = 0;
1259 $slot[$proc{$pid}->{slot}]->{node}->{fail_count} = 0;
1260 push @jobstep_done, $jobstepid;
1261 Log ($jobstepid, "success in $elapsed seconds");
1263 $Jobstep->{exitcode} = $childstatus;
1264 $Jobstep->{finishtime} = time;
1265 $Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
1266 $Jobstep->{'arvados_task'}->save;
1267 process_stderr ($jobstepid, $task_success);
1268 Log ($jobstepid, sprintf("task output (%d bytes): %s",
1269 length($Jobstep->{'arvados_task'}->{output}),
1270 $Jobstep->{'arvados_task'}->{output}));
1272 close $reader{$jobstepid};
1273 delete $reader{$jobstepid};
1274 delete $slot[$proc{$pid}->{slot}]->{pid};
1275 push @freeslot, $proc{$pid}->{slot};
1278 if ($task_success) {
1280 my $newtask_list = [];
1281 my $newtask_results;
1283 $newtask_results = api_call(
1286 'created_by_job_task_uuid' => $Jobstep->{'arvados_task'}->{uuid}
1288 'order' => 'qsequence',
1289 'offset' => scalar(@$newtask_list),
1291 push(@$newtask_list, @{$newtask_results->{items}});
1292 } while (@{$newtask_results->{items}});
1293 foreach my $arvados_task (@$newtask_list) {
1295 'level' => $arvados_task->{'sequence'},
1297 'arvados_task' => $arvados_task
1299 push @jobstep, $jobstep;
1300 push @jobstep_todo, $#jobstep;
1304 $progress_is_dirty = 1;
1308 sub check_refresh_wanted
1310 my @stat = stat $ENV{"CRUNCH_REFRESH_TRIGGER"};
1311 if (@stat && $stat[9] > $latest_refresh) {
1312 $latest_refresh = scalar time;
1313 my $Job2 = api_call("jobs/get", uuid => $jobspec);
1314 for my $attr ('cancelled_at',
1315 'cancelled_by_user_uuid',
1316 'cancelled_by_client_uuid',
1318 $Job->{$attr} = $Job2->{$attr};
1320 if ($Job->{'state'} ne "Running") {
1321 if ($Job->{'state'} eq "Cancelled") {
1322 Log (undef, "Job cancelled at " . $Job->{'cancelled_at'} . " by user " . $Job->{'cancelled_by_user_uuid'});
1324 Log (undef, "Job state unexpectedly changed to " . $Job->{'state'});
1327 $main::please_freeze = 1;
1334 my $last_squeue_check = $squeue_checked;
1336 # Do not call `squeue` or check the kill list more than once every
1338 return if $last_squeue_check > time - 15;
1339 $squeue_checked = time;
1341 # Look for children from which we haven't received stderr data since
1342 # the last squeue check. If no such children exist, all procs are
1343 # alive and there's no need to even look at squeue.
1345 # As long as the crunchstat poll interval (10s) is shorter than the
1346 # squeue check interval (15s) this should make the squeue check an
1348 my $silent_procs = 0;
1349 for my $procinfo (values %proc)
1351 my $jobstep = $jobstep[$procinfo->{jobstep}];
1352 if ($jobstep->{stderr_at} < $last_squeue_check)
1357 return if $silent_procs == 0;
1359 # use killem() on procs whose killtime is reached
1360 while (my ($pid, $procinfo) = each %proc)
1362 my $jobstep = $jobstep[$procinfo->{jobstep}];
1363 if (exists $procinfo->{killtime}
1364 && $procinfo->{killtime} <= time
1365 && $jobstep->{stderr_at} < $last_squeue_check)
1368 if ($jobstep->{stderr_at}) {
1369 $sincewhen = " in last " . (time - $jobstep->{stderr_at}) . "s";
1371 Log($procinfo->{jobstep}, "killing orphaned srun process $pid (task not in slurm queue, no stderr received$sincewhen)");
1378 # here is an opportunity to check for mysterious problems with local procs
1382 # Get a list of steps still running. Note: squeue(1) says --steps
1383 # selects a format (which we override anyway) and allows us to
1384 # specify which steps we're interested in (which we don't).
