X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/30b02581c938c05b804f7510a1fc8e850830b9cd..90fc79852a995fd8e665cf48ae20c49a9bbc78eb:/sdk/cli/bin/crunch-job diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job index 48a6c9dea7..0ba32b0588 100755 --- a/sdk/cli/bin/crunch-job +++ b/sdk/cli/bin/crunch-job @@ -76,11 +76,15 @@ use strict; use POSIX ':sys_wait_h'; use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK); use Arvados; +use Digest::MD5 qw(md5_hex); use Getopt::Long; use IPC::Open2; use IO::Select; use File::Temp; use Fcntl ':flock'; +use File::Path qw( make_path ); + +use constant EX_TEMPFAIL => 75; $ENV{"TMPDIR"} ||= "/tmp"; unless (defined $ENV{"CRUNCH_TMP"}) { @@ -90,20 +94,17 @@ unless (defined $ENV{"CRUNCH_TMP"}) { $ENV{"CRUNCH_TMP"} .= "-$<"; } } + +# Create the tmp directory if it does not exist +if ( ! -d $ENV{"CRUNCH_TMP"} ) { + make_path $ENV{"CRUNCH_TMP"} or die "Failed to create temporary working directory: " . $ENV{"CRUNCH_TMP"}; +} + $ENV{"JOB_WORK"} = $ENV{"CRUNCH_TMP"} . "/work"; $ENV{"CRUNCH_INSTALL"} = "$ENV{CRUNCH_TMP}/opt"; $ENV{"CRUNCH_WORK"} = $ENV{"JOB_WORK"}; # deprecated mkdir ($ENV{"JOB_WORK"}); -my $arv_cli; - -if (defined $ENV{"ARV_CLI"}) { - $arv_cli = $ENV{"ARV_CLI"}; -} -else { - $arv_cli = 'arv'; -} - my $force_unlock; my $git_dir; my $jobspec; @@ -139,7 +140,7 @@ $SIG{'USR2'} = sub my $arv = Arvados->new('apiVersion' => 'v1'); -my $metastream; +my $local_logfile; my $User = $arv->{'users'}->{'current'}->execute; @@ -151,17 +152,25 @@ if ($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'}) { - croak("Job is locked: " . $Job->{'is_locked_by_uuid'}); + Log(undef, "Job is locked by " . $Job->{'is_locked_by_uuid'}); + exit EX_TEMPFAIL; } if ($Job->{'success'} ne undef) { - croak("Job 'success' flag (" . $Job->{'success'} . ") is not null"); + Log(undef, "Job 'success' flag (" . $Job->{'success'} . ") is not null"); + exit EX_TEMPFAIL; } if ($Job->{'running'}) { - croak("Job 'running' flag is already set"); + Log(undef, "Job 'running' flag is already set"); + exit EX_TEMPFAIL; } if ($Job->{'started_at'}) { - croak("Job 'started_at' time is already set (" . $Job->{'started_at'} . ")"); + Log(undef, "Job 'started_at' time is already set (" . $Job->{'started_at'} . ")"); + exit EX_TEMPFAIL; } } } @@ -185,7 +194,7 @@ else $job_id = $Job->{'uuid'}; my $keep_logfile = $job_id . '.log.txt'; -my $local_logfile = File::Temp->new(); +$local_logfile = File::Temp->new(); $Job->{'runtime_constraints'} ||= {}; $Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0; @@ -274,7 +283,8 @@ if ($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'}) { - croak("Error while updating / locking job"); + Log(undef, "Error while updating / locking job, exiting ".EX_TEMPFAIL); + exit EX_TEMPFAIL; } $Job->update_attributes('started_at' => scalar gmtime, 'running' => 1, @@ -384,7 +394,7 @@ else if ($cleanpid == 0) { srun (["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}], - ['bash', '-c', 'if mount | grep -q $JOB_WORK/; then sudo /bin/umount $JOB_WORK/* 2>/dev/null; fi; sleep 1; rm -rf $JOB_WORK $CRUNCH_TMP/opt $CRUNCH_TMP/src*']); + ['bash', '-c', 'if mount | grep -q $JOB_WORK/; then for i in $JOB_WORK/*keep; do /bin/fusermount -z -u $i; done; fi; sleep 1; rm -rf $JOB_WORK $CRUNCH_TMP/opt $CRUNCH_TMP/src*']); exit (1); } while (1) @@ -477,6 +487,8 @@ else croak ("could not figure out commit id for $treeish"); } + # Note: this section is almost certainly unnecessary if we're + # running tasks in docker containers. my $installpid = fork(); if ($installpid == 0) { @@ -498,7 +510,39 @@ if (!