use POSIX qw(strftime);
use Fcntl qw(F_GETFL F_SETFL O_NONBLOCK);
use Arvados;
+use Cwd qw(realpath);
use Data::Dumper;
use Digest::MD5 qw(md5_hex);
use Getopt::Long;
use Fcntl ':flock';
use File::Path qw( make_path remove_tree );
+use constant TASK_TEMPFAIL => 111;
use constant EX_TEMPFAIL => 75;
$ENV{"TMPDIR"} ||= "/tmp";
$Job->{'runtime_constraints'}->{'max_tasks_per_node'} ||= 0;
my $max_ncpus = $Job->{'runtime_constraints'}->{'max_tasks_per_node'};
+my $gem_versions = `gem list --quiet arvados-cli 2>/dev/null`;
+if ($? == 0) {
+ $gem_versions =~ s/^arvados-cli \(/ with arvados-cli Gem version(s) /;
+ chomp($gem_versions);
+ chop($gem_versions); # Closing parentheses
+} else {
+ $gem_versions = "";
+}
+Log(undef,
+ "running from " . ((-e $0) ? realpath($0) : "stdin") . $gem_versions);
Log (undef, "check slurm allocation");
my @slot;
must_lock_now("$ENV{CRUNCH_TMP}/.lock", "a job is already running here.");
}
-
-my $build_script;
-do {
- local $/ = undef;
- $build_script = <DATA>;
-};
+my $build_script = handle_readall(\*DATA);
my $nodelist = join(",", @node);
+my $git_tar_count = 0;
if (!defined $no_clear_tmp) {
# Clean out crunch_tmp/work, crunch_tmp/opt, crunch_tmp/src*
my $cleanpid = fork();
if ($cleanpid == 0)
{
+ # Find FUSE mounts that look like Keep mounts (the mount path has the
+ # word "keep") and unmount them. Then clean up work directories.
+ # TODO: When #5036 is done and widely deployed, we can get rid of the
+ # regular expression and just unmount everything with type fuse.keep.
srun (["srun", "--nodelist=$nodelist", "-D", $ENV{'TMPDIR'}],
- ['bash', '-c', 'if mount | grep -q $JOB_WORK/; then for i in $JOB_WORK/*keep $CRUNCH_TMP/task/*.keep; do /bin/fusermount -z -u $i; done; fi; sleep 1; rm -rf $JOB_WORK $CRUNCH_INSTALL $CRUNCH_TMP/task $CRUNCH_TMP/src*']);
+ ['bash', '-ec', 'mount -t fuse,fuse.keep | awk \'($3 ~ /\ykeep\y/){print $3}\' | xargs -r -n 1 fusermount -u -z; sleep 1; rm -rf $JOB_WORK $CRUNCH_INSTALL $CRUNCH_TMP/task $CRUNCH_TMP/src* $CRUNCH_TMP/*.cid']);
exit (1);
}
while (1)
Log (undef, "Cleanup command exited ".exit_status_s($?));
}
+# If this job requires a Docker image, install that.
+my $docker_bin = "/usr/bin/docker.io";
+my ($docker_locator, $docker_stream, $docker_hash, $docker_limitmem);
+if ($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 ! $docker_bin images -q --no-trunc --all | grep -qxF \Q$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 $docker_locator exited "
+ .exit_status_s($?));
+ }
+
+ # Determine whether this version of Docker supports memory+swap limits.
+ srun(["srun", "--nodelist=" . $node[0]],
+ ["/bin/sh", "-ec", "$docker_bin run --help | grep -qe --memory-swap="],
+ {fork => 1});
+ $docker_limitmem = ($? == 0);
+
+ if ($Job->{arvados_sdk_version}) {
+ # The job also specifies an Arvados SDK version. Add the SDKs to the
+ # tar file for the build script to install.