1385 # Importantly, it also changes the meaning of %j from "job name" to
1386 # "step name" and (although this isn't mentioned explicitly in the
1387 # docs) switches from "one line per job" mode to "one line per step"
1388 # mode. Without it, we'd just get a list of one job, instead of a
1390 my @squeue = `squeue --jobs=\Q$ENV{SLURM_JOB_ID}\E --steps --format='%j' --noheader`;
1393 Log(undef, "warning: squeue exit status $? ($!)");
1398 # which of my jobsteps are running, according to squeue?
1400 for my $jobstepname (@squeue)
1402 $ok{$jobstepname} = 1;
1405 # Check for child procs >60s old and not mentioned by squeue.
1406 while (my ($pid, $procinfo) = each %proc)
1408 if ($procinfo->{time} < time - 60
1409 && $procinfo->{jobstepname}
1410 && !exists $ok{$procinfo->{jobstepname}}
1411 && !exists $procinfo->{killtime})
1413 # According to slurm, this task has ended (successfully or not)
1414 # -- but our srun child hasn't exited. First we must wait (30
1415 # seconds) in case this is just a race between communication
1416 # channels. Then, if our srun child process still hasn't
1417 # terminated, we'll conclude some slurm communication
1418 # error/delay has caused the task to die without notifying srun,
1419 # and we'll kill srun ourselves.
1420 $procinfo->{killtime} = time + 30;
1421 Log($procinfo->{jobstep}, "notice: task is not in slurm queue but srun process $pid has not exited");
1427 sub release_allocation
1431 Log (undef, "release job allocation");
1432 system "scancel $ENV{SLURM_JOB_ID}";
1440 foreach my $job (keys %reader)
1443 if (0 < sysread ($reader{$job}, $buf, 65536))
1445 print STDERR $buf if $ENV{CRUNCH_DEBUG};
1446 $jobstep[$job]->{stderr_at} = time;
1447 $jobstep[$job]->{stderr} .= $buf;
1449 # Consume everything up to the last \n
1450 preprocess_stderr ($job);
1452 if (length ($jobstep[$job]->{stderr}) > 16384)
1454 # If we get a lot of stderr without a newline, chop off the
1455 # front to avoid letting our buffer grow indefinitely.
1456 substr ($jobstep[$job]->{stderr},
1457 0, length($jobstep[$job]->{stderr}) - 8192) = "";
1466 sub preprocess_stderr
1470 while ($jobstep[$job]->{stderr} =~ /^(.*?)\n/) {
1472 substr $jobstep[$job]->{stderr}, 0, 1+length($line), "";
1473 Log ($job, "stderr $line");
1474 if ($line =~ /srun: error: (SLURM job $ENV{SLURM_JOB_ID} has expired|Unable to confirm allocation for job $ENV{SLURM_JOB_ID})/) {
1476 $main::please_freeze = 1;
1478 elsif ($line =~ /srun: error: (Node failure on|Aborting, .*\bio error\b)/) {
1479 my $job_slot_index = $jobstep[$job]->{slotindex};
1480 $slot[$job_slot_index]->{node}->{fail_count}++;
1481 $jobstep[$job]->{tempfail} = 1;
1482 ban_node_by_slot($job_slot_index);
1484 elsif ($line =~ /srun: error: (Unable to create job step|.*: Communication connection failure)/) {
1485 $jobstep[$job]->{tempfail} = 1;
1486 ban_node_by_slot($jobstep[$job]->{slotindex});
1488 elsif ($line =~ /\bKeep(Read|Write|Request)Error:/) {
1489 $jobstep[$job]->{tempfail} = 1;
1498 my $task_success = shift;
1499 preprocess_stderr ($job);
1502 Log ($job, "stderr $_");
1503 } split ("\n", $jobstep[$job]->{stderr});
1510 if (!open($keep, "-|", "arv-get", "--retries", retry_count(), $hash)) {
1511 Log(undef, "fetch_block run error from arv-get $hash: $!");
1514 my $output_block = "";
1517 my $bytes = sysread($keep, $buf, 1024 * 1024);
1518 if (!defined $bytes) {
1519 Log(undef, "fetch_block read error from arv-get: $!");
1520 $output_block = undef;
1522 } elsif ($bytes == 0) {
1523 # sysread returns 0 at the end of the pipe.
1526 # some bytes were read into buf.
