Queen.pm 35.6 KB
Newer Older
Jessica Severin's avatar
Jessica Severin committed
1 2 3 4 5 6
#
# You may distribute this module under the same terms as perl itself

=pod 

=head1 NAME
7

8
  Bio::EnsEMBL::Hive::Queen
Jessica Severin's avatar
Jessica Severin committed
9 10

=head1 SYNOPSIS
11

12
  The Queen of the Hive based job control system
Jessica Severin's avatar
Jessica Severin committed
13 14

=head1 DESCRIPTION
15

16 17 18 19 20 21 22 23 24 25 26 27
  The Queen of the Hive based job control system is responsible to 'birthing' the
  correct number of workers of the right type so that they can find jobs to do.
  It will also free up jobs of Workers that died unexpectantly so that other workers
  can claim them to do.

  Hive based processing is a concept based on a more controlled version
  of an autonomous agent type system.  Each worker is not told what to do
  (like a centralized control system - like the current pipeline system)
  but rather queries a central database for jobs (give me jobs).

  Each worker is linked to an analysis_id, registers its self on creation
  into the Hive, creates a RunnableDB instance of the Analysis->module,
28 29
  gets $analysis->stats->batch_size jobs from the job table, does its work,
  creates the next layer of job entries by interfacing to
30 31 32
  the DataflowRuleAdaptor to determine the analyses it needs to pass its
  output data to and creates jobs on the next analysis database.
  It repeats this cycle until it has lived its lifetime or until there are no
33 34 35 36 37 38 39 40 41 42
  more jobs left.
  The lifetime limit is just a safety limit to prevent these from 'infecting'
  a system.

  The Queens job is to simply birth Workers of the correct analysis_id to get the
  work down.  The only other thing the Queen does is free up jobs that were
  claimed by Workers that died unexpectantly so that other workers can take
  over the work.

  The Beekeeper is in charge of interfacing between the Queen and a compute resource
43
  or 'compute farm'.  Its job is to query Queens if they need any workers and to
44 45 46
  send the requested number of workers to open machines via the runWorker.pl script.
  It is also responsible for interfacing with the Queen to identify worker which died
  unexpectantly.
Jessica Severin's avatar
Jessica Severin committed
47 48

=head1 CONTACT
49

50
  Please contact ehive-users@ebi.ac.uk mailing list with questions/suggestions.
Jessica Severin's avatar
Jessica Severin committed
51 52

=head1 APPENDIX
53

54 55
  The rest of the documentation details each of the object methods. 
  Internal methods are usually preceded with a _
Jessica Severin's avatar
Jessica Severin committed
56 57 58

=cut

59

Jessica Severin's avatar
Jessica Severin committed
60 61 62
package Bio::EnsEMBL::Hive::Queen;

use strict;
63
use POSIX;
64 65 66
use Bio::EnsEMBL::Utils::Argument;
use Bio::EnsEMBL::Utils::Exception;

67
use Bio::EnsEMBL::Hive::Utils ('destringify', 'dir_revhash');  # import 'destringify()' and 'dir_revhash()'
68
use Bio::EnsEMBL::Hive::AnalysisJob;
Jessica Severin's avatar
Jessica Severin committed
69 70
use Bio::EnsEMBL::Hive::Worker;

71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
use base ('Bio::EnsEMBL::Hive::DBSQL::ObjectAdaptor');


sub default_table_name {
    return 'worker';
}


sub default_insertion_method {
    return 'INSERT';
}


sub object_class {
    return 'Bio::EnsEMBL::Hive::Worker';
}
87

Jessica Severin's avatar
Jessica Severin committed
88

89
############################
Jessica Severin's avatar
Jessica Severin committed
90
#
91
# PUBLIC API
Jessica Severin's avatar
Jessica Severin committed
92
#
93
############################
Jessica Severin's avatar
Jessica Severin committed
94

95

Jessica Severin's avatar
Jessica Severin committed
96 97
=head2 create_new_worker

98 99 100 101
  Description: Creates an entry in the worker table,
               populates some non-storable attributes
               and returns a Worker object based on that insert.
               This guarantees that each worker registered in this Queen's hive is properly registered.
Jessica Severin's avatar
Jessica Severin committed
102
  Returntype : Bio::EnsEMBL::Hive::Worker
103
  Caller     : runWorker.pl
Jessica Severin's avatar
Jessica Severin committed
104 105 106 107

=cut

sub create_new_worker {
108
    my ($self, @args) = @_;
109

110 111 112 113 114 115
    my ($meadow_type, $meadow_name, $process_id, $exec_host, $resource_class_id, $resource_class_name,
        $no_write, $debug, $worker_log_dir, $hive_log_dir, $job_limit, $life_span, $no_cleanup, $retry_throwing_jobs, $compile_module_once) =

    rearrange([qw(meadow_type meadow_name process_id exec_host resource_class_id resource_class_name
                no_write debug worker_log_dir hive_log_dir job_limit life_span no_cleanup retry_throwing_jobs compile_module_once) ], @args);