+ Log(undef, sprintf("Packing Arvados SDK version %s for installation",
+ $Job->{arvados_sdk_version}));
+ add_git_archive("git", "--git-dir=$git_dir", "archive",
+ "--prefix=.arvados.sdk/",
+ $Job->{arvados_sdk_version}, "sdk");
+ }
+}
-my $git_archive;
if (!defined $git_dir && $Job->{'script_version'} =~ m{^/}) {
# If script_version looks like an absolute path, *and* the --git-dir
# argument was not given -- which implies we were not invoked by
}
$ENV{"CRUNCH_SRC_COMMIT"} = $commit;
- $git_archive = `$gitcmd archive ''\Q$commit\E`;
- if ($?) {
- croak("Error: $gitcmd archive exited ".exit_status_s($?));
- }
+ add_git_archive("$gitcmd archive ''\Q$commit\E");
}
+my $git_archive = combined_git_archive();
if (!defined $git_archive) {
Log(undef, "Skip install phase (no git archive)");
if ($have_slurm) {
}
my $install_exited = $?;
Log (undef, "Install script exited ".exit_status_s($install_exited));
- exit (1) if $install_exited != 0;
-}
-
-if (!$have_slurm)
-{
- # Grab our lock again (we might have deleted and re-created CRUNCH_TMP above)
- 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 $docker_locator exited "
- .exit_status_s($?));
+ foreach my $tar_filename (map { tar_filename_n($_); } (1..$git_tar_count)) {
+ unlink($tar_filename);
}
+ exit (1) if $install_exited != 0;
}
foreach (qw (script script_version script_parameters runtime_constraints))
THISROUND:
+my $tasks_this_level = 0;
+foreach my $id (@jobstep_todo) {
+ $tasks_this_level++ if ($jobstep[$id]->{level} == $level);
+}
for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
{
my $id = $jobstep_todo[$todo_ptr];
my $childslotname = join (".",
$slot[$childslot]->{node}->{name},
$slot[$childslot]->{cpu});
+
my $childpid = fork();
if ($childpid == 0)
{
$ENV{"GZIP"} = "-n";
+ my $max_node_concurrent_tasks = $ENV{CRUNCH_NODE_SLOTS};
+ if ($tasks_this_level < $max_node_concurrent_tasks) {
+ $max_node_concurrent_tasks = $tasks_this_level;
+ }
+
my @srunargs = (
"srun",
"--nodelist=".$childnode->{name},
my $command =
"if [ -e $ENV{TASK_WORK} ]; then rm -rf $ENV{TASK_WORK}; fi; "
."mkdir -p $ENV{CRUNCH_TMP} $ENV{JOB_WORK} $ENV{TASK_WORK} $ENV{TASK_KEEPMOUNT} "
- ."&& cd $ENV{CRUNCH_TMP} ";
+ ."&& cd $ENV{CRUNCH_TMP} "
+ # These environment variables get used explicitly later in
+ # $command. No tool is expected to read these values directly.
+ .q{&& MEM=$(awk '($1 == "MemTotal:"){print $2}' </proc/meminfo) }
+ .q{&& SWAP=$(awk '($1 == "SwapTotal:"){print $2}' </proc/meminfo) }
+ ."&& MEMLIMIT=\$(( (\$MEM * 95) / ($max_node_concurrent_tasks * 100) )) "
+ ."&& let SWAPLIMIT=\$MEMLIMIT+\$SWAP ";
$command .= "&& exec arv-mount --by-id --allow-other $ENV{TASK_KEEPMOUNT} --exec ";
if ($docker_hash)
{
- my $cidfile = "$ENV{CRUNCH_TMP}/$ENV{TASK_UUID}.cid";
+ my $cidfile = "$ENV{CRUNCH_TMP}/$Jobstep->{arvados_task}->{uuid}-$Jobstep->{failures}.cid";
$command .= "crunchstat -cgroup-root=/sys/fs/cgroup -cgroup-parent=docker -cgroup-cid=$cidfile -poll=10000 ";
$command .= "$docker_bin run --rm=true --attach=stdout --attach=stderr --attach=stdin -i --user=crunch --cidfile=$cidfile --sig-proxy ";
+ # We only set memory limits if Docker lets us limit both memory and swap.
+ # Memory limits alone have been supported longer, but subprocesses tend
+ # to get SIGKILL if they exceed that without any swap limit set.
+ # See #5642 for additional background.