1527 $output_block .= $buf;
1532 Log(undef, "fetch_block arv-get exited " . exit_status_s($?));
1533 $output_block = undef;
1535 return $output_block;
1538 # Create a collection by concatenating the output of all tasks (each
1539 # task's output is either a manifest fragment, a locator for a
1540 # manifest fragment stored in Keep, or nothing at all). Return the
1541 # portable_data_hash of the new collection.
1542 sub create_output_collection
1544 Log (undef, "collate");
1546 my ($child_out, $child_in);
1547 my $pid = open2($child_out, $child_in, 'python', '-c', q{
1550 print (arvados.api("v1").collections().
1551 create(body={"manifest_text": sys.stdin.read()}).
1552 execute(num_retries=int(sys.argv[1]))["portable_data_hash"])
1556 my $manifest_size = 0;
1560 my $output = $_->{'arvados_task'}->{output};
1561 next if (!defined($output));
1563 if ($output =~ /^[0-9a-f]{32}(\+\S+)*$/) {
1564 $next_write = fetch_block($output);
1566 $next_write = $output;
1568 if (defined($next_write)) {
1569 if (!defined(syswrite($child_in, $next_write))) {
1570 # There's been an error writing. Stop the loop.
1571 # We'll log details about the exit code later.
1574 $manifest_size += length($next_write);
1577 my $uuid = $_->{'arvados_task'}->{'uuid'};
1578 Log (undef, "Error retrieving '$output' output by task $task_idx ($uuid)");
1583 Log(undef, "collated output manifest text to send to API server is $manifest_size bytes with access tokens");
1586 my $s = IO::Select->new($child_out);
1587 if ($s->can_read(120)) {
1588 sysread($child_out, $joboutput, 1024 * 1024);
1591 Log(undef, "output collection creation exited " . exit_status_s($?));
1597 Log (undef, "timed out while creating output collection");
1598 foreach my $signal (2, 2, 2, 15, 15, 9) {
1599 kill($signal, $pid);
1600 last if waitpid($pid, WNOHANG) == -1;
1609 # Calls create_output_collection, logs the result, and returns it.
1610 # If that was successful, save that as the output in the job record.
1611 sub save_output_collection {
1612 my $collated_output = create_output_collection();
1614 if (!$collated_output) {
1615 Log(undef, "Failed to write output collection");
1618 Log(undef, "job output $collated_output");
1619 $Job->update_attributes('output' => $collated_output);
1621 return $collated_output;
1628 my $sig = 2; # SIGINT first
1629 if (exists $proc{$_}->{"sent_$sig"} &&
1630 time - $proc{$_}->{"sent_$sig"} > 4)
1632 $sig = 15; # SIGTERM if SIGINT doesn't work
1634 if (exists $proc{$_}->{"sent_$sig"} &&
1635 time - $proc{$_}->{"sent_$sig"} > 4)
1637 $sig = 9; # SIGKILL if SIGTERM doesn't work
1639 if (!exists $proc{$_}->{"sent_$sig"})
1641 Log ($proc{$_}->{jobstep}, "sending 2x signal $sig to pid $_");
1643 select (undef, undef, undef, 0.1);
1646 kill $sig, $_; # srun wants two SIGINT to really interrupt
1648 $proc{$_}->{"sent_$sig"} = time;
1649 $proc{$_}->{"killedafter"} = time - $proc{$_}->{"time"};
1659 vec($bits,fileno($_),1) = 1;
1665 # Send log output to Keep via arv-put.
1667 # $log_pipe_in and $log_pipe_out are the input and output filehandles to the arv-put pipe.
1668 # $log_pipe_out_buf is a string containing all output read from arv-put so far.
1669 # $log_pipe_out_select is an IO::Select object around $log_pipe_out.
1670 # $log_pipe_pid is the pid of the arv-put subprocess.
1672 # The only functions that should access these variables directly are:
1674 # log_writer_start($logfilename)
1675 # Starts an arv-put pipe, reading data on stdin and writing it to
1676 # a $logfilename file in an output collection.
1678 # log_writer_read_output([$timeout])
1679 # Read output from $log_pipe_out and append it to $log_pipe_out_buf.
1680 # Passes $timeout to the select() call, with a default of 0.01.
1681 # Returns the result of the last read() call on $log_pipe_out, or
1682 # -1 if read() wasn't called because select() timed out.
1683 # Only other log_writer_* functions should need to call this.
1685 # log_writer_send($txt)
1686 # Writes $txt to the output log collection.