116
    if( defined($resource_class_name) ) {
117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170
        my $rc = $self->db->get_ResourceClassAdaptor->fetch_by_name($resource_class_name)
            or die "resource_class with name='$resource_class_name' could not be fetched from the database";

        $resource_class_id = $rc->dbID;
    }

    my $sql = q{INSERT INTO worker (born, last_check_in, meadow_type, meadow_name, host, process_id, resource_class_id)
              VALUES (CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, ?, ?, ?, ?, ?)};

    my $sth = $self->prepare($sql);
    $sth->execute($meadow_type, $meadow_name, $exec_host, $process_id, $resource_class_id);
    my $worker_id = $self->dbc->db_handle->last_insert_id(undef, undef, 'worker', 'worker_id')
        or die "Could not create a new worker";
    $sth->finish;

    if($hive_log_dir or $worker_log_dir) {
        my $dir_revhash = dir_revhash($worker_id);
        $worker_log_dir ||= $hive_log_dir .'/'. ($dir_revhash ? "$dir_revhash/" : '') .'worker_id_'.$worker_id;

            # Note: the following die-message will not reach the log files for circular reason!
        system("mkdir -p $worker_log_dir") && die "Could not create '$worker_log_dir' because: $!";

        my $sth_add_log = $self->prepare( "UPDATE worker SET log_dir=? WHERE worker_id=?" );
        $sth_add_log->execute($worker_log_dir, $worker_id);
        $sth_add_log->finish;
    }

    my $worker = $self->fetch_by_dbID($worker_id)
        or die "Could not fetch worker with dbID=$worker_id";

    $worker->init;

    if($job_limit) {
      $worker->job_limit($job_limit);
      $worker->life_span(0);
    }

    $worker->life_span($life_span * 60)                 if($life_span);

    $worker->execute_writes(0)                          if($no_write);

    $worker->perform_cleanup(0)                         if($no_cleanup);

    $worker->debug($debug)                              if($debug);

    $worker->retry_throwing_jobs($retry_throwing_jobs)  if(defined $retry_throwing_jobs);

    $worker->compile_module_once($compile_module_once)  if(defined $compile_module_once);

    return $worker;
}


=head2 specialize_new_worker
171

172 173
  Description: If analysis_id or logic_name is specified it will try to specialize the Worker into this analysis.
               If not specified the Queen will analyze the hive and pick the most suitable analysis.
174
  Caller     : Bio::EnsEMBL::Hive::Worker
175 176 177 178 179 180 181 182

=cut

sub specialize_new_worker {
    my ($self, $worker, @args) = @_;

    my ($analysis_id, $logic_name, $job_id, $force) =
        rearrange([qw(analysis_id logic_name job_id force) ], @args);
183

184 185
    if( scalar( grep {defined($_)} ($analysis_id, $logic_name, $job_id) ) > 1) {
        die "At most one of the options {-analysis_id, -logic_name, -job_id} can be set to pre-specialize a Worker";
Leo Gordon's avatar
Leo Gordon committed
186 187
    }

188
    my ($analysis, $stats, $special_batch);
189
    my $analysis_stats_adaptor = $self->db->get_AnalysisStatsAdaptor;
Jessica Severin's avatar
Jessica Severin committed
190

191
    if($job_id or $analysis_id or $logic_name) {    # probably pre-specialized from command-line
192

193
        if($job_id) {
194 195 196
            print "resetting and fetching job for job_id '$job_id'\n";

            my $job_adaptor = $self->db->get_AnalysisJobAdaptor;
197

198 199 200 201 202 203 204 205 206
            my $job = $job_adaptor->fetch_by_dbID( $job_id )
                or die "Could not fetch job with dbID='$job_id'";
            my $job_status = $job->status();

            if($job_status =~/(CLAIMED|PRE_CLEANUP|FETCH_INPUT|RUN|WRITE_OUTPUT|POST_CLEANUP)/ ) {
                die "Job with dbID='$job_id' is already in progress, cannot run";   # FIXME: try GC first, then complain
            } elsif($job_status =~/(DONE|SEMAPHORED)/ and !$force) {
                die "Job with dbID='$job_id' is $job_status, please use -force 1 to override";
            }
207

208 209 210 211 212 213 214 215
            if(($job_status eq 'DONE') and $job->semaphored_job_id) {
                warn "Increasing the semaphore count of the dependent job";
                $job_adaptor->increase_semaphore_count_for_jobid( $job->semaphored_job_id );
            }

            my $worker_id = $worker->dbID;
            if($job = $job_adaptor->reset_or_grab_job_by_dbID($job_id, $worker_id)) {
                $special_batch = [ $job ];
216 217
                $analysis_id = $job->analysis_id;
            } else {
218
                die "Could not claim job with dbID='$job_id' for worker with dbID='$worker_id'";
219
            }
220 221 222 223 224 225 226 227 228 229 230 231 232
        }

        if($logic_name) {
            $analysis = $self->db->get_AnalysisAdaptor->fetch_by_logic_name($logic_name)
                or die "analysis with name='$logic_name' could not be fetched from the database";

