3 # jobd - perform Girocco maintenance jobs
5 # Run with --help for details
12 use POSIX
":sys_wait_h";
23 my $max_par_intensive = 1;
24 my $load_triggers = '10,2';
25 my $lockfile = "/tmp/jobd.lock";
26 my $restart_delay = 60;
30 my ($load_trig, $load_untrig);
36 my $p = $job->{'project'};
37 check_project_exists
($job) || return;
38 if (-e get_project_path
($p).".nofetch") {
40 return setup_gc
($job);
42 if (-e get_project_path
($p).".clone_in_progress") {
43 job_skip
($job, "initial mirroring not complete yet");
46 if (my $ts = is_operation_uptodate
($p, 'lastrefresh', $Girocco::Config
::min_mirror_interval
)) {
47 job_skip
($job, "not needed right now, last run at $ts");
51 if (is_svn_clone
($p)) {
52 # git svn can be very, very slow at times
53 $job->{'timeout_factor'} = 3;
55 exec_job_command
($job, ["$Girocco::Config::basedir/jobd/update.sh", $p], $quiet);
60 my $p = $job->{'project'};
61 check_project_exists
($job) || return;
62 if (my $ts = is_operation_uptodate
($p, 'lastgc', $Girocco::Config
::min_gc_interval
)) {
63 job_skip
($job, "not needed right now, last run at $ts");
66 exec_job_command
($job, ["$Girocco::Config::basedir/jobd/gc.sh", $p], $quiet);
72 project
=> $job->{'project'},
74 command
=> \
&gc_project
,
79 sub check_project_exists
{
81 my $p = $job->{'project'};
82 if (!-d get_project_path
($p)) {
83 job_skip
($job, "non-existent project");
89 sub get_project_path
{
90 "$Girocco::Config::reporoot/".shift().".git/";
93 sub is_operation_uptodate
{
94 my ($project, $which, $threshold) = @_;
95 my $path = get_project_path
($project);
96 my $timestamp = `GIT_DIR="$path" $Girocco::Config::git_bin config "gitweb.$which"`;
97 my $unix_ts = `date +%s -d "$timestamp"`;
98 (time - $unix_ts) <= $threshold ?
$timestamp : undef;
103 my $path = get_project_path
($project);
104 my $baseurl = `GIT_DIR="$path" $Girocco::Config::git_bin config "gitweb.baseurl"`;
105 my $svnurl = `GIT_DIR="$path" $Girocco::Config::git_bin config "svn-remote.svn.url"`;
106 return $baseurl =~ /^svn[:+]/i && $svnurl;
114 command
=> \
&update_project
,
115 on_success
=> \
&setup_gc
,
116 on_error
=> \
&setup_gc
,
121 queue_one
($_) for (Girocco
::Project
->get_full_list());
124 ######### Daemon operation {{{1
134 sub handle_softexit
{
135 error
("Waiting for outstanding jobs to finish... ".
136 "^C again to exit immediately");
139 $SIG{'INT'} = \
&handle_exit
;
143 error
("Killing outstanding jobs...");
144 $SIG{'TERM'} = 'IGNORE';
146 kill 'KILL', -($_->{'pid'});
148 unlink $lockfile if ($locked);
154 $opts{'queued_at'} = time;
155 $opts{'dont_run'} = 0;
156 $opts{'intensive'} = 0 unless exists $opts{'intensive'};
164 $job->{'command'}->($job);
165 if ($job->{'dont_run'}) {
174 "[".$job->{'type'}."::".$job->{'project'}."]";
177 # Only one of those per job!