1688 # log_writer_finish()
1689 # Closes the arv-put pipe and returns the output that it produces.
1691 # log_writer_is_active()
1692 # Returns a true value if there is currently a live arv-put
1693 # process, false otherwise.
1695 my ($log_pipe_in, $log_pipe_out, $log_pipe_out_buf, $log_pipe_out_select,
1698 sub log_writer_start($)
1700 my $logfilename = shift;
1701 $log_pipe_pid = open2($log_pipe_out, $log_pipe_in,
1705 '--filename', $logfilename,
1707 $log_pipe_out_buf = "";
1708 $log_pipe_out_select = IO::Select->new($log_pipe_out);
1711 sub log_writer_read_output {
1712 my $timeout = shift || 0.01;
1714 while ($read && $log_pipe_out_select->can_read($timeout)) {
1715 $read = read($log_pipe_out, $log_pipe_out_buf, 65536,
1716 length($log_pipe_out_buf));
1718 if (!defined($read)) {
1719 Log(undef, "error reading log manifest from arv-put: $!");
1724 sub log_writer_send($)
1727 print $log_pipe_in $txt;
1728 log_writer_read_output();
1731 sub log_writer_finish()
1733 return unless $log_pipe_pid;
1735 close($log_pipe_in);
1737 my $logger_failed = 0;
1738 my $read_result = log_writer_read_output(120);
1739 if ($read_result == -1) {
1740 $logger_failed = -1;
1741 Log (undef, "timed out reading from 'arv-put'");
1742 } elsif ($read_result != 0) {
1743 $logger_failed = -2;
1744 Log(undef, "failed to read arv-put log manifest to EOF");
1747 waitpid($log_pipe_pid, 0);
1749 $logger_failed ||= $?;
1750 Log(undef, "log_writer_finish: arv-put exited " . exit_status_s($?))
1753 close($log_pipe_out);
1754 my $arv_put_output = $logger_failed ? undef : $log_pipe_out_buf;
1755 $log_pipe_pid = $log_pipe_in = $log_pipe_out = $log_pipe_out_buf =
1756 $log_pipe_out_select = undef;
1758 return $arv_put_output;
1761 sub log_writer_is_active() {
1762 return $log_pipe_pid;
1765 sub Log # ($jobstep_id, $logmessage)
1767 if ($_[1] =~ /\n/) {
1768 for my $line (split (/\n/, $_[1])) {
1773 my $fh = select STDERR; $|=1; select $fh;
1774 my $message = sprintf ("%s %d %s %s", $job_id, $$, @_);
1775 $message =~ s{([^ -\176])}{"\\" . sprintf ("%03o", ord($1))}ge;
1778 if (log_writer_is_active() || -t STDERR) {
1779 my @gmtime = gmtime;
1780 $datetime = sprintf ("%04d-%02d-%02d_%02d:%02d:%02d",
1781 $gmtime[5]+1900, $gmtime[4]+1, @gmtime[3,2,1,0]);
1783 print STDERR ((-t STDERR) ? ($datetime." ".$message) : $message);
1785 if (log_writer_is_active()) {
1786 log_writer_send($datetime . " " . $message);
1793 my ($package, $file, $line) = caller;
1794 my $message = "@_ at $file line $line\n";
1795 Log (undef, $message);
1796 freeze() if @jobstep_todo;
1797 create_output_collection() if @jobstep_todo;
1807 if ($Job->{'state'} eq 'Cancelled') {
1808 $Job->update_attributes('finished_at' => scalar gmtime);
1810 $Job->update_attributes('state' => 'Failed');
1817 my $justcheckpoint = shift; # false if this will be the last meta saved
1818 return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm
1819 return unless log_writer_is_active();
1820 my $log_manifest = log_writer_finish();
1821 return unless defined($log_manifest);
1824 my $prev_log_coll = api_call("collections/get", uuid => $Job->{log});
1825 $log_manifest = $prev_log_coll->{manifest_text} . $log_manifest;
1828 my $log_coll = api_call(
1829 "collections/create", ensure_unique_name => 1, collection => {
1830 manifest_text => $log_manifest,
1831 owner_uuid => $Job->{owner_uuid},
1832 name => sprintf("Log from %s job %s", $Job->{script}, $Job->{uuid}),
1834 Log(undef, "log collection is " . $log_coll->{portable_data_hash});
1835 $Job->update_attributes('log' => $log_coll->{portable_data_hash});
1839 sub freeze_if_want_freeze