            $analysis_id = $analysis->dbID;

        } elsif($analysis_id) {
            $analysis = $self->db->get_AnalysisAdaptor->fetch_by_dbID($analysis_id)
                or die "analysis with dbID='$analysis_id' could not be fetched from the database";
        }

233 234 235
        if( $worker->resource_class_id
        and $worker->resource_class_id != $analysis->resource_class_id) {
                die "resource_class of analysis ".$analysis->logic_name." is incompatible with this Worker's resource_class";
236 237
        }

238 239 240
        $stats = $analysis_stats_adaptor->fetch_by_analysis_id($analysis_id);
        $self->safe_synchronize_AnalysisStats($stats);

241
        unless($special_batch or $force) {    # do we really need to run this analysis?
242
            if($self->get_hive_current_load() >= 1.1) {
243
                die "Hive is overloaded, can't specialize a worker";
244 245
            }
            if($stats->status eq 'BLOCKED') {
246
                die "Analysis is BLOCKED, can't specialize a worker";
247 248
            }
            if($stats->num_required_workers <= 0) {
249
                die "Analysis requires 0 workers at the moment";
250 251
            }
            if($stats->status eq 'DONE') {
252
                die "Analysis is DONE, and doesn't require workers";
253 254
            }
        }
255

256
    } else {    # probably scheduled by beekeeper.pl
257

258
        $stats = $self->suggest_analysis_to_specialize_by_rc_id($worker->resource_class_id)
259
            or die "Scheduler failed to pick an analysis for the worker";
260

261
        print "Scheduler picked analysis with dbID=".$stats->analysis_id." for the worker\n";
262 263

        $analysis_id = $stats->analysis_id;
264
    }
265

266
        # now set it in the $worker:
Jessica Severin's avatar
Jessica Severin committed
267

268
    $worker->analysis_id( $analysis_id );
Jessica Severin's avatar
Jessica Severin committed
269

270 271 272 273
    my $sth_update_analysis_id = $self->prepare( "UPDATE worker SET analysis_id=? WHERE worker_id=?" );
    $sth_update_analysis_id->execute($worker->analysis_id, $worker->dbID);
    $sth_update_analysis_id->finish;

274 275
    if($special_batch) {
        $worker->special_batch( $special_batch );
276
    } else {    # count it as autonomous worker sharing the load of that analysis:
277 278 279

        $stats->update_status('WORKING');

280
        $analysis_stats_adaptor->decrease_required_workers($worker->analysis_id);
281 282 283 284 285 286 287 288 289
    }

        # The following increment used to be done only when no specific task was given to the worker,
        # thereby excluding such "special task" workers from being counted in num_running_workers.
        #
        # However this may be tricky to emulate by triggers that know nothing about "special tasks",
        # so I am (temporarily?) simplifying the accounting algorithm.
        #
    unless( $self->db->hive_use_triggers() ) {
290
        $analysis_stats_adaptor->increase_running_workers($worker->analysis_id);
291
    }
Jessica Severin's avatar
Jessica Severin committed
292 293
}

294

Jessica Severin's avatar
Jessica Severin committed
295
sub register_worker_death {
296
    my ($self, $worker) = @_;
297

298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321
    return unless($worker);

    my $cod = $worker->cause_of_death() || 'UNKNOWN';    # make sure we do not attempt to insert a void

    my $sql = qq{UPDATE worker SET died=CURRENT_TIMESTAMP
                    ,last_check_in=CURRENT_TIMESTAMP
                    ,status='DEAD'
                    ,work_done='}. $worker->work_done . qq{'
                    ,cause_of_death='$cod'
                WHERE worker_id='}. $worker->dbID . qq{'};
    $self->dbc->do( $sql );

    if(my $analysis_id = $worker->analysis_id) {
        my $analysis_stats_adaptor = $self->db->get_AnalysisStatsAdaptor;

        unless( $self->db->hive_use_triggers() ) {
            $analysis_stats_adaptor->decrease_running_workers($worker->analysis_id);
        }

        if($cod eq 'NO_WORK') {
            $analysis_stats_adaptor->update_status($worker->analysis_id, 'ALL_CLAIMED');
        } elsif($cod eq 'UNKNOWN'
            or $cod eq 'MEMLIMIT'
            or $cod eq 'RUNLIMIT'
322 323
            or $cod eq 'KILLED_BY_USER'
            or $cod eq 'CONTAMINATED') {
324 325 326 327 328 329 330 331 332 333
                $self->db->get_AnalysisJobAdaptor->release_undone_jobs_from_worker($worker);
        }

