X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/638c2c2258e7004f9a86277588d50f1243a297c8..5d0474c43296f2bac56c5b40e1d50ab0022a28b1:/sdk/cli/bin/crunch-job diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job index 4d2aad997a..e4a2e543fd 100755 --- a/sdk/cli/bin/crunch-job +++ b/sdk/cli/bin/crunch-job @@ -74,6 +74,7 @@ behavior (e.g., cancel job if cancelled_at becomes non-nil). use strict; use POSIX ':sys_wait_h'; +use POSIX qw(strftime); use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK); use Arvados; use Digest::MD5 qw(md5_hex); @@ -152,26 +153,15 @@ if ($job_has_uuid) { $Job = $arv->{'jobs'}->{'get'}->execute('uuid' => $jobspec); if (!$force_unlock) { - # If some other crunch-job process has grabbed this job (or we see - # other evidence that the job is already underway) we exit - # EX_TEMPFAIL so crunch-dispatch (our parent process) doesn't - # mark the job as failed. - if ($Job->{'is_locked_by_uuid'}) { - Log(undef, "Job is locked by " . $Job->{'is_locked_by_uuid'}); + # Claim this job, and make sure nobody else does + eval { + # lock() sets is_locked_by_uuid and changes state to Running. + $arv->{'jobs'}->{'lock'}->execute('uuid' => $Job->{'uuid'}) + }; + if ($@) { + Log(undef, "Error while locking job, exiting ".EX_TEMPFAIL); exit EX_TEMPFAIL; - } - if ($Job->{'success'} ne undef) { - Log(undef, "Job 'success' flag (" . $Job->{'success'} . ") is not null"); - exit EX_TEMPFAIL; - } - if ($Job->{'running'}) { - Log(undef, "Job 'running' flag is already set"); - exit EX_TEMPFAIL; - } - if ($Job->{'started_at'}) { - Log(undef, "Job 'started_at' time is already set (" . $Job->{'started_at'} . ")"); - exit EX_TEMPFAIL; - } + }; } } else @@ -276,25 +266,11 @@ foreach (@sinfo) @slot = sort { $a->{cpu} <=> $b->{cpu} } @slot; - -my $jobmanager_id; -if ($job_has_uuid) -{ - # Claim this job, and make sure nobody else does - unless ($Job->update_attributes('is_locked_by_uuid' => $User->{'uuid'}) && - $Job->{'is_locked_by_uuid'} == $User->{'uuid'}) { - Log(undef, "Error while updating / locking job, exiting ".EX_TEMPFAIL); - exit EX_TEMPFAIL; - } - $Job->update_attributes('started_at' => scalar gmtime, - 'running' => 1, - 'success' => undef, - 'tasks_summary' => { 'failed' => 0, - 'todo' => 1, - 'running' => 0, - 'done' => 0 }); -} - +$Job->update_attributes( + 'tasks_summary' => { 'failed' => 0, + 'todo' => 1, + 'running' => 0, + 'done' => 0 }); Log (undef, "start"); $SIG{'INT'} = sub { $main::please_freeze = 1; }; @@ -447,9 +423,10 @@ else if ($treeish =~ m{^\d{1,4}$}) { my $gitlog = `git --git-dir=\Q$repo\E log --pretty="format:%H" --grep="git-svn-id:.*\@"\Q$treeish\E" " master`; chomp $gitlog; - if ($gitlog =~ /^[a-f0-9]{40}$/) { + Log(undef, "git Subversion search exited $?"); + if (($? == 0) && ($gitlog =~ /^[a-f0-9]{40}$/)) { $commit = $gitlog; - Log (undef, "Using commit $commit for script_version $treeish"); + Log(undef, "Using commit $commit for Subversion revision $treeish"); } } @@ -458,21 +435,19 @@ else if (!defined $commit) { my $found = `git --git-dir=\Q$repo\E rev-list -1 ''\Q$treeish\E`; chomp $found; - if ($found =~ /^[0-9a-f]{40}$/s) { + Log(undef, "git rev-list exited $? with result '$found'"); + if (($? == 0) && ($found =~ /^[0-9a-f]{40}$/s)) { $commit = $found; + Log(undef, "Using commit $commit for tree-ish $treeish"); if ($commit ne $treeish) { # Make sure we record the real commit id in the database, # frozentokey, logs, etc. -- instead of an abbreviation or a # branch name which can become ambiguous or point to a # different commit in the future. - $ENV{"CRUNCH_SRC_COMMIT"} = $commit; - Log (undef, "Using commit $commit for tree-ish $treeish"); - if ($commit ne $treeish) { - $Job->{'script_version'} = $commit; - !