1841 if ($main::please_freeze)
1843 release_allocation();
1846 # kill some srun procs before freeze+stop
1847 map { $proc{$_} = {} } @_;
1850 killem (keys %proc);
1851 select (undef, undef, undef, 0.1);
1853 while (($died = waitpid (-1, WNOHANG)) > 0)
1855 delete $proc{$died};
1860 create_output_collection();
1870 Log (undef, "Freeze not implemented");
1877 croak ("Thaw not implemented");
1893 $s =~ s{\\(.)}{$1 eq "n" ? "\n" : $1}ge;
1900 my $srunargs = shift;
1901 my $execargs = shift;
1902 my $opts = shift || {};
1904 my $args = $have_slurm ? [@$srunargs, @$execargs] : $execargs;
1906 $Data::Dumper::Terse = 1;
1907 $Data::Dumper::Indent = 0;
1908 my $show_cmd = Dumper($args);
1909 $show_cmd =~ s/(TOKEN\\*=)[^\s\']+/${1}[...]/g;
1910 $show_cmd =~ s/\n/ /g;
1911 if ($opts->{fork}) {
1912 Log(undef, "starting: $show_cmd");
1914 # This is a child process: parent is in charge of reading our
1915 # stderr and copying it to Log() if needed.
1916 warn "starting: $show_cmd\n";
1919 if (defined $stdin) {
1920 my $child = open STDIN, "-|";
1921 defined $child or die "no fork: $!";
1923 print $stdin or die $!;
1924 close STDOUT or die $!;
1929 return system (@$args) if $opts->{fork};
1932 warn "ENV size is ".length(join(" ",%ENV));
1933 die "exec failed: $!: @$args";
1937 sub ban_node_by_slot {
1938 # Don't start any new jobsteps on this node for 60 seconds
1940 $slot[$slotid]->{node}->{hold_until} = 60 + scalar time;
1941 $slot[$slotid]->{node}->{hold_count}++;
1942 Log (undef, "backing off node " . $slot[$slotid]->{node}->{name} . " for 60 seconds");
1947 my ($lockfile, $error_message) = @_;
1948 open L, ">", $lockfile or croak("$lockfile: $!");
1949 if (!flock L, LOCK_EX|LOCK_NB) {
1950 croak("Can't lock $lockfile: $error_message\n");
1954 sub find_docker_image {
1955 # Given a Keep locator, check to see if it contains a Docker image.
1956 # If so, return its stream name and Docker hash.
1957 # If not, return undef for both values.
1958 my $locator = shift;
1959 my ($streamname, $filename);
1960 my $image = api_call("collections/get", uuid => $locator);
1962 foreach my $line (split(/\n/, $image->{manifest_text})) {
1963 my @tokens = split(/\s+/, $line);
1965 $streamname = shift(@tokens);
1966 foreach my $filedata (grep(/^\d+:\d+:/, @tokens)) {
1967 if (defined($filename)) {
1968 return (undef, undef); # More than one file in the Collection.
1970 $filename = (split(/:/, $filedata, 3))[2];
1975 if (defined($filename) and ($filename =~ /^([0-9A-Fa-f]{64})\.tar$/)) {
1976 return ($streamname, $1);
1978 return (undef, undef);
1983 # Calculate the number of times an operation should be retried,
1984 # assuming exponential backoff, and that we're willing to retry as
1985 # long as tasks have been running. Enforce a minimum of 3 retries.
1986 my ($starttime, $endtime, $timediff, $retries);
1988 $starttime = $jobstep[0]->{starttime};
1989 $endtime = $jobstep[-1]->{finishtime};
1991 if (!defined($starttime)) {
1993 } elsif (!defined($endtime)) {
1994 $timediff = time - $starttime;
1996 $timediff = ($endtime - $starttime) - (time - $endtime);
1998 if ($timediff > 0) {
1999 $retries = int(log($timediff) / log(2));
2001 $retries = 1; # Use the minimum.
2003 return ($retries > 3) ? $retries : 3;
2007 # Pass in two function references.
2008 # This method will be called with the remaining arguments.
2009 # If it dies, retry it with exponential backoff until it succeeds,
2010 # or until the current retry_count is exhausted. After each failure
2011 # that can be retried, the second function will be called with
2012 # the current try count (0-based), next try time, and error message.