            # re-sync the analysis_stats when a worker dies as part of dynamic sync system
        if($self->safe_synchronize_AnalysisStats($worker->analysis->stats)->status ne 'DONE') {
            # since I'm dying I should make sure there is someone to take my place after I'm gone ...
            # above synch still sees me as a 'living worker' so I need to compensate for that
            $analysis_stats_adaptor->increase_required_workers($worker->analysis_id);
        }
    }
Jessica Severin's avatar
Jessica Severin committed
334 335
}

336

337
sub check_for_dead_workers {    # scans the whole Valley for lost Workers (but ignores unreachagle ones)
338
    my ($self, $valley, $check_buried_in_haste) = @_;
Leo Gordon's avatar
Leo Gordon committed
339

340
    warn "GarbageCollector:\tChecking for lost Workers...\n";
341

342 343 344 345
    my $queen_worker_list           = $self->fetch_overdue_workers(0);
    my %mt_and_pid_to_worker_status = ();
    my %worker_status_counts        = ();
    my %mt_and_pid_to_lost_worker   = ();
Leo Gordon's avatar
Leo Gordon committed
346

347
    warn "GarbageCollector:\t[Queen:] we have ".scalar(@$queen_worker_list)." Workers alive.\n";
348

Leo Gordon's avatar
Leo Gordon committed
349 350
    foreach my $worker (@$queen_worker_list) {

351 352 353
        my $meadow_type = $worker->meadow_type;
        if(my $meadow = $valley->find_available_meadow_responsible_for_worker($worker)) {
            $mt_and_pid_to_worker_status{$meadow_type} ||= $meadow->status_of_all_our_workers;
Leo Gordon's avatar
Leo Gordon committed
354
        } else {
355
            $worker_status_counts{$meadow_type}{'UNREACHABLE'}++;
356

357
            next;   # Worker is unreachable from this Valley
Leo Gordon's avatar
Leo Gordon committed
358 359
        }

360 361 362 363 364
        my $process_id = $worker->process_id;
        if(my $status = $mt_and_pid_to_worker_status{$meadow_type}{$process_id}) { # can be RUN|PEND|xSUSP
            $worker_status_counts{$meadow_type}{$status}++;
        } else {
            $worker_status_counts{$meadow_type}{'LOST'}++;
365

366
            $mt_and_pid_to_lost_worker{$meadow_type}{$process_id} = $worker;
367
        }
368 369 370 371 372 373
    }

        # just a quick summary report:
    foreach my $meadow_type (keys %worker_status_counts) {
        warn "GarbageCollector:\t[$meadow_type Meadow:]\t".join(', ', map { "$_:$worker_status_counts{$meadow_type}{$_}" } keys %{$worker_status_counts{$meadow_type}})."\n\n";
    }
374

375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391
    while(my ($meadow_type, $pid_to_lost_worker) = each %mt_and_pid_to_lost_worker) {
        my $this_meadow = $valley->available_meadow_hash->{$meadow_type};

        if(my $lost_this_meadow = scalar(keys %$pid_to_lost_worker) ) {
            warn "GarbageCollector:\tDiscovered $lost_this_meadow lost $meadow_type Workers\n";

            my $wpid_to_cod = {};
            if($this_meadow->can('find_out_causes')) {
                $wpid_to_cod = $this_meadow->find_out_causes( keys %$pid_to_lost_worker );
                my $lost_with_known_cod = scalar(keys %$wpid_to_cod);
                warn "GarbageCollector:\tFound why $lost_with_known_cod of $meadow_type Workers died\n";
            } else {
                warn "GarbageCollector:\t$meadow_type meadow does not support post-mortem examination\n";
            }

            warn "GarbageCollector:\tReleasing the jobs\n";
            while(my ($process_id, $worker) = each %$pid_to_lost_worker) {
392
                $worker->cause_of_death( $wpid_to_cod->{$process_id} || 'UNKNOWN');
393 394
                $self->register_worker_death($worker);
            }
395 396 397
        }
    }

398
        # the following bit is completely Meadow-agnostic and only restores database integrity:
Leo Gordon's avatar
Leo Gordon committed
399
    if($check_buried_in_haste) {
400
        warn "GarbageCollector:\tChecking for Workers buried in haste...\n";
401
        my $buried_in_haste_list = $self->fetch_all_dead_workers_with_jobs();
Leo Gordon's avatar
Leo Gordon committed
402
        if(my $bih_number = scalar(@$buried_in_haste_list)) {
403
            warn "GarbageCollector:\tfound $bih_number jobs, reclaiming.\n\n";
Leo Gordon's avatar
Leo Gordon committed
404 405 406
            if($bih_number) {
                my $job_adaptor = $self->db->get_AnalysisJobAdaptor();
                foreach my $worker (@$buried_in_haste_list) {
Leo Gordon's avatar
Leo Gordon committed
407
                    $job_adaptor->release_undone_jobs_from_worker($worker);
Leo Gordon's avatar
Leo Gordon committed
408 409 410
                }
            }
        } else {
411
            warn "GarbageCollector:\tfound none\n";
Leo Gordon's avatar
Leo Gordon committed
412 413 414
        }
    }
}
Jessica Severin's avatar
Jessica Severin committed
415 416