$have_slurm) must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here."); } - +# If this job requires a Docker image, install that. +my $docker_bin = "/usr/bin/docker.io"; +my ($docker_locator, $docker_stream, $docker_hash); +if ($docker_locator = $Job->{docker_image_locator}) { + ($docker_stream, $docker_hash) = find_docker_image($docker_locator); + if (!$docker_hash) + { + croak("No Docker image hash found from locator $docker_locator"); + } + $docker_stream =~ s/^\.//; + my $docker_install_script = qq{ +if ! $docker_bin images -q --no-trunc | grep -qxF \Q$docker_hash\E; then + arv-get \Q$docker_locator$docker_stream/$docker_hash.tar\E | $docker_bin load +fi +}; + my $docker_pid = fork(); + if ($docker_pid == 0) + { + srun (["srun", "--nodelist=" . join(',', @node)], + ["/bin/sh", "-ec", $docker_install_script]); + exit ($?); + } + while (1) + { + last if $docker_pid == waitpid (-1, WNOHANG); + freeze_if_want_freeze ($docker_pid); + select (undef, undef, undef, 0.1); + } + if ($? != 0) + { + croak("Installing Docker image from $docker_locator returned exit code $?"); + } +} foreach (qw (script script_version script_parameters runtime_constraints)) { @@ -590,6 +634,7 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++) $ENV{"TASK_SLOT_NODE"} = $slot[$childslot]->{node}->{name}; $ENV{"TASK_SLOT_NUMBER"} = $slot[$childslot]->{cpu}; $ENV{"TASK_WORK"} = $ENV{"JOB_WORK"}."/$id.$$"; + $ENV{"HOME"} = $ENV{"TASK_WORK"}; $ENV{"TASK_KEEPMOUNT"} = $ENV{"TASK_WORK"}.".keep"; $ENV{"TASK_TMPDIR"} = $ENV{"TASK_WORK"}; # deprecated $ENV{"CRUNCH_NODE_SLOTS"} = $slot[$childslot]->{node}->{ncpus}; @@ -603,11 +648,10 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++) qw(-n1 -c1 -N1 -D), $ENV{'TMPDIR'}, "--job-name=$job_id.$id.$$", ); - my @execargs = qw(sh); my $build_script_to_send = ""; my $command = "if [ -e $ENV{TASK_WORK} ]; then rm -rf $ENV{TASK_WORK}; fi; " - ."mkdir -p $ENV{JOB_WORK} $ENV{CRUNCH_TMP} $ENV{TASK_WORK} $ENV{TASK_KEEPMOUNT} " + ."mkdir -p $ENV{CRUNCH_TMP} $ENV{JOB_WORK} $ENV{TASK_WORK} $ENV{TASK_KEEPMOUNT} " ."&& cd $ENV{CRUNCH_TMP} "; if ($build_script) { @@ -615,11 +659,51 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++) $command .= "&& perl -"; } - $command .= - "&& exec arv-mount $ENV{TASK_KEEPMOUNT} --exec $ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"}; + $command .= "&& exec arv-mount --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 "; + $command .= "$docker_bin run --rm=true --attach=stdout --attach=stderr --user=crunch --cidfile=$ENV{TASK_WORK}/docker.cid "; + # Dynamically configure the container to use the host system as its + # DNS server. Get the host's global addresses from the ip command, + # and turn them into docker --dns options using gawk. + $command .= + q{$(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_WORK") { + $command .= "--env=\QTASK_WORK=/tmp/crunch-job\E "; + } + elsif ($env_key eq "TASK_KEEPMOUNT") { + $command .= "--env=\QTASK_KEEPMOUNT=/keep\E "; + } + else { + $command .= "--env=\Q$env_key=$env_val\E "; + } + } + } + $command .= "--env=\QCRUNCH_NODE_SLOTS=$ENV{CRUNCH_NODE_SLOTS}\E "; + $command .= "--env=\QCRUNCH_SRC=/tmp/crunch-src\E "; + $command .= "\Q$docker_hash\E "; + $command .= "stdbuf --output=0 --error=0 "; + $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 "; + $command .