2013 my $operation = shift;
2014 my $retry_callback = shift;
2015 my $retries = retry_count();
2016 foreach my $try_count (0..$retries) {
2017 my $next_try = time + (2 ** $try_count);
2018 my $result = eval { $operation->(@_); };
2021 } elsif ($try_count < $retries) {
2022 $retry_callback->($try_count, $next_try, $@);
2023 my $sleep_time = $next_try - time;
2024 sleep($sleep_time) if ($sleep_time > 0);
2027 # Ensure the error message ends in a newline, so Perl doesn't add
2028 # retry_op's line number to it.
2034 # Pass in a /-separated API method name, and arguments for it.
2035 # This function will call that method, retrying as needed until
2036 # the current retry_count is exhausted, with a log on the first failure.
2037 my $method_name = shift;
2038 my $log_api_retry = sub {
2039 my ($try_count, $next_try_at, $errmsg) = @_;
2040 $errmsg =~ s/\s*\bat \Q$0\E line \d+\.?\s*//;
2041 $errmsg =~ s/\s/ /g;
2042 $errmsg =~ s/\s+$//;
2044 if ($next_try_at < time) {
2045 $retry_msg = "Retrying.";
2047 my $next_try_fmt = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($next_try_at);
2048 $retry_msg = "Retrying at $next_try_fmt.";
2050 Log(undef, "API method $method_name failed: $errmsg. $retry_msg");
2053 foreach my $key (split(/\//, $method_name)) {
2054 $method = $method->{$key};
2056 return retry_op(sub { $method->execute(@_); }, $log_api_retry, @_);
2060 # Given a $?, return a human-readable exit code string like "0" or
2061 # "1" or "0 with signal 1" or "1 with signal 11".
2062 my $exitcode = shift;
2063 my $s = $exitcode >> 8;
2064 if ($exitcode & 0x7f) {
2065 $s .= " with signal " . ($exitcode & 0x7f);
2067 if ($exitcode & 0x80) {
2068 $s .= " with core dump";
2073 sub handle_readall {
2074 # Pass in a glob reference to a file handle.
2075 # Read all its contents and return them as a string.
2076 my $fh_glob_ref = shift;
2078 return <$fh_glob_ref>;
2081 sub tar_filename_n {
2083 return sprintf("%s/git.%s.%d.tar", $ENV{CRUNCH_TMP}, $job_id, $n);
2086 sub add_git_archive {
2087 # Pass in a git archive command as a string or list, a la system().
2088 # This method will save its output to be included in the archive sent to the
2092 if (!open(GIT_ARCHIVE, ">", tar_filename_n($git_tar_count))) {
2093 croak("Failed to save git archive: $!");
2095 my $git_pid = open2(">&GIT_ARCHIVE", $git_input, @_);
2097 waitpid($git_pid, 0);
2100 croak("Failed to save git archive: git exited " . exit_status_s($?));
2104 sub combined_git_archive {
2105 # Combine all saved tar archives into a single archive, then return its
2106 # contents in a string. Return undef if no archives have been saved.
2107 if ($git_tar_count < 1) {
2110 my $base_tar_name = tar_filename_n(1);
2111 foreach my $tar_to_append (map { tar_filename_n($_); } (2..$git_tar_count)) {
2112 my $tar_exit = system("tar", "-Af", $base_tar_name, $tar_to_append);
2113 if ($tar_exit != 0) {
2114 croak("Error preparing build archive: tar -A exited " .
2115 exit_status_s($tar_exit));
2118 if (!open(GIT_TAR, "<", $base_tar_name)) {
2119 croak("Could not open build archive: $!");
2121 my $tar_contents = handle_readall(\*GIT_TAR);
2123 return $tar_contents;
2126 sub set_nonblocking {
2128 my $flags = fcntl ($fh, F_GETFL, 0) or croak ($!);
2129 fcntl ($fh, F_SETFL, $flags | O_NONBLOCK) or croak ($!);
2135 # This is crunch-job's internal dispatch script. crunch-job running on the API
2136 # server invokes this script on individual compute nodes, or localhost if we're
2137 # running a job locally. It gets called in two modes:
2139 # * No arguments: Installation mode. Read a tar archive from the DATA
2140 # file handle; it includes the Crunch script's source code, and
2141 # maybe SDKs as well. Those should be installed in the proper
2142 # locations. This runs outside of any Docker container, so don't try to
2143 # introspect Crunch's runtime environment.