417 418 419
    # a new version that both checks in and updates the status
sub check_in_worker {
    my ($self, $worker) = @_;
Jessica Severin's avatar
Jessica Severin committed
420

421
    $self->dbc->do("UPDATE worker SET last_check_in=CURRENT_TIMESTAMP, status='".$worker->status."', work_done='".$worker->work_done."' WHERE worker_id='".$worker->dbID."'");
Jessica Severin's avatar
Jessica Severin committed
422 423 424
}


425
=head2 reset_job_by_dbID_and_sync
426

427
  Arg [1]: int $job_id
428
  Example: 
429
    my $job = $queen->reset_job_by_dbID_and_sync($job_id);
430
  Description: 
431
    For the specified job_id it will fetch just that job, 
432 433 434
    reset it completely as if it has never run, and return it.  
    Specifying a specific job bypasses the safety checks, 
    thus multiple workers could be running the 
435
    same job simultaneously (use only for debugging).
436
  Returntype : none
437
  Exceptions :
438
  Caller     : beekeeper.pl
439 440 441

=cut

442 443
sub reset_job_by_dbID_and_sync {
    my ($self, $job_id) = @_;
444

445
    my $job_adaptor = $self->db->get_AnalysisJobAdaptor;
446
    my $job = $job_adaptor->reset_or_grab_job_by_dbID($job_id); 
447 448 449

    my $stats = $self->db->get_AnalysisStatsAdaptor->fetch_by_analysis_id($job->analysis_id);
    $self->synchronize_AnalysisStats($stats);
450 451 452
}


453 454 455 456 457 458
######################################
#
# Public API interface for beekeeper
#
######################################

459

460 461 462
    # Note: asking for Queen->fetch_overdue_workers(0) essentially means
    #       "fetch all workers known to the Queen not to be officially dead"
    #
463
sub fetch_overdue_workers {
464
    my ($self,$overdue_secs) = @_;
465

466
    $overdue_secs = 3600 unless(defined($overdue_secs));
467

468
    my $constraint = "status!='DEAD' AND ".
469
                    ( ($self->dbc->driver eq 'sqlite')
470 471 472
                        ? "(strftime('%s','now')-strftime('%s',last_check_in))>$overdue_secs"
                        : "(UNIX_TIMESTAMP()-UNIX_TIMESTAMP(last_check_in))>$overdue_secs");
    return $self->fetch_all( $constraint );
473 474
}

475 476 477

sub fetch_all_dead_workers_with_jobs {
    my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
478

479
    return $self->fetch_all( "JOIN job j USING(worker_id) WHERE worker.status='DEAD' AND j.status NOT IN ('DONE', 'READY', 'FAILED', 'PASSED_ON') GROUP BY worker_id" );
Leo Gordon's avatar
Leo Gordon committed
480
}
Jessica Severin's avatar
Jessica Severin committed
481

482

483 484
=head2 synchronize_hive

Leo Gordon's avatar
Leo Gordon committed
485
  Arg [1]    : $filter_analysis (optional)
486 487
  Example    : $queen->synchronize_hive();
  Description: Runs through all analyses in the system and synchronizes
488 489
              the analysis_stats summary with the states in the job 
              and worker tables.  Then follows by checking all the blocking rules
490 491 492 493 494 495 496
              and blocks/unblocks analyses as needed.
  Exceptions : none
  Caller     : general

=cut

sub synchronize_hive {
Leo Gordon's avatar
Leo Gordon committed
497
  my $self          = shift;
Leo Gordon's avatar
Leo Gordon committed
498
  my $filter_analysis = shift; # optional parameter
499

500
  my $start_time = time();
501

Leo Gordon's avatar
Leo Gordon committed
502
  my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
503

Leo Gordon's avatar
Leo Gordon committed
504
  print STDERR "\nSynchronizing the hive (".scalar(@$list_of_analyses)." analyses this time):\n";
Leo Gordon's avatar
Leo Gordon committed
505 506
  foreach my $analysis (@$list_of_analyses) {
    $self->synchronize_AnalysisStats($analysis->stats);
507
    print STDERR ( ($analysis->stats()->status eq 'BLOCKED') ? 'x' : 'o');
508
  }
Leo Gordon's avatar
Leo Gordon committed
509
  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
510

Leo Gordon's avatar
Leo Gordon committed
511
  print STDERR ''.((time() - $start_time))." seconds to synchronize_hive\n\n";
512
}
513

