Queen.pm 34.1 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 for Workers
Jessica Severin's avatar
Jessica Severin committed
92
#
93
############################
Jessica Severin's avatar
Jessica Severin committed
94 95 96

=head2 create_new_worker

97
  Arg [1]    : $analysis_id (optional)
Jessica Severin's avatar
Jessica Severin committed
98
  Example    :
99 100 101
  Description: If analysis_id is specified it will try to create a worker based
               on that analysis.  If not specified the queen will analyze the hive
               and pick the analysis that has the most amount of work to be done.
102
               It creates an entry in the worker table, and returns a Worker object 
103 104
               based on that insert.  This guarantees that each worker registered
               in this queens hive is properly registered.
Jessica Severin's avatar
Jessica Severin committed
105 106 107 108 109 110 111
  Returntype : Bio::EnsEMBL::Hive::Worker
  Exceptions :
  Caller     :

=cut

sub create_new_worker {
112 113
  my ($self, @args) = @_;

114
  my (  $meadow_type, $meadow_name, $process_id, $exec_host,
Leo Gordon's avatar
Leo Gordon committed
115
        $rc_id, $rc_name, $analysis_id, $logic_name, $job_id, $input_id,
116
        $no_write, $debug, $worker_log_dir, $hive_log_dir, $job_limit, $life_span, $no_cleanup, $retry_throwing_jobs, $compile_module_once) =
117

118
 rearrange([qw(meadow_type meadow_name process_id exec_host
Leo Gordon's avatar
Leo Gordon committed
119
        rc_id rc_name analysis_id logic_name job_id input_id
120
        no_write debug worker_log_dir hive_log_dir job_limit life_span no_cleanup retry_throwing_jobs compile_module_once) ], @args);
121

Leo Gordon's avatar
Leo Gordon committed
122 123 124 125 126 127 128 129 130 131 132
    if($rc_name) {
        if($rc_id) {
            die "You should either define -rc_id or -rc_name, but not both\n";
        }
        if(my $rc = $self->db->get_ResourceClassAdaptor->fetch_by_name($rc_name)) {
            $rc_id = $rc->dbID;
        } else {
            die "rc_name '$rc_name' could not be fetched from the database\n";
        }
    }

133 134 135 136 137 138 139 140 141 142
    if($logic_name) {
        if($analysis_id) {
            die "You should either define -analysis_id or -logic_name, but not both\n";
        }
        if(my $analysis = $self->db->get_AnalysisAdaptor->fetch_by_logic_name($logic_name)) {
            $analysis_id = $analysis->dbID;
        } else {
            die "logic_name '$logic_name' could not be fetched from the database\n";
        }
    }
Jessica Severin's avatar
Jessica Severin committed
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
    my $job;

    if($input_id) {
        if($job_id) {
            die "You should either define -input_id or -job_id, but not both\n";

        } elsif($analysis_id) {
            $job = Bio::EnsEMBL::Hive::AnalysisJob->new(
                -INPUT_ID       => $input_id,
                -ANALYSIS_ID    => $analysis_id,
                -DBID           => -1,
            );
            print "creating a job outside the database\n";
            $job->print_job;
            $debug=1 unless(defined($debug));
        } else {
            die "For creating a job outside the database either -analysis_id or -logic_name must also be defined\n";
        }
    }

    if($job_id) {
        if($analysis_id) {
            die "When you specify -job_id, please omit both -logic_name and -analysis_id to avoid confusion\n";
        } else {
168 169 170 171 172
            print "resetting and fetching job for job_id '$job_id'\n";

            my $job_adaptor = $self->db->get_AnalysisJobAdaptor;
            $job_adaptor->reset_job_by_dbID($job_id); 
            if($job = $job_adaptor->fetch_by_dbID($job_id)) {
173 174 175 176 177 178
                $analysis_id = $job->analysis_id;
            } else {
                die "job_id '$job_id' could not be fetched from the database\n";
            }
        }
    }
179

180
  
181
  my $analysis_stats_adaptor = $self->db->get_AnalysisStatsAdaptor or return undef;
182 183
  my $analysisStats;
  if($analysis_id) {
184
    $analysisStats = $analysis_stats_adaptor->fetch_by_analysis_id($analysis_id);
185
    $self->safe_synchronize_AnalysisStats($analysisStats);
186
    #return undef unless(($analysisStats->status ne 'BLOCKED') and ($analysisStats->num_required_workers > 0));
187
  } else {
188
    if( $analysisStats = $self->specialize_new_worker($rc_id) ) {
189 190 191 192
        print "Scheduler picked analysis_id=".$analysisStats->analysis_id()." for the worker\n";
    } else {
        print "Scheduler failed to pick analysis_id for the worker\n";
    }
193
  }
194
  return undef unless($analysisStats);
195