2145 # * With arguments: Crunch script run mode. This script should set up the
2146 # environment, then run the command specified in the arguments. This runs
2147 # inside any Docker container.
2150 use File::Path qw( make_path remove_tree );
2151 use POSIX qw(getcwd);
2153 use constant TASK_TEMPFAIL => 111;
2155 # Map SDK subdirectories to the path environments they belong to.
2156 my %SDK_ENVVARS = ("perl/lib" => "PERLLIB", "ruby/lib" => "RUBYLIB");
2158 my $destdir = $ENV{"CRUNCH_SRC"};
2159 my $archive_hash = $ENV{"CRUNCH_GIT_ARCHIVE_HASH"};
2160 my $repo = $ENV{"CRUNCH_SRC_URL"};
2161 my $install_dir = $ENV{"CRUNCH_INSTALL"} || (getcwd() . "/opt");
2162 my $job_work = $ENV{"JOB_WORK"};
2163 my $task_work = $ENV{"TASK_WORK"};
2165 open(STDOUT_ORIG, ">&", STDOUT);
2166 open(STDERR_ORIG, ">&", STDERR);
2168 for my $dir ($destdir, $job_work, $task_work) {
2171 -e $dir or die "Failed to create temporary directory ($dir): $!";
2176 remove_tree($task_work, {keep_root => 1});
2179 ### Crunch script run mode
2181 # We want to do routine logging during task 0 only. This gives the user
2182 # the information they need, but avoids repeating the information for every
2185 if ($ENV{TASK_SEQUENCE} eq "0") {
2188 printf STDERR_ORIG "[Crunch] $msg\n", @_;
2194 my $python_src = "$install_dir/python";
2195 my $venv_dir = "$job_work/.arvados.venv";
2196 my $venv_built = -e "$venv_dir/bin/activate";
2197 if ((!$venv_built) and (-d $python_src) and can_run("virtualenv")) {
2198 shell_or_die(undef, "virtualenv", "--quiet", "--system-site-packages",
2199 "--python=python2.7", $venv_dir);
2200 shell_or_die(TASK_TEMPFAIL, "$venv_dir/bin/pip", "--quiet", "install", "-I", $python_src);
2202 $Log->("Built Python SDK virtualenv");
2205 my @pysdk_version_cmd = ("python", "-c",
2206 "from pkg_resources import get_distribution as get; print get('arvados-python-client').version");
2208 $Log->("Running in Python SDK virtualenv");
2209 @pysdk_version_cmd = ();
2210 my $orig_argv = join(" ", map { quotemeta($_); } @ARGV);
2211 @ARGV = ("/bin/sh", "-ec",
2212 ". \Q$venv_dir/bin/activate\E; exec $orig_argv");
2213 } elsif (-d $python_src) {
2214 $Log->("Warning: virtualenv not found inside Docker container default " .
2215 "\$PATH. Can't install Python SDK.");
2218 if (@pysdk_version_cmd) {
2219 open(my $pysdk_version_pipe, "-|", @pysdk_version_cmd);
2220 my $pysdk_version = <$pysdk_version_pipe>;
2221 close($pysdk_version_pipe);
2223 chomp($pysdk_version);
2224 $Log->("Using Arvados SDK version $pysdk_version");
2226 # A lot could've gone wrong here, but pretty much all of it means that
2227 # Python won't be able to load the Arvados SDK.
2228 $Log->("Warning: Arvados SDK not found");
2232 while (my ($sdk_dir, $sdk_envkey) = each(%SDK_ENVVARS)) {
2233 my $sdk_path = "$install_dir/$sdk_dir";
2235 if ($ENV{$sdk_envkey}) {
2236 $ENV{$sdk_envkey} = "$sdk_path:" . $ENV{$sdk_envkey};
2238 $ENV{$sdk_envkey} = $sdk_path;
2240 $Log->("Arvados SDK added to %s", $sdk_envkey);
2245 die "Cannot exec `@ARGV`: $!";
2248 ### Installation mode
2249 open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
2251 if (readlink ("$destdir.archive_hash") eq $archive_hash && -d $destdir) {
2252 # This exact git archive (source + arvados sdk) is already installed
2253 # here, so there's no need to reinstall it.