514

515 516 517
=head2 safe_synchronize_AnalysisStats

  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
518
  Example    : $self->safe_synchronize_AnalysisStats($stats);
519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535
  Description: Prewrapper around synchronize_AnalysisStats that does
               checks and grabs sync_lock before proceeding with sync.
               Used by distributed worker sync system to avoid contention.
  Exceptions : none
  Caller     : general

=cut

sub safe_synchronize_AnalysisStats {
  my $self = shift;
  my $stats = shift;

  return $stats unless($stats->analysis_id);
  return $stats if($stats->status eq 'SYNCHING');
  return $stats if($stats->status eq 'DONE');
  return $stats if($stats->sync_lock);
  return $stats if(($stats->status eq 'WORKING') and
536
                   ($stats->seconds_since_last_update < 3*60));
537 538 539 540

  # OK try to claim the sync_lock
  my $sql = "UPDATE analysis_stats SET status='SYNCHING', sync_lock=1 ".
            "WHERE sync_lock=0 and analysis_id=" . $stats->analysis_id;
541
  #print("$sql\n");
542
  my $row_count = $self->dbc->do($sql);  
543
  return $stats unless($row_count == 1);        # return the un-updated status if locked
544
  #printf("got sync_lock on analysis_stats(%d)\n", $stats->analysis_id);
545
  
546
      # since we managed to obtain the lock, let's go and perform the sync:
547
  $self->synchronize_AnalysisStats($stats);
548

549 550 551 552
  return $stats;
}


553
=head2 synchronize_AnalysisStats
554

555 556
  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
  Example    : $self->synchronize($analysisStats);
557
  Description: Queries the job and worker tables to get summary counts
558 559 560 561 562
               and rebuilds the AnalysisStats object.  Then updates the
               analysis_stats table with the new summary info
  Returntype : newly synced Bio::EnsEMBL::Hive::AnalysisStats object
  Exceptions : none
  Caller     : general
563

564
=cut
565

566
sub synchronize_AnalysisStats {
567
  my $self = shift;
568
  my $analysisStats = shift;
569

570
  return $analysisStats unless($analysisStats);
571
  return $analysisStats unless($analysisStats->analysis_id);
572

573
  $analysisStats->refresh(); ## Need to get the new hive_capacity for dynamic analyses
574 575
  my $hive_capacity = $analysisStats->hive_capacity;

576 577
  if($self->db->hive_use_triggers()) {

578 579
            my $job_count = $analysisStats->ready_job_count();
            my $required_workers = $hive_capacity && POSIX::ceil( $job_count / $analysisStats->get_or_estimate_batch_size() );
580

581 582
                # adjust_stats_for_living_workers:
            if($hive_capacity > 0) {
583
                my $unfulfilled_capacity = $hive_capacity - $analysisStats->num_running_workers();
584

585 586
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
587 588 589 590 591 592
                }
            }
            $analysisStats->num_required_workers( $required_workers );

  } else {
      $analysisStats->total_job_count(0);
593 594
      $analysisStats->semaphored_job_count(0);
      $analysisStats->ready_job_count(0);
595 596 597
      $analysisStats->done_job_count(0);
      $analysisStats->failed_job_count(0);
      $analysisStats->num_required_workers(0);
598

599
            # ask for analysis_id to force MySQL to use existing index on (analysis_id, status)
600
      my $sql = "SELECT analysis_id, status, count(*) FROM job WHERE analysis_id=? GROUP BY analysis_id, status";
601 602
      my $sth = $self->prepare($sql);
      $sth->execute($analysisStats->analysis_id);
603

604 605 606
      my $done_here       = 0;
      my $done_elsewhere  = 0;
      my $total_job_count = 0;
607
      while (my ($dummy_analysis_id, $status, $job_count)=$sth->fetchrow_array()) {
608
    # print STDERR "$status: $job_count\n";
609

610
        $total_job_count += $job_count;
611

612
        if($status eq 'READY') {
613
            $analysisStats->ready_job_count($job_count);
614

615
            my $required_workers = $hive_capacity && POSIX::ceil( $job_count / $analysisStats->get_or_estimate_batch_size() );
616 617 618

                # adjust_stats_for_living_workers:
            if($hive_capacity > 0) {
619
                my $unfulfilled_capacity = $hive_capacity - $self->count_running_workers( $analysisStats->analysis_id() );
620

621 622
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
623
                }
624
            }
625 626
            $analysisStats->num_required_workers( $required_workers );

627 628
        } elsif($status eq 'SEMAPHORED') {
            $analysisStats->semaphored_job_count($job_count);
629
        } elsif($status eq 'DONE') {
630
            $done_here = $job_count;
631
        } elsif($status eq 'PASSED_ON') {
632 633 634
            $done_elsewhere = $job_count;
        } elsif ($status eq 'FAILED') {
            $analysisStats->failed_job_count($job_count);
635
        }
636
      } # /while
637
      $sth->finish;
638

