BEX module split

[bex.git] / brun

diff --git a/brun b/brun
index 48afd8c8b5ec77d817a263b83a511fd2b11019e6..760bf9ca172f1f7e8f0bcaf92df7c64109bb0386 100755 (executable)
--- a/brun
+++ b/brun
@@ -9,86 +9,120 @@ use Getopt::Long;
 use lib 'lib';
 use BEX;
 
+my $given_job;
 my $queue_name;
+my $status_fifo;
 
 GetOptions(
+       "j|job=s" => \$given_job,
        "q|queue=s" => \$queue_name,
+       "s|status-fifo=s" => \$status_fifo,
 ) or die <<AMEN ;
 Usage: brun [<options>] [[!]<machine-or-class> ...]
 
 Options:
--q, --queue=<name>     Run jobs in the given queue
+-j, --job=<id>         Run only the specified job
+-q, --queue=<name>     Select job queue
+    --status-fifo=<f>  Send status updates to the given named pipe
 AMEN
 
+my $status_fd;
+if (defined $status_fifo) {
+       open $status_fd, '>>', $status_fifo or die "Cannot open status FIFO: $!";
+       autoflush $status_fd, 1;
+}
+
+sub update_status($$$$;$) {
+       my ($mach, $job, $status, $log_on_queue, $msg) = @_;
+       if ($status_fd) {
+               print $status_fd "! $mach $job $status\n";
+       }
+       if ($log_on_queue) {
+               $log_on_queue->log($mach, $job, $status, $msg);
+       }
+}
+
 sub ping_machine($) {
        my ($mach) = @_;
+       update_status($mach, '-', 'PING', undef);
        `ping -c1 -n $mach >/dev/null 2>/dev/null`;
        return !$?;
 }
 
 sub run_job($$$) {
        my ($job, $queue, $mach) = @_;
+       my $jid = $job->{'ID'};
+
        # FIXME: rsyncing, rsync-only jobs
        # FIXME: Locking
 
-       my $tmp = $queue->temp_file($mach, $job->{'ID'});
+       my $tmp = $queue->temp_file($mach, $jid);
        open T, '>', $tmp or die;
        if (defined $BEX::Config::job_prolog) {
-               open P, $BEX::Config::job_prolog or return "Cannot open prolog: $!";
+               open P, $BEX::Config::job_prolog or return ('INTERR', "Cannot open prolog: $!");
                while (<P>) { print T; }
                close P;
        } else {
                print T "#!/bin/sh\n";
        }
-       print T "# BEX job ", $job->{'ID'}, "\n";
+       print T "# BEX job ", $jid, "\n";
        print T $job->{'body'};
        if (defined $BEX::Config::job_epilog) {
-               open E, $BEX::Config::job_epilog or return "Cannot open epilog: $!";
+               open E, $BEX::Config::job_epilog or return ('INTERR', "Cannot open epilog: $!");
                while (<E>) { print T; }
                close E;
        }
        close T;
 
+       update_status($mach, $jid, 'SEND', undef);
        my $cmd = 't=$(mktemp -t bex-XXXXXXXX) && cat >$t && chmod u+x $t && echo $t';
        my $rtmp = `ssh <$tmp $mach '$cmd'`;
-       !$? && defined($rtmp) && $rtmp ne '' or return "Transfer failed";
+       !$? && defined($rtmp) && $rtmp ne '' or return ('NOXFER', 'Transfer failed');
        chomp $rtmp;
 
+       update_status($mach, $jid, 'RUN', $queue);
        system 'ssh', '-t', $mach, "$rtmp ; e=\$? ; rm -f $rtmp ; exit \$e";
        if ($?) {
-               return 'Failed';
+               return ('FAILED', 'Job failed');
        } else {
-               return 'OK';
+               return ('OK', undef);
        }
 }
 
 my @machines = BEX::Config::parse_machine_list(@ARGV ? @ARGV : '*');
 my $queue = BEX::Queue->new($queue_name);
+
 for my $mach (@machines) {
        my @q = $queue->scan($mach) or next;
+       update_status($mach, '-', 'INIT', undef);
        my $ping;
        for my $jid (@q) {
+               if (defined $given_job) {
+                       $jid eq $given_job or next;
+               }
                my $job = BEX::Job->new_from_file($queue->job_file($jid));
                my $stat = {
                        'Time' => time,
                };
                print "### Running $jid (", $job->attr('Subject'), ") on $mach ###\n";
                $ping //= ping_machine($mach);
-               my $s;
+               my $s, $msg;
                if (!$ping) {
-                       $s = 'No ping';
+                       ($s, $msg) = ('NOPING', 'Does not ping');
                } else {
-                       $s = run_job($job, $queue, $mach);
+                       ($s, $msg) = run_job($job, $queue, $mach);
                }
+               update_status($mach, $jid, $s, $queue, $msg);
 
-               BEX::log("$mach $jid $s");
                if ($s eq 'OK') {
                        print "+++ OK\n";
                        $queue->remove($mach, $jid);
                } else {
-                       print "--- $s\n";
+                       print "--- $s: $msg\n";
                        $stat->{'Status'} = $s;
+                       $stat->{'Message'} = $msg;
                        $queue->write_job_status($mach, $jid, $stat);
                }
        }
+       update_status($mach, '-', 'DONE', undef);
 }