196 197
  unless($job) {
    #go into autonomous mode
198 199 200 201 202

    if($self->get_hive_current_load() >= 1.1) {
        print "Hive is overloaded, can't create a worker\n";
        return;
    }
203
    if($analysisStats->status eq 'BLOCKED') {
204 205
      print "Analysis is BLOCKED, can't create workers\n";
      return;
206 207
    }
    if($analysisStats->status eq 'DONE') {
208 209
      print "Analysis is DONE, don't need to create workers\n";
      return;
210
    }
211

212 213
    $analysis_stats_adaptor->decrease_required_workers($analysisStats->analysis_id);

214
    $analysisStats->print_stats;
215
  }
216
  
217 218 219 220 221 222 223 224 225 226
    # 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() ) {
        $analysis_stats_adaptor->increase_running_workers($analysisStats->analysis_id);
  }

227
    my $sql = q{INSERT INTO worker (born, last_check_in, meadow_type, meadow_name, process_id, host, analysis_id)
228
              VALUES (CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, ?,?,?,?,?)};
Jessica Severin's avatar
Jessica Severin committed
229

230 231 232 233
    my $sth = $self->prepare($sql);
    $sth->execute($meadow_type, $meadow_name, $process_id, $exec_host, $analysisStats->analysis_id);
    my $worker_id = $self->dbc->db_handle->last_insert_id(undef, undef, 'worker', 'worker_id');
    $sth->finish;
Jessica Severin's avatar
Jessica Severin committed
234

235 236 237
    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;
Jessica Severin's avatar
Jessica Severin committed
238

239 240
            # 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: $!";
241

242 243 244 245 246 247 248 249 250 251 252 253
        my $sql_add_log = "UPDATE worker SET log_dir=? WHERE worker_id=?";
        my $sth_add_log = $self->prepare($sql_add_log);
        $sth_add_log->execute($worker_log_dir, $worker_id);
        $sth_add_log->finish;
    }

    my $worker = $self->fetch_by_dbID($worker_id);

    $worker->init;

    $worker->_specific_job($job) if($job);
    $worker->execute_writes(0) if($no_write);
254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270

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


    if($job_limit) {
      $worker->job_limit($job_limit);
      $worker->life_span(0);
    }
    if($life_span) {
      $worker->life_span($life_span * 60);
    }
    if($no_cleanup) { 
      $worker->perform_cleanup(0); 
    }
    if(defined $retry_throwing_jobs) {
        $worker->retry_throwing_jobs($retry_throwing_jobs);
    }
271 272 273
    if(defined $compile_module_once) {
        $worker->compile_module_once($compile_module_once);
    }
274

275 276
    $analysisStats->update_status('WORKING');

277
    return $worker;
Jessica Severin's avatar
Jessica Severin committed
278 279
}

280

Jessica Severin's avatar
Jessica Severin committed
281 282 283 284
sub register_worker_death {
  my ($self, $worker) = @_;

  return unless($worker);
Leo Gordon's avatar
Leo Gordon committed
285

Leo Gordon's avatar
bugfix  
Leo Gordon committed
286 287
  my $cod = $worker->cause_of_death();

288
  my $sql = "UPDATE worker SET died=CURRENT_TIMESTAMP, last_check_in=CURRENT_TIMESTAMP";
289
  $sql .= " ,status='DEAD'";
Jessica Severin's avatar
Jessica Severin committed
290
  $sql .= " ,work_done='" . $worker->work_done . "'";
Leo Gordon's avatar
bugfix  
Leo Gordon committed
291
  $sql .= " ,cause_of_death='$cod'";
292
  $sql .= " WHERE worker_id='" . $worker->dbID ."'";
Jessica Severin's avatar
Jessica Severin committed
293

Leo Gordon's avatar
bugfix  
Leo Gordon committed
294
  $self->dbc->do( $sql );
295

296 297 298 299
  unless( $self->db->hive_use_triggers() ) {
      $worker->analysis->stats->adaptor->decrease_running_workers($worker->analysis->stats->analysis_id);
  }