$job_has_uuid or - $Job->update_attributes('script_version' => $commit) or - croak("Error while updating job"); - } + $Job->{'script_version'} = $commit; + !$job_has_uuid or + $Job->update_attributes('script_version' => $commit) or + croak("Error while updating job"); } } } @@ -482,6 +457,7 @@ else @execargs = ("sh", "-c", "mkdir -p $ENV{CRUNCH_INSTALL} && cd $ENV{CRUNCH_TMP} && perl -"); $git_archive = `git --git-dir=\Q$repo\E archive ''\Q$commit\E`; + croak("git archive failed: exit " . ($? >> 8)) if ($? != 0); } else { croak ("could not figure out commit id for $treeish"); @@ -656,8 +632,10 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++) if ($build_script) { $build_script_to_send = $build_script; + $command .= + "&& perl -"; } - $command .= "&& exec arv-mount --allow-other $ENV{TASK_KEEPMOUNT} --exec "; + $command .= "&& exec arv-mount --by-id --allow-other $ENV{TASK_KEEPMOUNT} --exec "; if ($docker_hash) { $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -cgroup-parent=docker -cgroup-cid=$ENV{TASK_WORK}/docker.cid -poll=10000 "; @@ -668,12 +646,16 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++) $command .= q{$(ip -o address show scope global | gawk 'match($4, /^([0-9\.:]+)\//, x){print "--dns", x[1]}') }; + $command .= "--volume=\Q$ENV{CRUNCH_SRC}:/tmp/crunch-src:ro\E "; $command .= "--volume=\Q$ENV{TASK_KEEPMOUNT}:/keep:ro\E "; $command .= "--env=\QHOME=/home/crunch\E "; while (my ($env_key, $env_val) = each %ENV) { if ($env_key =~ /^(ARVADOS|JOB|TASK)_/) { - if ($env_key eq "TASK_KEEPMOUNT") { + if ($env_key eq "TASK_WORK") { + $command .= "--env=\QTASK_WORK=/tmp/crunch-job\E "; + } + elsif ($env_key eq "TASK_KEEPMOUNT") { $command .= "--env=\QTASK_KEEPMOUNT=/keep\E "; } else { @@ -685,13 +667,11 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++) $command .= "--env=\QCRUNCH_SRC=/tmp/crunch-src\E "; $command .= "\Q$docker_hash\E "; $command .= "stdbuf --output=0 --error=0 "; - $command .= "perl - "; $command .= "/tmp/crunch-src/crunch_scripts/" . $Job->{"script"}; } else { # Non-docker run $command .= "crunchstat -cgroup-root=/sys/fs/cgroup -poll=10000 "; $command .= "stdbuf --output=0 --error=0 "; - $command .= "perl - "; $command .= "$ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"}; } @@ -725,6 +705,9 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++) delete $Jobstep->{stderr}; delete $Jobstep->{finishtime}; + $Jobstep->{'arvados_task'}->{started_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{starttime}); + $Jobstep->{'arvados_task'}->save; + splice @jobstep_todo, $todo_ptr, 1; --$todo_ptr; @@ -840,12 +823,6 @@ release_allocation(); freeze(); my $collated_output = &collate_output(); -if ($job_has_uuid) { - $Job->update_attributes('running' => 0, - 'success' => $collated_output && $main::success, - 'finished_at' => scalar gmtime) -} - if (!$collated_output) { Log(undef, "output undef"); } @@ -860,7 +837,8 @@ else { my $output = $arv->{'collections'}->{'create'}->execute('collection' => { 'manifest_text' => $orig_manifest_text, }); - Log(undef, "output " . $output->{portable_data_hash}); + Log(undef, "output uuid " . $output->{uuid}); + Log(undef, "output hash " . $output->{portable_data_hash}); $Job->update_attributes('output' => $output->{portable_data_hash}) if $job_has_uuid; }; if ($@) { @@ -871,7 +849,16 @@ else { Log (undef, "finish"); save_meta(); -exit ($Job->{'success'} ? 1 : 0); + +if ($job_has_uuid) { + if ($collated_output && $main::success) { + $Job->update_attributes('state' => 'Complete') + } else { + $Job->update_attributes('state' => 'Failed') + } +} + +exit ($Job->{'state'} != 'Complete' ? 