+ if ($docker_limitmem) {
+ $command .= "--memory=\${MEMLIMIT}k --memory-swap=\${SWAPLIMIT}k ";
+ }
# Dynamically configure the container to use the host system as its
# DNS server. Get the host's global addresses from the ip command,
||
(@slot > @freeslot && $todo_ptr+1 > $#jobstep_todo))
{
- last THISROUND if $main::please_freeze;
+ last THISROUND if $main::please_freeze || defined($main::success);
if ($main::please_info)
{
$main::please_info = 0;
Log (undef, "Failed to write output collection");
}
else {
- Log(undef, "output hash " . $collated_output);
+ Log(undef, "job output $collated_output");
$Job->update_attributes('output' => $collated_output);
}
{
my $temporary_fail;
$temporary_fail ||= $Jobstep->{node_fail};
- $temporary_fail ||= ($exitvalue == 111);
+ $temporary_fail ||= ($exitvalue == TASK_TEMPFAIL);
++$thisround_failed;
++$thisround_failed_multiple if $Jobstep->{'failures'} >= 1;
if (!$temporary_fail || $Jobstep->{'failures'} >= 3) {
# Give up on this task, and the whole job
$main::success = 0;
- $main::please_freeze = 1;
}
# Put this task back on the todo queue
push @jobstep_todo, $jobstepid;
$Jobstep->{'arvados_task'}->{finished_at} = strftime "%Y-%m-%dT%H:%M:%SZ", gmtime($Jobstep->{finishtime});
$Jobstep->{'arvados_task'}->save;
process_stderr ($jobstepid, $task_success);
- Log ($jobstepid, "output " . $Jobstep->{'arvados_task'}->{output});
+ Log ($jobstepid, sprintf("task output (%d bytes): %s",
+ length($Jobstep->{'arvados_task'}->{output}),
+ $Jobstep->{'arvados_task'}->{output}));
close $reader{$jobstepid};
delete $reader{$jobstepid};
# whoa.
$main::please_freeze = 1;
}
- elsif ($line =~ /srun: error: (Node failure on|Unable to create job step) /) {
+ elsif ($line =~ /(srun: error: (Node failure on|Unable to create job step|.*: Communication connection failure))|arvados.errors.Keep/) {
$jobstep[$job]->{node_fail} = 1;
ban_node_by_slot($jobstep[$job]->{slotindex});
}
sub fetch_block
{
my $hash = shift;
- my ($keep, $child_out, $output_block);
-
- my $cmd = "arv-get \Q$hash\E";
- open($keep, '-|', $cmd) or die "fetch_block: $cmd: $!";
- $output_block = '';
+ my $keep;
+ if (!open($keep, "-|", "arv-get", "--retries", retry_count(), $hash)) {
+ Log(undef, "fetch_block run error from arv-get $hash: $!");
+ return undef;
+ }
+ my $output_block = "";
while (1) {
my $buf;
my $bytes = sysread($keep, $buf, 1024 * 1024);
if (!defined $bytes) {
- die "reading from arv-get: $!";
+ Log(undef, "fetch_block read error from arv-get: $!");
+ $output_block = undef;
+ last;
} elsif ($bytes == 0) {
# sysread returns 0 at the end of the pipe.
last;
}
}
close $keep;
+ if ($?) {
+ Log(undef, "fetch_block arv-get exited " . exit_status_s($?));
+ $output_block = undef;
+ }
return $output_block;
}
-# create_output_collections generates a new collection containing the
-# output of each successfully completed task, and returns the
-# portable_data_hash for the new collection.
-#
+# Create a collection by concatenating the output of all tasks (each
+# task's output is either a manifest fragment, a locator for a
+# manifest fragment stored in Keep, or nothing at all). Return the
+# portable_data_hash of the new collection.
sub create_output_collection
{
Log (undef, "collate");
my ($child_out, $child_in);
- my $pid = open2($child_out, $child_in, 'python', '-c',
- 'import arvados; ' .
- 'import sys; ' .
- 'print arvados.api()' .
- '.collections()' .
- '.create(body={"manifest_text":sys.stdin.read()})' .
- '.execute()["portable_data_hash"]'
- );
-
+ my $pid = open2($child_out, $child_in, 'python', '-c', q{
+import arvados
+import sys
+print (arvados.api("v1").collections().
+ create(body={"manifest_text": sys.stdin.read()}).
+ execute(num_retries=int(sys.argv[1]))["portable_data_hash"])
+}, retry_count());
+
+ my $task_idx = -1;
+ my $manifest_size = 0;
for (@jobstep)
{
- next if (!exists $_->{'arvados_task'}->{'output'} ||
- !$_->{'arvados_task'}->{'success'});
+ ++$task_idx;
my $output = $_->{'arvados_task'}->{output};
- if ($output !~ /^[0-9a-f]{32}(\+\S+)*$/)
- {
- print $child_in $output;
- }
- elsif (defined (my $outblock = fetch_block ($output)))
- {
- print $child_in $outblock;
+ next if (!defined($output));
+ my $next_write;
+ if ($output =~ /^[0-9a-f]{32}(\+\S+)*$/) {
+ $next_write = fetch_block($output);
+ } else {
+ $next_write = $output;
}
- else
- {
- Log (undef, "XXX fetch_block($output) failed XXX");
+ if (defined($next_write)) {
+ if (!defined(syswrite($child_in, $next_write))) {
+ # There's been an error writing. Stop the loop.