Leo Gordon's avatar
bugfix  
Leo Gordon committed
300
  if($cod eq 'NO_WORK') {
301
    $self->db->get_AnalysisStatsAdaptor->update_status($worker->analysis->dbID, 'ALL_CLAIMED');
302
  }
Leo Gordon's avatar
bugfix  
Leo Gordon committed
303 304 305 306
  if($cod eq 'FATALITY'
  or $cod eq 'MEMLIMIT'
  or $cod eq 'RUNLIMIT'
  or $cod eq 'KILLED_BY_USER') {
Leo Gordon's avatar
Leo Gordon committed
307
    $self->db->get_AnalysisJobAdaptor->release_undone_jobs_from_worker($worker);
308
  }
309
  
310
  # re-sync the analysis_stats when a worker dies as part of dynamic sync system
311
  if($self->safe_synchronize_AnalysisStats($worker->analysis->stats)->status ne 'DONE') {
312 313
    # 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
314
    $self->db->get_AnalysisStatsAdaptor->increase_required_workers($worker->analysis->dbID);
315 316
  }

Jessica Severin's avatar
Jessica Severin committed
317 318
}

319

320
sub check_for_dead_workers {    # scans the whole Valley for lost Workers (but ignores unreachagle ones)
321
    my ($self, $valley, $check_buried_in_haste) = @_;
Leo Gordon's avatar
Leo Gordon committed
322

323
    warn "GarbageCollector:\tChecking for lost Workers...\n";
324

325 326 327 328
    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
329

330
    warn "GarbageCollector:\t[Queen:] we have ".scalar(@$queen_worker_list)." Workers alive.\n";
331

Leo Gordon's avatar
Leo Gordon committed
332 333
    foreach my $worker (@$queen_worker_list) {

334 335 336
        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
337
        } else {
338
            $worker_status_counts{$meadow_type}{'UNREACHABLE'}++;
339

340
            next;   # Worker is unreachable from this Valley
Leo Gordon's avatar
Leo Gordon committed
341 342
        }

343 344 345 346 347
        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'}++;
348

349
            $mt_and_pid_to_lost_worker{$meadow_type}{$process_id} = $worker;
350
        }
351 352 353 354 355 356
    }

        # 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";
    }
357

358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377
    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) {
                $worker->cause_of_death( $wpid_to_cod->{$process_id} || 'FATALITY');
                $self->register_worker_death($worker);
            }
378 379 380
        }
    }

381
        # the following bit is completely Meadow-agnostic and only restores database integrity:
Leo Gordon's avatar
Leo Gordon committed
382
    if($check_buried_in_haste) {
383
        warn "GarbageCollector:\tChecking for Workers buried in haste...\n";
384
        my $buried_in_haste_list = $self->fetch_all_dead_workers_with_jobs();
Leo Gordon's avatar
Leo Gordon committed
385
        if(my $bih_number = scalar(@$buried_in_haste_list)) {
386
            warn "GarbageCollector:\tfound $bih_number jobs, reclaiming.\n\n";
Leo Gordon's avatar
Leo Gordon committed
387 388 389
            if($bih_number) {
                my $job_adaptor = $self->db->get_AnalysisJobAdaptor();
                foreach my $worker (@$buried_in_haste_list) {
Leo Gordon's avatar
Leo Gordon committed
390
                    $job_adaptor->release_undone_jobs_from_worker($worker);
Leo Gordon's avatar
Leo Gordon committed
391 392 393
                }
            }
        } else {
394
            warn "GarbageCollector:\tfound none\n";
Leo Gordon's avatar
Leo Gordon committed
395 396 397
        }
    }
}
Jessica Severin's avatar
Jessica Severin committed
398 399


400 401 402
    # 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
403

404
    $self->dbc->do("UPDATE worker SET last_check_in=CURRENT_TIMESTAMP, status='".$worker->status."', work_done='".$worker->work_done."' WHERE worker_id='".$worker->dbID."'");
Jessica Severin's avatar
Jessica Severin committed
405 406 407
}


408
=head2 reset_job_by_dbID_and_sync
409

410
  Arg [1]: int $job_id
411
  Example: 
412
    my $job = $queen->reset_job_by_dbID_and_sync($job_id);
413
  Description: 
414
    For the specified job_id it will fetch just that job, 
415 416 417
    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 
418
    same job simultaneously (use only for debugging).
419
  Returntype : none
420
  Exceptions :
421
  Caller     : beekeeper.pl
422 423 424