639 640
      $analysisStats->total_job_count( $total_job_count );
      $analysisStats->done_job_count( $done_here + $done_elsewhere );
641
  } # /unless $self->{'_hive_use_triggers'}
642

643
  $analysisStats->check_blocking_control_rules();
644

645 646 647
  if($analysisStats->status ne 'BLOCKED') {
    $analysisStats->determine_status();
  }
648

649
  # $analysisStats->sync_lock(0); ## do we perhaps need it here?
650
  $analysisStats->update;  #update and release sync_lock
651

652
  return $analysisStats;
653
}
654 655


656 657 658 659 660 661 662 663 664 665 666 667 668 669
=head2 get_num_failed_analyses

  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object (optional)
  Example    : if( $self->get_num_failed_analyses( $my_analysis )) { do_something; }
  Example    : my $num_failed_analyses = $self->get_num_failed_analyses();
  Description: Reports all failed analyses and returns
                either the number of total failed (if no $filter_analysis was provided)
                or 1/0, depending on whether $filter_analysis failed or not.
  Returntype : int
  Exceptions : none
  Caller     : general

=cut

Leo Gordon's avatar
Leo Gordon committed
670
sub get_num_failed_analyses {
671
    my ($self, $filter_analysis) = @_;
672

673 674 675 676 677 678 679 680 681 682 683 684
    my $failed_analyses = $self->db->get_AnalysisAdaptor->fetch_all_failed_analyses();

    my $filter_analysis_failed = 0;

    foreach my $failed_analysis (@$failed_analyses) {
        print "\t##########################################################\n";
        print "\t# Too many jobs in analysis '".$failed_analysis->logic_name."' FAILED #\n";
        print "\t##########################################################\n\n";
        if($filter_analysis and ($filter_analysis->dbID == $failed_analysis)) {
            $filter_analysis_failed = 1;
        }
    }
685

686
    return $filter_analysis ? $filter_analysis_failed : scalar(@$failed_analyses);
687 688 689
}


690
sub get_hive_current_load {
691 692 693
    my $self = shift;
    my $sql = qq{
        SELECT sum(1/hive_capacity)
694 695 696 697
        FROM worker w
        JOIN analysis_stats USING(analysis_id)
        WHERE w.status!='DEAD'
        AND hive_capacity>0
698 699 700 701 702 703
    };
    my $sth = $self->prepare($sql);
    $sth->execute();
    my ($load)=$sth->fetchrow_array();
    $sth->finish;
    return ($load || 0);
704 705 706
}


707 708 709
sub count_running_workers {
    my ($self, $analysis_id) = @_;

710 711 712 713 714
    my $sql = qq{
            SELECT count(*)
            FROM worker
            WHERE status!='DEAD'
        } . ($analysis_id ? " AND analysis_id='$analysis_id'" : '');
715 716 717 718 719 720 721

    my $sth = $self->prepare($sql);
    $sth->execute();
    (my $running_workers_count)=$sth->fetchrow_array();
    $sth->finish();

    return $running_workers_count || 0;
722 723
}

724

725
=head2 schedule_workers
726

727
  Arg[1]     : Bio::EnsEMBL::Hive::Analysis object (optional)
728
  Example    : $count = $queen->schedule_workers();
729 730 731
  Description: Runs through the analyses in the system which are waiting
               for workers to be created for them.  Calculates the maximum
               number of workers needed to fill the current needs of the system
732
               If Arg[1] is defined, does it only for the given analysis.
733
  Exceptions : none
734
  Caller     : beekeepers and other external processes
735 736 737

=cut

738
sub schedule_workers {
739
  my ($self, $filter_analysis, $orig_pending_by_rc_name, $available_submit_limit) = @_;
740

741 742 743
  my @suitable_analyses         = $filter_analysis
                                    ? ( $filter_analysis->stats )
                                    : @{ $self->db->get_AnalysisStatsAdaptor->fetch_all_by_suitability_rc_id() };
744

745 746 747 748
  my $analysis_id2rc_id         = $self->db->get_AnalysisAdaptor->fetch_HASHED_FROM_analysis_id_TO_resource_class_id();
  my $rc_id2name                = $self->db->get_ResourceClassAdaptor->fetch_HASHED_FROM_resource_class_id_TO_name();
        # combined mapping
  my %analysis_id2rc_name       = map { $_ => $rc_id2name->{ $analysis_id2rc_id->{ $_ }} } keys %$analysis_id2rc_id;
749

750
  return {} unless(@suitable_analyses);
751

752
  my %pending_by_rc_name        = %{ $orig_pending_by_rc_name || {} };  # NB: make our own copy to be able to modify it
753
  my $total_workers_to_run      = 0;
754
  my %workers_to_run_by_rc_name = ();
755
  my $available_load            = 1.0 - $self->get_hive_current_load();
756