+ # We'll log details about the exit code later.
+ last;
+ } else {
+ $manifest_size += length($next_write);
+ }
+ } else {
+ my $uuid = $_->{'arvados_task'}->{'uuid'};
+ Log (undef, "Error retrieving '$output' output by task $task_idx ($uuid)");
$main::success = 0;
}
}
- $child_in->close;
+ close($child_in);
+ Log(undef, "collated output manifest text to send to API server is $manifest_size bytes with access tokens");
my $joboutput;
my $s = IO::Select->new($child_out);
if ($s->can_read(120)) {
- sysread($child_out, $joboutput, 64 * 1024 * 1024);
- chomp($joboutput);
- # TODO: Ensure exit status == 0.
+ sysread($child_out, $joboutput, 1024 * 1024);
+ waitpid($pid, 0);
+ if ($?) {
+ Log(undef, "output collection creation exited " . exit_status_s($?));
+ $joboutput = undef;
+ } else {
+ chomp($joboutput);
+ }
} else {
Log (undef, "timed out while creating output collection");
+ foreach my $signal (2, 2, 2, 15, 15, 9) {
+ kill($signal, $pid);
+ last if waitpid($pid, WNOHANG) == -1;
+ sleep(1);
+ }
}
- # TODO: kill $pid instead of waiting, now that we've decided to
- # ignore further output.
- waitpid($pid, 0);
+ close($child_out);
return $joboutput;
}
{
my $logfilename = shift;
$log_pipe_pid = open2($log_pipe_out, $log_pipe_in,
- 'arv-put', '--portable-data-hash',
+ 'arv-put',
+ '--portable-data-hash',
+ '--project-uuid', $Job->{owner_uuid},
'--retries', '3',
+ '--name', $logfilename,
'--filename', $logfilename,
'-');
}
return $s;
}
+sub handle_readall {
+ # Pass in a glob reference to a file handle.
+ # Read all its contents and return them as a string.
+ my $fh_glob_ref = shift;
+ local $/ = undef;
+ return <$fh_glob_ref>;
+}
+
+sub tar_filename_n {
+ my $n = shift;
+ return sprintf("%s/git.%s.%d.tar", $ENV{CRUNCH_TMP}, $job_id, $n);
+}
+
+sub add_git_archive {
+ # Pass in a git archive command as a string or list, a la system().
+ # This method will save its output to be included in the archive sent to the
+ # build script.
+ my $git_input;
+ $git_tar_count++;
+ if (!open(GIT_ARCHIVE, ">", tar_filename_n($git_tar_count))) {
+ croak("Failed to save git archive: $!");
+ }
+ my $git_pid = open2(">&GIT_ARCHIVE", $git_input, @_);
+ close($git_input);
+ waitpid($git_pid, 0);
+ close(GIT_ARCHIVE);
+ if ($?) {
+ croak("Failed to save git archive: git exited " . exit_status_s($?));
+ }
+}
+
+sub combined_git_archive {
+ # Combine all saved tar archives into a single archive, then return its
+ # contents in a string. Return undef if no archives have been saved.
+ if ($git_tar_count < 1) {
+ return undef;
+ }
+ my $base_tar_name = tar_filename_n(1);
+ foreach my $tar_to_append (map { tar_filename_n($_); } (2..$git_tar_count)) {
+ my $tar_exit = system("tar", "-Af", $base_tar_name, $tar_to_append);
+ if ($tar_exit != 0) {
+ croak("Error preparing build archive: tar -A exited " .
+ exit_status_s($tar_exit));
+ }
+ }
+ if (!open(GIT_TAR, "<", $base_tar_name)) {
+ croak("Could not open build archive: $!");
+ }
+ my $tar_contents = handle_readall(\*GIT_TAR);
+ close(GIT_TAR);
+ return $tar_contents;
+}
+
__DATA__
#!/usr/bin/perl
-
-# checkout-and-build
+#
+# This is crunch-job's internal dispatch script. crunch-job running on the API
+# server invokes this script on individual compute nodes, or localhost if we're
+# running a job locally. It gets called in two modes:
+#
+# * No arguments: Installation mode. Read a tar archive from the DATA
+# file handle; it includes the Crunch script's source code, and
+# maybe SDKs as well. Those should be installed in the proper
+# locations. This runs outside of any Docker container, so don't try to
+# introspect Crunch's runtime environment.