= "$ENV{CRUNCH_SRC}/crunch_scripts/" . $Job->{"script"}; + } + my @execargs = ('bash', '-c', $command); srun (\@srunargs, \@execargs, undef, $build_script_to_send); - exit (111); + # exec() failed, we assume nothing happened. + Log(undef, "srun() failed on build script"); + die; } close("writer"); if (!defined $childpid) @@ -758,31 +842,31 @@ goto ONELEVEL if !defined $main::success; release_allocation(); freeze(); +my $collated_output = &collate_output(); + if ($job_has_uuid) { - $Job->update_attributes('output' => &collate_output(), - 'running' => 0, - 'success' => $Job->{'output'} && $main::success, + $Job->update_attributes('running' => 0, + 'success' => $collated_output && $main::success, 'finished_at' => scalar gmtime) } -if ($Job->{'output'}) -{ +if (!$collated_output) { + Log(undef, "output undef"); +} +else { eval { - my $manifest_text = `arv keep get ''\Q$Job->{'output'}\E`; - $arv->{'collections'}->{'create'}->execute('collection' => { - 'uuid' => $Job->{'output'}, - 'manifest_text' => $manifest_text, - }); - if ($Job->{'output_is_persistent'}) { - $arv->{'links'}->{'create'}->execute('link' => { - 'tail_kind' => 'arvados#user', - 'tail_uuid' => $User->{'uuid'}, - 'head_kind' => 'arvados#collection', - 'head_uuid' => $Job->{'output'}, - 'link_class' => 'resources', - 'name' => 'wants', - }); + open(my $orig_manifest, '-|', 'arv-get', $collated_output) + or die "failed to get collated manifest: $!"; + my $orig_manifest_text = ''; + while (my $manifest_line = <$orig_manifest>) { + $orig_manifest_text .= $manifest_line; } + my $output = $arv->{'collections'}->{'create'}->execute('collection' => { + 'manifest_text' => $orig_manifest_text, + }); + Log(undef, "output uuid " . $output->{uuid}); + Log(undef, "output hash " . $output->{portable_data_hash}); + $Job->update_attributes('output' => $output->{portable_data_hash}) if $job_has_uuid; }; if ($@) { Log (undef, "Failed to register output manifest: $@"); @@ -792,7 +876,7 @@ if ($Job->{'output'}) Log (undef, "finish"); save_meta(); -exit 0; +exit ($Job->{'success'} ? 1 : 0); @@ -904,21 +988,29 @@ sub reapchildren push @freeslot, $proc{$pid}->{slot}; delete $proc{$pid}; - # Load new tasks - my $newtask_list = $arv->{'job_tasks'}->{'list'}->execute( - 'where' => { - 'created_by_job_task_uuid' => $Jobstep->{'arvados_task'}->{uuid} - }, - 'order' => 'qsequence' - ); - foreach my $arvados_task (@{$newtask_list->{'items'}}) { - my $jobstep = { - 'level' => $arvados_task->{'sequence'}, - 'failures' => 0, - 'arvados_task' => $arvados_task - }; - push @jobstep, $jobstep; - push @jobstep_todo, $#jobstep; + if ($task_success) { + # Load new tasks + my $newtask_list = []; + my $newtask_results; + do { + $newtask_results = $arv->{'job_tasks'}->{'list'}->execute( + 'where' => { + 'created_by_job_task_uuid' => $Jobstep->{'arvados_task'}->{uuid} + }, + 'order' => 'qsequence', + 'offset' => scalar(@$newtask_list), + ); + push(@$newtask_list, @{$newtask_results->{items}}); + } while (@{$newtask_results->{items}}); + foreach my $arvados_task (@$newtask_list) { + my $jobstep = { + 'level' => $arvados_task->{'sequence'}, + 'failures' => 0, + 'arvados_task' => $arvados_task + }; + push @jobstep, $jobstep; + push @jobstep_todo, $#jobstep; + } } $progress_is_dirty = 1; @@ -1083,9 +1175,22 @@ sub fetch_block my $hash = shift; my ($keep, $child_out, $output_block); - my $cmd = "$arv_cli keep get \Q$hash\E"; + my $cmd = "arv-get \Q$hash\E"; open($keep, '-|', $cmd) or die "fetch_block: $cmd: $!"; - sysread($keep, $output_block, 64 * 1024 * 1024); + $output_block = ''; + while (1) { + my $buf; + my $bytes = sysread($keep, $buf, 1024 * 1024); + if (!defined $bytes) { + die "reading from arv-get: $!"; + } elsif ($bytes == 0) { + # sysread returns 0 at the end of the pipe. + last; + } else { + # some bytes were read into buf. + $output_block .