=cut

425 426
sub reset_job_by_dbID_and_sync {
    my ($self, $job_id) = @_;
427

428 429 430 431 432 433
    my $job_adaptor = $self->db->get_AnalysisJobAdaptor;
    $job_adaptor->reset_job_by_dbID($job_id); 

    my $job = $job_adaptor->fetch_by_dbID($job_id); 
    my $stats = $self->db->get_AnalysisStatsAdaptor->fetch_by_analysis_id($job->analysis_id);
    $self->synchronize_AnalysisStats($stats);
434 435 436
}


437 438 439 440 441 442
######################################
#
# Public API interface for beekeeper
#
######################################

443

444 445 446
    # Note: asking for Queen->fetch_overdue_workers(0) essentially means
    #       "fetch all workers known to the Queen not to be officially dead"
    #
447
sub fetch_overdue_workers {
448
    my ($self,$overdue_secs) = @_;
449

450
    $overdue_secs = 3600 unless(defined($overdue_secs));
451

452
    my $constraint = "cause_of_death='' AND ".
453
                    ( ($self->dbc->driver eq 'sqlite')
454 455 456
                        ? "(strftime('%s','now')-strftime('%s',last_check_in))>$overdue_secs"
                        : "(UNIX_TIMESTAMP()-UNIX_TIMESTAMP(last_check_in))>$overdue_secs");
    return $self->fetch_all( $constraint );
457 458
}

459

460
sub fetch_failed_workers {
461
    my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
462

463
    return $self->fetch_all( "cause_of_death='FATALITY'" );
464 465
}

Leo Gordon's avatar
Leo Gordon committed
466

467 468
sub fetch_all_dead_workers_with_jobs {
    my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
469

470
    return $self->fetch_all( "JOIN job j USING(worker_id) WHERE j.status NOT IN ('DONE', 'READY', 'FAILED', 'PASSED_ON') GROUP BY worker_id" );
Leo Gordon's avatar
Leo Gordon committed
471
}
Jessica Severin's avatar
Jessica Severin committed
472

473

474 475
=head2 synchronize_hive

Leo Gordon's avatar
Leo Gordon committed
476
  Arg [1]    : $filter_analysis (optional)
477 478
  Example    : $queen->synchronize_hive();
  Description: Runs through all analyses in the system and synchronizes
479 480
              the analysis_stats summary with the states in the job 
              and worker tables.  Then follows by checking all the blocking rules
481 482 483 484 485 486 487
              and blocks/unblocks analyses as needed.
  Exceptions : none
  Caller     : general

=cut

sub synchronize_hive {
Leo Gordon's avatar
Leo Gordon committed
488
  my $self          = shift;
Leo Gordon's avatar
Leo Gordon committed
489
  my $filter_analysis = shift; # optional parameter
490

491
  my $start_time = time();
492

Leo Gordon's avatar
Leo Gordon committed
493
  my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
494

Leo Gordon's avatar
Leo Gordon committed
495
  print STDERR "\nSynchronizing the hive (".scalar(@$list_of_analyses)." analyses this time):\n";
Leo Gordon's avatar
Leo Gordon committed
496 497
  foreach my $analysis (@$list_of_analyses) {
    $self->synchronize_AnalysisStats($analysis->stats);
498
    print STDERR ( ($analysis->stats()->status eq 'BLOCKED') ? 'x' : 'o');
499
  }
Leo Gordon's avatar
Leo Gordon committed
500
  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
501

Leo Gordon's avatar
Leo Gordon committed
502
  print STDERR ''.((time() - $start_time))." seconds to synchronize_hive\n\n";
503
}
504

505

506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527
=head2 safe_synchronize_AnalysisStats

  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
  Example    : $self->synchronize($analysisStats);
  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);
  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
528
                   ($stats->seconds_since_last_update < 3*60));
529 530 531 532

  # 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=" . $stats->analysis_id;
533
  #print("$sql\n");
534
  my $row_count = $self->dbc->do($sql);  
535
  return $stats unless($row_count == 1);        # return the un-updated status if locked
536
  #printf("got sync_lock on analysis_stats(%d)\n", $stats->analysis_id);
537
  