+#
+# * With arguments: Crunch script run mode. This script should set up the
+# environment, then run the command specified in the arguments. This runs
+# inside any Docker container.
use Fcntl ':flock';
use File::Path qw( make_path remove_tree );
+use POSIX qw(getcwd);
+
+use constant TASK_TEMPFAIL => 111;
+
+# Map SDK subdirectories to the path environments they belong to.
+my %SDK_ENVVARS = ("perl/lib" => "PERLLIB", "ruby/lib" => "RUBYLIB");
my $destdir = $ENV{"CRUNCH_SRC"};
my $commit = $ENV{"CRUNCH_SRC_COMMIT"};
my $repo = $ENV{"CRUNCH_SRC_URL"};
+my $install_dir = $ENV{"CRUNCH_INSTALL"} || (getcwd() . "/opt");
my $job_work = $ENV{"JOB_WORK"};
my $task_work = $ENV{"TASK_WORK"};
remove_tree($task_work, {keep_root => 1});
}
-my @git_archive_data = <DATA>;
-if (!@git_archive_data) {
- # Nothing to extract -> nothing to install.
- run_argv_and_exit();
+open(STDOUT_ORIG, ">&", STDOUT);
+open(STDERR_ORIG, ">&", STDERR);
+open(STDOUT, ">>", "$destdir.log");
+open(STDERR, ">&", STDOUT);
+
+### Crunch script run mode
+if (@ARGV) {
+ # We want to do routine logging during task 0 only. This gives the user
+ # the information they need, but avoids repeating the information for every
+ # task.
+ my $Log;
+ if ($ENV{TASK_SEQUENCE} eq "0") {
+ $Log = sub {
+ my $msg = shift;
+ printf STDERR_ORIG "[Crunch] $msg\n", @_;
+ };
+ } else {
+ $Log = sub { };
+ }
+
+ my $python_src = "$install_dir/python";
+ my $venv_dir = "$job_work/.arvados.venv";
+ my $venv_built = -e "$venv_dir/bin/activate";
+ if ((!$venv_built) and (-d $python_src) and can_run("virtualenv")) {
+ shell_or_die(undef, "virtualenv", "--quiet", "--system-site-packages",
+ "--python=python2.7", $venv_dir);
+ shell_or_die(TASK_TEMPFAIL, "$venv_dir/bin/pip", "--quiet", "install", "-I", $python_src);
+ $venv_built = 1;
+ $Log->("Built Python SDK virtualenv");
+ }
+
+ my $pip_bin = "pip";
+ if ($venv_built) {
+ $Log->("Running in Python SDK virtualenv");
+ $pip_bin = "$venv_dir/bin/pip";
+ my $orig_argv = join(" ", map { quotemeta($_); } @ARGV);
+ @ARGV = ("/bin/sh", "-ec",
+ ". \Q$venv_dir/bin/activate\E; exec $orig_argv");
+ } elsif (-d $python_src) {
+ $Log->("Warning: virtualenv not found inside Docker container default " .
+ "\$PATH. Can't install Python SDK.");
+ }
+
+ my $pkgs = `(\Q$pip_bin\E freeze 2>/dev/null | grep arvados) || dpkg-query --show '*arvados*'`;
+ if ($pkgs) {
+ $Log->("Using Arvados SDK:");
+ foreach my $line (split /\n/, $pkgs) {
+ $Log->($line);
+ }
+ } else {
+ $Log->("Arvados SDK packages not found");
+ }
+
+ while (my ($sdk_dir, $sdk_envkey) = each(%SDK_ENVVARS)) {
+ my $sdk_path = "$install_dir/$sdk_dir";
+ if (-d $sdk_path) {
+ if ($ENV{$sdk_envkey}) {
+ $ENV{$sdk_envkey} = "$sdk_path:" . $ENV{$sdk_envkey};
+ } else {
+ $ENV{$sdk_envkey} = $sdk_path;
+ }
+ $Log->("Arvados SDK added to %s", $sdk_envkey);
+ }
+ }
+
+ close(STDOUT);
+ close(STDERR);
+ open(STDOUT, ">&", STDOUT_ORIG);
+ open(STDERR, ">&", STDERR_ORIG);
+ exec(@ARGV);
+ die "Cannot exec `@ARGV`: $!";
}
+### Installation mode
open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
flock L, LOCK_EX;