= $buf; + } + } close $keep; return $output_block; } @@ -1095,13 +1200,12 @@ sub collate_output Log (undef, "collate"); my ($child_out, $child_in); - my $pid = open2($child_out, $child_in, $arv_cli, 'keep', 'put', '--raw'); + my $pid = open2($child_out, $child_in, 'arv-put', '--raw'); my $joboutput; for (@jobstep) { - next if (!exists $_->{'arvados_task'}->{output} || - !$_->{'arvados_task'}->{'success'} || - $_->{'exitcode'} != 0); + next if (!exists $_->{'arvados_task'}->{'output'} || + !$_->{'arvados_task'}->{'success'}); my $output = $_->{'arvados_task'}->{output}; if ($output !~ /^[0-9a-f]{32}(\+\S+)*$/) { @@ -1132,20 +1236,11 @@ sub collate_output sysread($child_out, $joboutput, 64 * 1024 * 1024); chomp($joboutput); } else { - Log (undef, "timed out reading from 'arv keep put'"); + Log (undef, "timed out reading from 'arv-put'"); } } waitpid($pid, 0); - if ($joboutput) - { - Log (undef, "output $joboutput"); - $Job->update_attributes('output' => $joboutput) if $job_has_uuid; - } - else - { - Log (undef, "output undef"); - } return $joboutput; } @@ -1204,15 +1299,15 @@ sub Log # ($jobstep_id, $logmessage) $message =~ s{([^ -\176])}{"\\" . sprintf ("%03o", ord($1))}ge; $message .= "\n"; my $datetime; - if ($metastream || -t STDERR) { + if ($local_logfile || -t STDERR) { my @gmtime = gmtime; $datetime = sprintf ("%04d-%02d-%02d_%02d:%02d:%02d", $gmtime[5]+1900, $gmtime[4]+1, @gmtime[3,2,1,0]); } print STDERR ((-t STDERR) ? ($datetime." ".$message) : $message); - if ($metastream) { - print $metastream $datetime . " " . $message; + if ($local_logfile) { + print $local_logfile $datetime . " " . $message; } } @@ -1225,7 +1320,7 @@ sub croak freeze() if @jobstep_todo; collate_output() if @jobstep_todo; cleanup(); - save_meta() if $metastream; + save_meta() if $local_logfile; die; } @@ -1245,10 +1340,11 @@ sub save_meta return if $justcheckpoint; # checkpointing is not relevant post-Warehouse.pm $local_logfile->flush; - my $cmd = "$arv_cli keep put --filename ''\Q$keep_logfile\E " + my $cmd = "arv-put --portable-data-hash --filename ''\Q$keep_logfile\E " . quotemeta($local_logfile->filename); my $loglocator = `$cmd`; die "system $cmd failed: $?" if $?; + chomp($loglocator); $local_logfile = undef; # the temp file is automatically deleted Log (undef, "log manifest is $loglocator"); @@ -1364,21 +1460,62 @@ sub must_lock_now } } +sub find_docker_image { + # Given a Keep locator, check to see if it contains a 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)) { + 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(/:/, $filedata, 3))[2]; + } + } + } + } + if (defined($filename) and ($filename =~ /^([0-9A-Fa-f]{64})\.tar$/)) { + return ($streamname, $1); + } else { + return (undef, undef); + } +} + __DATA__ #!/usr/bin/perl # checkout-and-build use Fcntl ':flock'; +use File::Path qw( make_path ); my $destdir = $ENV{"CRUNCH_SRC"}; my $commit = $ENV{"CRUNCH_SRC_COMMIT"}; my $repo = $ENV{"CRUNCH_SRC_URL"}; +my $task_work = $ENV{"TASK_WORK"}; + +for my $dir ($destdir, $task_work) { + if ($dir) { + make_path $dir; + -e $dir or die "Failed to create temporary directory ($dir): $!"; + } +} open L, ">", "$destdir.lock" or die "$destdir.lock: $!"; flock L, LOCK_EX; if (readlink ("$destdir.commit") eq $commit && -d $destdir) { - exit 0; + if (@ARGV) { + exec(@ARGV); + die "Cannot exec `@ARGV`: $!"; + } else { + exit 0; + } } unlink "$destdir.commit"; @@ -1424,7 +1561,12 @@ if ($commit) { close L; -exit 0; +if (@ARGV) { + exec(@ARGV); + die "Cannot exec `@ARGV`: $!"; +} else { + exit 0; +} sub shell_or_die {