757

758
  foreach my $analysis_stats (@suitable_analyses) {
759 760
    last if ($available_load <= 0.0);
    last if (defined($available_submit_limit) and !$available_submit_limit);
761

Leo Gordon's avatar
Leo Gordon committed
762
        #digging deeper under the surface so need to sync
763
    if(($analysis_stats->status eq 'LOADING') or ($analysis_stats->status eq 'BLOCKED') or ($analysis_stats->status eq 'ALL_CLAIMED')) {
764 765 766 767 768 769
      $self->synchronize_AnalysisStats($analysis_stats);
    }

    next if($analysis_stats->status eq 'BLOCKED');
    next if($analysis_stats->num_required_workers == 0);

770
        # FIXME: the following call *sometimes* returns a stale number greater than the number of workers actually needed for an analysis; -sync fixes it
Leo Gordon's avatar
Leo Gordon committed
771
    my $workers_this_analysis = $analysis_stats->num_required_workers;
772

773 774 775 776 777 778
    if(defined($available_submit_limit)) {                              # submit_limit total capping, if available
        if($workers_this_analysis > $available_submit_limit) {
            $workers_this_analysis = $available_submit_limit;
        }
        $available_submit_limit -= $workers_this_analysis;
    }
Leo Gordon's avatar
Leo Gordon committed
779

780 781 782 783 784
    if((my $hive_capacity = $analysis_stats->hive_capacity) > 0) {      # per-analysis hive_capacity capping, if available
        my $remaining_capacity_for_this_analysis = int($available_load * $hive_capacity);

        if($workers_this_analysis > $remaining_capacity_for_this_analysis) {
            $workers_this_analysis = $remaining_capacity_for_this_analysis;
Leo Gordon's avatar
Leo Gordon committed
785 786
        }

787
        $available_load -= 1.0*$workers_this_analysis/$hive_capacity;
788
    }
Leo Gordon's avatar
Leo Gordon committed
789

790
    my $curr_rc_name    = $analysis_id2rc_name{ $analysis_stats->analysis_id };
791

792 793 794 795 796 797
    if($pending_by_rc_name{ $curr_rc_name }) {                              # per-rc_name capping by pending processes, if available
        my $pending_this_analysis = ($pending_by_rc_name{ $curr_rc_name } < $workers_this_analysis) ? $pending_by_rc_name{ $curr_rc_name } : $workers_this_analysis;

        print "Scheduler detected $pending_this_analysis pending workers with resource_class_name=$curr_rc_name\n";
        $workers_this_analysis                  -= $pending_this_analysis;
        $pending_by_rc_name{ $curr_rc_name }    -= $pending_this_analysis;
798 799
    }

800 801
    next unless($workers_this_analysis);    # do not autovivify the hash by a zero

802
    $total_workers_to_run += $workers_this_analysis;
803
    $workers_to_run_by_rc_name{ $curr_rc_name } += $workers_this_analysis;
804
    $analysis_stats->print_stats();
805
    printf("Scheduler suggests adding %d more workers of resource_class_name=%s for analysis_id=%d [%1.3f hive_load remaining]\n", $workers_this_analysis, $curr_rc_name, $analysis_stats->analysis_id, $available_load);
806
  }
807

808
  printf("Scheduler suggests adding a total of %d workers [%1.5f hive_load remaining]\n", $total_workers_to_run, $available_load);
809
  return \%workers_to_run_by_rc_name;
Leo Gordon's avatar
Leo Gordon committed
810 811
}

812

813
sub schedule_workers_resync_if_necessary {
814 815 816
    my ($self, $valley, $analysis) = @_;

    my $meadow              = $valley->get_current_meadow();
Leo Gordon's avatar
Leo Gordon committed
817

818
    my $pending_by_rc_name  = ($meadow->can('count_pending_workers_by_rc_name') and $meadow->config_get('PendingAdjust')) ? $meadow->count_pending_workers_by_rc_name() : {};
819 820
    my $submit_limit        = $meadow->config_get('SubmitWorkersMax');
    my $meadow_limit        = ($meadow->can('count_running_workers') and defined($meadow->config_get('TotalRunningWorkersMax'))) ? $meadow->config_get('TotalRunningWorkersMax') - $meadow->count_running_workers : undef;
Leo Gordon's avatar
Leo Gordon committed
821

822 823 824 825
    my $available_submit_limit = ($submit_limit and $meadow_limit)
                                    ? (($submit_limit<$meadow_limit) ? $submit_limit : $meadow_limit)
                                    : (defined($submit_limit) ? $submit_limit : $meadow_limit);

826
    my $workers_to_run_by_rc_name = $self->schedule_workers($analysis, $pending_by_rc_name, $available_submit_limit);
827

828
    unless( keys %$workers_to_run_by_rc_name or $self->get_hive_current_load() or $self->count_running_workers() ) {
829
        print "\nScheduler: nothing is running and nothing to do (according to analysis_stats) => executing garbage collection and sync\n" ;