1 : 0); @@ -974,6 +961,8 @@ sub reapchildren } $Jobstep->{exitcode} = $childstatus; $Jobstep->{finishtime} = time; + $Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime}); + $Jobstep->{'arvados_task'}->save; process_stderr ($jobstepid, $task_success); Log ($jobstepid, "output " . $Jobstep->{'arvados_task'}->{output}); @@ -1021,12 +1010,16 @@ sub check_refresh_wanted my $Job2 = $arv->{'jobs'}->{'get'}->execute('uuid' => $jobspec); for my $attr ('cancelled_at', 'cancelled_by_user_uuid', - 'cancelled_by_client_uuid') { + 'cancelled_by_client_uuid', + 'state') { $Job->{$attr} = $Job2->{$attr}; } - if ($Job->{'cancelled_at'}) { - Log (undef, "Job cancelled at " . $Job->{cancelled_at} . - " by user " . $Job->{cancelled_by_user_uuid}); + if ($Job->{'state'} ne "Running") { + if ($Job->{'state'} eq "Cancelled") { + Log (undef, "Job cancelled at " . $Job->{'cancelled_at'} . " by user " . $Job->{'cancelled_by_user_uuid'}); + } else { + Log (undef, "Job state unexpectedly changed to " . $Job->{'state'}); + } $main::success = 0; $main::please_freeze = 1; } @@ -1195,7 +1188,8 @@ sub collate_output Log (undef, "collate"); my ($child_out, $child_in); - my $pid = open2($child_out, $child_in, 'arv-put', '--raw'); + my $pid = open2($child_out, $child_in, 'arv-put', '--raw', + '--retries', put_retry_count()); my $joboutput; for (@jobstep) { @@ -1323,9 +1317,11 @@ sub croak sub cleanup { return if !$job_has_uuid; - $Job->update_attributes('running' => 0, - 'success' => 0, - 'finished_at' => scalar gmtime); + if ($Job->{'state'} eq 'Cancelled') { + $Job->update_attributes('finished_at' => scalar gmtime); + } else { + $Job->update_attributes('state' => 'Failed'); + } } @@ -1335,8 +1331,9 @@ sub save_meta return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm $local_logfile->flush; - my $cmd = "arv-put --filename ''\Q$keep_logfile\E " - . quotemeta($local_logfile->filename); + my $retry_count = put_retry_count(); + my $cmd = "arv-put --portable-data-hash --retries $retry_count " . + "--filename ''\Q$keep_logfile\E " . quotemeta($local_logfile->filename); my $loglocator = `$cmd`; die "system $cmd failed: $?" if $?; chomp($loglocator); @@ -1460,14 +1457,40 @@ sub find_docker_image { # If so, return its stream name and Docker hash. # If not, return undef for both values. my $locator = shift; + my ($streamname, $filename); if (my $image = $arv->{collections}->{get}->execute(uuid => $locator)) { - my @file_list = @{$image->{files}}; - if ((scalar(@file_list) == 1) && - ($file_list[0][1] =~ /^([0-9A-Fa-f]{64})\.tar$/)) { - return ($file_list[0][0], $1); + foreach my $line (split(/\n/, $image->{manifest_text})) { + my @tokens = split(/\s+/, $line); + next if (!@tokens); + $streamname = shift(@tokens); + foreach my $filedata (grep(/^\d+:\d+:/, @tokens)) { + if (defined($filename)) { + return (undef, undef); # More than one file in the Collection. + } else { + $filename = (split(/:/, $filedata, 3))[2]; + } + } } } - return (undef, undef); + if (defined($filename) and ($filename =~ /^([0-9A-Fa-f]{64})\.tar$/)) { + return ($streamname, $1); + } else { + return (undef, undef); + } +} + +sub put_retry_count { + # Calculate a --retries argument for arv-put that will have it try + # approximately as long as this Job has been running. + my $stoptime = shift || time; + my $starttime = $jobstep[0]->{starttime}; + my $timediff = defined($starttime) ? ($stoptime - $starttime) : 1; + my $retries = 0; + while ($timediff >= 2) { + $retries++; + $timediff /= 2; + } + return ($retries > 3) ? $retries : 3; } __DATA__ @@ -1493,8 +1516,12 @@ for my $dir ($destdir, $task_work) { open L, ">", "$destdir.lock" or die "$destdir.lock: $!"; flock L, LOCK_EX; if (readlink ("$destdir.commit") eq $commit && -d $destdir) { - exec(@ARGV) if @ARGV; - exit 0; + if (@ARGV) { + exec(@ARGV); + die "Cannot exec `@ARGV`: $!"; + } else { + exit 0; + } } unlink "$destdir.commit"; @@ -1542,6 +1569,7 @@ close L; if (@ARGV) { exec(@ARGV); + die "Cannot exec `@ARGV`: $!"; } else { exit 0; }