538
      # since we managed to obtain the lock, let's go and perform the sync:
539
  $self->synchronize_AnalysisStats($stats);
540

541 542 543 544
  return $stats;
}


545
=head2 synchronize_AnalysisStats
546

547 548
  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
  Example    : $self->synchronize($analysisStats);
549
  Description: Queries the job and worker tables to get summary counts
550 551 552 553 554
               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
555

556
=cut
557

558
sub synchronize_AnalysisStats {
559
  my $self = shift;
560
  my $analysisStats = shift;
561

562
  return $analysisStats unless($analysisStats);
563
  return $analysisStats unless($analysisStats->analysis_id);
564

565
  $analysisStats->refresh(); ## Need to get the new hive_capacity for dynamic analyses
566 567
  my $hive_capacity = $analysisStats->hive_capacity;

568 569
  if($self->db->hive_use_triggers()) {

570 571
            my $job_count = $analysisStats->ready_job_count();
            my $required_workers = $hive_capacity && POSIX::ceil( $job_count / $analysisStats->get_or_estimate_batch_size() );
572

573 574
                # adjust_stats_for_living_workers:
            if($hive_capacity > 0) {
575
                my $unfulfilled_capacity = $hive_capacity - $analysisStats->num_running_workers();
576

577 578
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
579 580 581 582 583 584
                }
            }
            $analysisStats->num_required_workers( $required_workers );

  } else {
      $analysisStats->total_job_count(0);
585 586
      $analysisStats->semaphored_job_count(0);
      $analysisStats->ready_job_count(0);
587 588 589
      $analysisStats->done_job_count(0);
      $analysisStats->failed_job_count(0);
      $analysisStats->num_required_workers(0);
590

591
            # ask for analysis_id to force MySQL to use existing index on (analysis_id, status)
592
      my $sql = "SELECT analysis_id, status, count(*) FROM job WHERE analysis_id=? GROUP BY analysis_id, status";
593 594
      my $sth = $self->prepare($sql);
      $sth->execute($analysisStats->analysis_id);
595

596 597 598
      my $done_here       = 0;
      my $done_elsewhere  = 0;
      my $total_job_count = 0;
599
      while (my ($dummy_analysis_id, $status, $job_count)=$sth->fetchrow_array()) {
600
    # print STDERR "$status: $job_count\n";
601

602
        $total_job_count += $job_count;
603

604
        if($status eq 'READY') {
605
            $analysisStats->ready_job_count($job_count);
606

607
            my $required_workers = $hive_capacity && POSIX::ceil( $job_count / $analysisStats->get_or_estimate_batch_size() );
608 609 610

                # adjust_stats_for_living_workers:
            if($hive_capacity > 0) {
611
                my $unfulfilled_capacity = $hive_capacity - $self->count_running_workers( $analysisStats->analysis_id() );
612

613 614
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
615
                }
616
            }
617 618
            $analysisStats->num_required_workers( $required_workers );

619 620
        } elsif($status eq 'SEMAPHORED') {
            $analysisStats->semaphored_job_count($job_count);
621
        } elsif($status eq 'DONE') {
622
            $done_here = $job_count;
623
        } elsif($status eq 'PASSED_ON') {
624 625 626
            $done_elsewhere = $job_count;
        } elsif ($status eq 'FAILED') {
            $analysisStats->failed_job_count($job_count);
627
        }
628
      } # /while
629
      $sth->finish;
630

631 632
      $analysisStats->total_job_count( $total_job_count );
      $analysisStats->done_job_count( $done_here + $done_elsewhere );
633
  } # /unless $self->{'_hive_use_triggers'}
634

635
  $analysisStats->check_blocking_control_rules();
636

637 638 639
  if($analysisStats->status ne 'BLOCKED') {
    $analysisStats->determine_status();
  }
640

641
  # $analysisStats->sync_lock(0); ## do we perhaps need it here?
642
  $analysisStats->update;  #update and release sync_lock
643

644
  return $analysisStats;
645
}
646 647


648 649 650 651 652 653 654 655 656 657 658 659 660 661
=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
662
sub get_num_failed_analyses {
663
    my ($self, $filter_analysis) = @_;
664

665 666 667 668 669 670 671 672 673 674 675 676
    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;
        }
    }
677

678
    return $filter_analysis ? $filter_analysis_failed : scalar(@$failed_analyses);
679 680 681
}