178 sub exec_job_command
{
179 my ($job, $command, $err_only) = @_;
182 if (!defined($pid = fork)) {
183 error
(_job_name
($job) ." Can't fork job: $!");
184 $job->{'finished'} = 1;
188 open STDIN
, '/dev/null' || do {
189 error
(_job_name
($job) ."Can't read from /dev/null: $!");
190 $job->{'finished'} = 1;
194 open STDOUT
, '>/dev/null' || do {
195 error
(_job_name
($job) ." Can't write to /dev/null: $!");
196 $job->{'finished'} = 1;
200 # New process group so we can keep track of all of its children
201 if (!defined(POSIX
::setpgid
(0, 0))) {
202 error
(_job_name
($job) ." Can't create process group: $!");
203 $job->{'finished'} = 1;
207 select(undef, undef, undef, 0.1);
209 # Stop perl from complaining
212 $job->{'pid'} = $pid;
213 $job->{'finished'} = 0;
214 $job->{'started_at'} = time;
218 my ($job, $msg) = @_;
219 $job->{'dont_run'} = 1;
220 error
(_job_name
($job) ." Skipping job: $msg") unless $quiet || !$msg;
223 sub reap_hanging_jobs
{
225 my $factor = $_->{'timeout_factor'} || 1;
226 if (defined($_->{'started_at'}) && (time - $_->{'started_at'}) > ($kill_after * $factor)) {
227 $_->{'finished'} = 1;
228 kill 'KILL', -($_->{'pid'});
229 error
(_job_name
($_) ." KILLED due to timeout");
230 push @jobs_killed, _job_name
($_);
235 sub reap_finished_jobs
{
237 my $finished_any = 0;
239 $pid = waitpid(-1, WNOHANG
);
243 my @child = grep { $_->{'pid'} && $_->{'pid'} == $pid } @running;
245 # XXX- we currently don't care
247 if (@child && !$child[0]->{'finished'}) {
248 $child[0]->{'on_success'}->($child[0]) if defined($child[0]->{'on_success'});
249 $child[0]->{'finished'} = 1;
252 $child[0]->{'on_error'}->($child[0]) if defined($child[0]->{'on_error'});
255 @running = grep { $_->{'finished'} == 0 } @running;
259 sub have_intensive_jobs
{
260 grep { $_->{'intensive'} == 1 } @running;
264 "[". scalar(localtime) ."] ";
268 if ($^O
eq "linux") {
269 # Read /proc/loadavg on Linux
270 open(LOADAV
, '<', '/proc/loadavg') or return undef;
271 my $loadinfo = <LOADAV
>;
273 return (split(/\s/, $loadinfo, 4))[0..2];
275 # Read the output of uptime everywhere else (works on Linux too)
276 open(LOADAV
, '-|', 'uptime') or return undef;
277 my $loadinfo = <LOADAV
>;
279 $loadinfo =~ /load average[^0-9.]*([0-9.]+)[^0-9.]+([0-9.]+)[^0-9.]+([0-9.]+)/iso or return undef;
285 my $last_progress = time;
286 my $last_checkload = time - 5;
287 my $current_load = $load_trig;
294 ferror
("--- Processing %d queued jobs", scalar(@queue));
296 $SIG{'INT'} = \
&handle_softexit
;
297 $SIG{'TERM'} = \
&handle_exit
;
298 while (@queue || @running) {
300 my $proceed_immediately = reap_finished_jobs
();
301 # Check current system load
302 if ($load_trig && (time - $last_checkload) >= 5 && defined((my @loadinfo = get_load_info
())[0])) {
303 my $current_load = $loadinfo[0];
304 if ($current_load > $load_trig && !$overloaded) {
306 error
("PAUSE: system load is at $current_load > $load_trig") if $progress;
307 } elsif ($current_load < $load_untrig && $overloaded) {
309 error
("RESUME: system load is at $current_load < $load_untrig") if $progress;
312 $load_info = ', paused (load '. $current_load .')';
314 $load_info = ', load '. $current_load;
316 $last_checkload = time;
319 if ($progress && (time - $last_progress) >= 60) {
320 ferror
("STATUS: %d queued, %d running, %d finished, %d skipped, %d killed$load_info", scalar(@queue), scalar(@running), $jobs_executed, $jobs_skipped, scalar(@jobs_killed));
324 push @run_status, _job_name
($_)." ". (time - $_->{'started_at'}) ."s";
326 error
("STATUS: currently running: ". join(', ', @run_status));
328 $last_progress = time;
330 # Back off if we're too busy
331 if (@running >= $max_par || have_intensive_jobs
() >= $max_par_intensive || !@queue || $overloaded) {