if (readlink ("$destdir.commit") eq $commit && -d $destdir) {
# This version already installed -> nothing to do.
- run_argv_and_exit();
+ exit(0);
}
unlink "$destdir.commit";
-open STDERR_ORIG, ">&STDERR";
-open STDOUT, ">", "$destdir.log";
-open STDERR, ">&STDOUT";
-
mkdir $destdir;
-open TARX, "|-", "tar", "-C", $destdir, "-xf", "-";
-print TARX @git_archive_data;
+
+if (!open(TARX, "|-", "tar", "-xC", $destdir)) {
+ die "Error launching 'tar -xC $destdir': $!";
+}
+# If we send too much data to tar in one write (> 4-5 MiB), it stops, and we
+# get SIGPIPE. We must feed it data incrementally.
+my $tar_input;
+while (read(DATA, $tar_input, 65536)) {
+ print TARX $tar_input;
+}
if(!close(TARX)) {
- die "'tar -C $destdir -xf -' exited $?: $!";
+ die "'tar -xC $destdir' exited $?: $!";
}
-my $pwd;
-chomp ($pwd = `pwd`);
-my $install_dir = $ENV{"CRUNCH_INSTALL"} || "$pwd/opt";
mkdir $install_dir;
-for my $src_path ("$destdir/arvados/sdk/python") {
- if (-d $src_path) {
- shell_or_die ("virtualenv", $install_dir);
- shell_or_die ("cd $src_path && ./build.sh && $install_dir/bin/python setup.py install");
+my $sdk_root = "$destdir/.arvados.sdk/sdk";
+if (-d $sdk_root) {
+ foreach my $sdk_lang (("python",
+ map { (split /\//, $_, 2)[0]; } keys(%SDK_ENVVARS))) {
+ if (-d "$sdk_root/$sdk_lang") {
+ if (!rename("$sdk_root/$sdk_lang", "$install_dir/$sdk_lang")) {
+ die "Failed to install $sdk_lang SDK: $!";
+ }
+ }
}
}
+my $python_dir = "$install_dir/python";
+if ((-d $python_dir) and can_run("python2.7") and
+ (system("python2.7", "$python_dir/setup.py", "--quiet", "egg_info") != 0)) {
+ # egg_info failed, probably when it asked git for a build tag.
+ # Specify no build tag.
+ open(my $pysdk_cfg, ">>", "$python_dir/setup.cfg");
+ print $pysdk_cfg "\n[egg_info]\ntag_build =\n";
+ close($pysdk_cfg);
+}
+
if (-e "$destdir/crunch_scripts/install") {
- shell_or_die ("$destdir/crunch_scripts/install", $install_dir);
+ shell_or_die (undef, "$destdir/crunch_scripts/install", $install_dir);
} elsif (!-e "./install.sh" && -e "./tests/autotests.sh") {
# Old version
- shell_or_die ("./tests/autotests.sh", $install_dir);
+ shell_or_die (undef, "./tests/autotests.sh", $install_dir);
} elsif (-e "./install.sh") {
- shell_or_die ("./install.sh", $install_dir);
+ shell_or_die (undef, "./install.sh", $install_dir);
}
if ($commit) {
close L;
-run_argv_and_exit();
-
-sub run_argv_and_exit
-{
- if (@ARGV) {
- exec(@ARGV);
- die "Cannot exec `@ARGV`: $!";
- } else {
- exit 0;
- }
+sub can_run {
+ my $command_name = shift;
+ open(my $which, "-|", "which", $command_name);
+ while (<$which>) { }
+ close($which);
+ return ($? == 0);
}
sub shell_or_die
{
+ my $exitcode = shift;
+
if ($ENV{"DEBUG"}) {
print STDERR "@_\n";
}
if (system (@_) != 0) {
my $err = $!;
- my $exitstatus = sprintf("exit %d signal %d", $? >> 8, $? & 0x7f);
+ my $code = $?;
+ my $exitstatus = sprintf("exit %d signal %d", $code >> 8, $code & 0x7f);
open STDERR, ">&STDERR_ORIG";
system ("cat $destdir.log >&2");
- die "@_ failed ($err): $exitstatus";
+ warn "@_ failed ($err): $exitstatus";
+ if (defined($exitcode)) {
+ exit $exitcode;
+ }
+ else {
+ exit (($code >> 8) || 1);
+ }
}
}
projects / arvados.git / blobdiff