682 683
sub get_hive_current_load {
  my $self = shift;
684 685 686
  my $sql = "SELECT sum(1/s.hive_capacity) FROM worker w, analysis_stats s ".
            "WHERE w.analysis_id=s.analysis_id and w.cause_of_death ='' ".
            "AND s.hive_capacity>0";
687 688 689 690 691
  my $sth = $self->prepare($sql);
  $sth->execute();
  (my $load)=$sth->fetchrow_array();
  $sth->finish;
  $load=0 unless($load);
692
#  print("current hive load = $load\n");
693 694 695 696
  return $load;
}


697 698 699 700 701 702 703 704 705 706 707 708
sub count_running_workers {
    my ($self, $analysis_id) = @_;

    my $sql = "SELECT count(*) FROM worker WHERE cause_of_death =''"
        . ($analysis_id ? " AND analysis_id='$analysis_id'" : '');

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

    return $running_workers_count || 0;
709 710
}

711

712
=head2 schedule_workers
713

714
  Arg[1]     : Bio::EnsEMBL::Hive::Analysis object (optional)
715
  Example    : $count = $queen->schedule_workers();
716 717 718
  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
719
               If Arg[1] is defined, does it only for the given analysis.
720
  Exceptions : none
721
  Caller     : beekeepers and other external processes
722 723 724

=cut

725
sub schedule_workers {
726
  my ($self, $filter_analysis, $orig_pending_by_rc_name, $available_submit_limit) = @_;
727

728 729 730
  my @suitable_analyses         = $filter_analysis
                                    ? ( $filter_analysis->stats )
                                    : @{ $self->db->get_AnalysisStatsAdaptor->fetch_all_by_suitability_rc_id() };
731

732 733 734 735
  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;
736

737
  return {} unless(@suitable_analyses);
738

739
  my %pending_by_rc_name        = %{ $orig_pending_by_rc_name || {} };  # NB: make our own copy to be able to modify it
740
  my $total_workers_to_run      = 0;
741
  my %workers_to_run_by_rc_name = ();
742
  my $available_load            = 1.0 - $self->get_hive_current_load();
743

744

745
  foreach my $analysis_stats (@suitable_analyses) {
746 747
    last if ($available_load <= 0.0);
    last if (defined($available_submit_limit) and !$available_submit_limit);
748

Leo Gordon's avatar
Leo Gordon committed
749
        #digging deeper under the surface so need to sync
750
    if(($analysis_stats->status eq 'LOADING') or ($analysis_stats->status eq 'BLOCKED') or ($analysis_stats->status eq 'ALL_CLAIMED')) {
751 752 753 754 755 756
      $self->synchronize_AnalysisStats($analysis_stats);
    }

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

757
        # 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
758
    my $workers_this_analysis = $analysis_stats->num_required_workers;
759

760 761 762 763 764 765
    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
766

767 768 769 770 771
    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
772 773
        }

774
        $available_load -= 1.0*$workers_this_analysis/$hive_capacity;
775
    }
Leo Gordon's avatar
Leo Gordon committed
776

777
    my $curr_rc_name    = $analysis_id2rc_name{ $analysis_stats->analysis_id };
778

779 780 781 782 783 784
    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;
785 786
    }

787 788
    next unless($workers_this_analysis);    # do not autovivify the hash by a zero

789
    $total_workers_to_run += $workers_this_analysis;
790
    $workers_to_run_by_rc_name{ $curr_rc_name } += $workers_this_analysis;
791
    $analysis_stats->print_stats();
792
    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);
793
  }
794

795
  printf("Scheduler suggests adding a total of %d workers [%1.5f hive_load remaining]\n", $total_workers_to_run, $available_load);
796
  return \%workers_to_run_by_rc_name;
Leo Gordon's avatar
Leo Gordon committed
797 798
}

799

800
sub schedule_workers_resync_if_necessary {
801 802 803
    my ($self, $valley, $analysis) = @_;

    my $meadow              = $valley->get_current_meadow();
Leo Gordon's avatar
Leo Gordon committed
804

805
    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() : {};
806 807
    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
808

809 810 811 812
    my $available_submit_limit = ($submit_limit and $meadow_limit)
                                    ? (($submit_limit<$meadow_limit) ? $submit_limit : $meadow_limit)
                                    : (defined($submit_limit) ? $submit_limit : $meadow_limit);

813
    my $workers_to_run_by_rc_name = $self->