332 sleep 1 unless $proceed_immediately;
336 run_job
(shift(@queue)) if @queue;
339 ferror
("--- Queue processed. %d jobs executed, %d skipped, %d killed.", $jobs_executed, $jobs_skipped, scalar(@jobs_killed));
343 sub run_perpetually
{
345 die "Lockfile exists. Please make sure no other instance of jobd is running.";
347 open LOCK
, '>', $lockfile || die "Cannot create lockfile $lockfile: $!";
355 sleep($restart_delay) if $perpetual; # Let the system breathe for a moment
360 ######### Helpers {{{1
363 print STDERR ts
().shift()."\n";
366 error
(sprintf($_[0], @_[1..$#_]));
376 Getopt
::Long
::Configure
('bundling');
377 my $parse_res = GetOptions
(
378 'help|?' => sub { pod2usage
(-verbose
=> 1, -exitval
=> 0); },
379 'quiet|q' => \
$quiet,
380 'progress|P' => \
$progress,
381 'kill-after|k=i' => \
$kill_after,
382 'max-parallel|p=i' => \
$max_par,
383 'max-intensive-parallel|i=i' => \
$max_par_intensive,
384 'load-triggers=s' => \
$load_triggers,
385 'restart-delay|d=i' => \
$restart_delay,
386 'lockfile|l=s' => \
$lockfile,
387 'all-once|a' => \
$all_once,
390 fatal
("Error: can only use one out of --all-once and --one")
391 if ($all_once && $one);
394 $ENV{'show_progress'} = '1';
398 $load_triggers = '0,0' if (!-f
'/proc/loadavg');
399 ($load_trig, $load_untrig) = split(/,/, $load_triggers);
415 ########## Documentation {{{1
421 jobd - Perform Girocco maintenance jobs
428 -h | --help detailed instructions
429 -q | --quiet run quietly
430 -P | --progress show occasional status updates
431 -k SECONDS | --kill-after SECONDS how long to wait before killing jobs
432 -p NUM | --max-parallel NUM how many jobs to run at the same time
433 -i NUM | --max-intensive-parallel NUM how many resource-hungry jobs to run
435 --load-triggers TRIG,UNTRIG stop queueing jobs at load above
436 TRIG and resume at load below UNTRIG
437 -d NUM | --restart-delay SECONDS wait for this many seconds between
439 -l FILE | --lockfile FILE create a lockfile in the given
441 -a | --all-once process the list only once
442 -o PRJNAME | --one PRJNAME process only one project
450 Print the full description of jobd's options.
454 Suppress non-error messages, e.g. for use when running this task as a cronjob.
458 Show information about the current status of the job queue occasionally. This
459 is automatically enabled if --quiet is not given.
461 =item B<--kill-after SECONDS>
463 Kill supervised jobs after a certain time to avoid hanging the daemon.
465 =item B<--max-parallel NUM>
467 Run no more than that many jobs at the same time.
469 =item B<--max-intensive-parallel NUM>
471 Run no more than that many resource-hungry jobs at the same time. Right now,
472 this refers to repacking jobs.
474 =item B<--load-triggers TRIG,UNTRIG>
476 If the first system load average (1 minute average) exceeds TRIG, don't queue
477 any more jobs until it goes below UNTRIG. This is currently only supported on
478 Linux and any other platforms that provide an uptime command with load average
481 If both values are zero, load checks are disabled. Note that this is not the
484 =item B<--restart-delay NUM>
486 After processing the queue, wait this many seconds until the queue is
489 =item B<--lockfile FILE>
491 For perpetual operation, create a lockfile in that place and clean it up after
496 Instead of perpetuously processing all projects over and over again, process
497 them just once and then exit.
499 =item B<--one PRJNAME>
501 Process only the given project (given as just the project name without C<.git>
502 suffix) and then exit.
508 jobd is Girocco's repositories maintenance servant; it periodically checks all
509 the repositories and updates mirrored repositories and repacks push-mode
510 repositories when needed.