2255 # We must consume our DATA section, though: otherwise the process
2256 # feeding it to us will get SIGPIPE.
2258 while (read(DATA, $buf, 65536)) { }
2263 unlink "$destdir.archive_hash";
2267 # Ignore SIGPIPE: we check retval of close() instead. See perlipc(1).
2268 local $SIG{PIPE} = "IGNORE";
2269 warn "Extracting archive: $archive_hash\n";
2270 # --ignore-zeros is necessary sometimes: depending on how much NUL
2271 # padding tar -A put on our combined archive (which in turn depends
2272 # on the length of the component archives) tar without
2273 # --ignore-zeros will exit before consuming stdin and cause close()
2274 # to fail on the resulting SIGPIPE.
2275 if (!open(TARX, "|-", "tar", "--ignore-zeros", "-xC", $destdir)) {
2276 die "Error launching 'tar -xC $destdir': $!";
2278 # If we send too much data to tar in one write (> 4-5 MiB), it stops, and we
2279 # get SIGPIPE. We must feed it data incrementally.
2281 while (read(DATA, $tar_input, 65536)) {
2282 print TARX $tar_input;
2285 die "'tar -xC $destdir' exited $?: $!";
2291 my $sdk_root = "$destdir/.arvados.sdk/sdk";
2293 foreach my $sdk_lang (("python",
2294 map { (split /\//, $_, 2)[0]; } keys(%SDK_ENVVARS))) {
2295 if (-d "$sdk_root/$sdk_lang") {
2296 if (!rename("$sdk_root/$sdk_lang", "$install_dir/$sdk_lang")) {
2297 die "Failed to install $sdk_lang SDK: $!";
2303 my $python_dir = "$install_dir/python";
2304 if ((-d $python_dir) and can_run("python2.7")) {
2305 open(my $egg_info_pipe, "-|",
2306 "python2.7 \Q$python_dir/setup.py\E egg_info 2>&1 >/dev/null");
2307 my @egg_info_errors = <$egg_info_pipe>;
2308 close($egg_info_pipe);
2311 if (@egg_info_errors and (($egg_info_errors[-1] =~ /\bgit\b/) or ($egg_info_errors[-1] =~ /\[Errno 2\]/))) {
2312 # egg_info apparently failed because it couldn't ask git for a build tag.
2313 # Specify no build tag.
2314 open(my $pysdk_cfg, ">>", "$python_dir/setup.cfg");
2315 print $pysdk_cfg "\n[egg_info]\ntag_build =\n";
2318 my $egg_info_exit = $? >> 8;
2319 foreach my $errline (@egg_info_errors) {
2322 warn "python setup.py egg_info failed: exit $egg_info_exit";
2323 exit ($egg_info_exit || 1);
2328 # Hide messages from the install script (unless it fails: shell_or_die
2329 # will show $destdir.log in that case).
2330 open(STDOUT, ">>", "$destdir.log");
2331 open(STDERR, ">&", STDOUT);
2333 if (-e "$destdir/crunch_scripts/install") {
2334 shell_or_die (undef, "$destdir/crunch_scripts/install", $install_dir);
2335 } elsif (!-e "./install.sh" && -e "./tests/autotests.sh") {
2337 shell_or_die (undef, "./tests/autotests.sh", $install_dir);
2338 } elsif (-e "./install.sh") {
2339 shell_or_die (undef, "./install.sh", $install_dir);
2342 if ($archive_hash) {
2343 unlink "$destdir.archive_hash.new";
2344 symlink ($archive_hash, "$destdir.archive_hash.new") or die "$destdir.archive_hash.new: $!";
2345 rename ("$destdir.archive_hash.new", "$destdir.archive_hash") or die "$destdir.archive_hash: $!";
2351 my $command_name = shift;
2352 open(my $which, "-|", "which", $command_name);
2353 while (<$which>) { }
2360 my $exitcode = shift;
2362 if ($ENV{"DEBUG"}) {
2363 print STDERR "@_\n";
2365 if (system (@_) != 0) {
2368 my $exitstatus = sprintf("exit %d signal %d", $code >> 8, $code & 0x7f);
2369 open STDERR, ">&STDERR_ORIG";
2370 system ("cat $destdir.log >&2");
2371 warn "@_ failed ($err): $exitstatus";
2372 if (defined($exitcode)) {
2376 exit (($code >> 8) || 1);