Queen.pm 37.6 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';  # import 'destringify()'
68
use Bio::EnsEMBL::Hive::AnalysisJob;
Jessica Severin's avatar
Jessica Severin committed
69
use Bio::EnsEMBL::Hive::Worker;
70
use Bio::EnsEMBL::Hive::DBSQL::AnalysisCtrlRuleAdaptor;
Jessica Severin's avatar
Jessica Severin committed
71

Leo Gordon's avatar
Leo Gordon committed
72
use base ('Bio::EnsEMBL::DBSQL::BaseAdaptor');
73

Jessica Severin's avatar
Jessica Severin committed
74

75
############################
Jessica Severin's avatar
Jessica Severin committed
76
#
77
# PUBLIC API for Workers
Jessica Severin's avatar
Jessica Severin committed
78
#
79
############################
Jessica Severin's avatar
Jessica Severin committed
80 81 82

=head2 create_new_worker

83
  Arg [1]    : $analysis_id (optional)
Jessica Severin's avatar
Jessica Severin committed
84
  Example    :
85 86 87
  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.
88
               It creates an entry in the worker table, and returns a Worker object 
89 90
               based on that insert.  This guarantees that each worker registered
               in this queens hive is properly registered.
Jessica Severin's avatar
Jessica Severin committed
91 92 93 94 95 96 97
  Returntype : Bio::EnsEMBL::Hive::Worker
  Exceptions :
  Caller     :

=cut

sub create_new_worker {
98 99
  my ($self, @args) = @_;

100
  my (  $meadow_type, $meadow_name, $process_id, $exec_host,
101
        $rc_id, $logic_name, $analysis_id, $input_id, $job_id,
102
        $no_write, $debug, $worker_output_dir, $hive_output_dir, $job_limit, $life_span, $no_cleanup, $retry_throwing_jobs) =
103

104
 rearrange([qw(meadow_type meadow_name process_id exec_host
105
        rc_id logic_name analysis_id input_id job_id
106
        no_write debug worker_output_dir hive_output_dir job_limit life_span no_cleanup retry_throwing_jobs) ], @args);
107 108 109 110 111 112 113 114 115 116 117

    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
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
    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));
            $hive_output_dir='' unless(defined($hive_output_dir)); # make it defined but empty/false
        } 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 {
144 145 146 147 148
            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)) {
149 150 151 152 153 154
                $analysis_id = $job->analysis_id;
            } else {
                die "job_id '$job_id' could not be fetched from the database\n";
            }
        }
    }
155

156
  
157
  my $analysis_stats_adaptor = $self->db->get_AnalysisStatsAdaptor or return undef;
158 159
  my $analysisStats;
  if($analysis_id) {
160
    $analysisStats = $analysis_stats_adaptor->fetch_by_analysis_id($analysis_id);
161
    $self->safe_synchronize_AnalysisStats($analysisStats);
162
    #return undef unless(($analysisStats->status ne 'BLOCKED') and ($analysisStats->num_required_workers > 0));
163
  } else {
164 165 166 167 168
    if( $analysisStats = $self->_pick_best_analysis_for_new_worker($rc_id) ) {
        print "Scheduler picked analysis_id=".$analysisStats->analysis_id()." for the worker\n";
    } else {
        print "Scheduler failed to pick analysis_id for the worker\n";
    }
169
  }
170
  return undef unless($analysisStats);
171

172 173
  unless($job) {
    #go into autonomous mode
174 175 176 177 178

    if($self->get_hive_current_load() >= 1.1) {
        print "Hive is overloaded, can't create a worker\n";
        return;
    }
179
    if($analysisStats->status eq 'BLOCKED') {
180 181
      print "Analysis is BLOCKED, can't create workers\n";
      return;
182 183
    }
    if($analysisStats->status eq 'DONE') {
184 185
      print "Analysis is DONE, don't need to create workers\n";
      return;
186
    }
187

188 189
    $analysis_stats_adaptor->decrease_required_workers($analysisStats->analysis_id);

190
    $analysisStats->print_stats;
191
  }
192
  
193 194 195 196 197 198 199 200 201 202
    # 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);
  }

203
  my $sql = q{INSERT INTO worker 
204 205
              (born, last_check_in, meadow_type, meadow_name, process_id, host, analysis_id)
              VALUES (CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, ?,?,?,?,?)};
Jessica Severin's avatar
Jessica Severin committed
206 207

  my $sth = $self->prepare($sql);
208
  $sth->execute($meadow_type, $meadow_name, $process_id, $exec_host, $analysisStats->analysis_id);
Leo Gordon's avatar
Leo Gordon committed
209
  my $worker_id = $self->dbc->db_handle->last_insert_id(undef, undef, 'worker', 'worker_id');
Jessica Severin's avatar
Jessica Severin committed
210 211
  $sth->finish;

212
  my $worker = $self->fetch_by_dbID($worker_id);
Jessica Severin's avatar
Jessica Severin committed
213 214
  $worker=undef unless($worker and $worker->analysis);

215
  if($worker and $analysisStats) {
216
    $analysisStats->update_status('WORKING');
Jessica Severin's avatar
Jessica Severin committed
217
  }
218
  
219
  $worker->_specific_job($job) if($job);
220
  $worker->execute_writes(0) if($no_write);
221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246

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

    unless(defined($hive_output_dir)) {
        my $arrRef = $self->db->get_MetaContainer->list_value_by_key( 'hive_output_dir' );
        if( @$arrRef ) {
            $hive_output_dir = destringify($arrRef->[0]);
        } 
    }
    $worker->hive_output_dir($hive_output_dir);

    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);
    }

Jessica Severin's avatar
Jessica Severin committed
247 248 249 250 251 252 253
  return $worker;
}

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

  return unless($worker);
Leo Gordon's avatar
Leo Gordon committed
254

Leo Gordon's avatar
bugfix  
Leo Gordon committed
255 256
  my $cod = $worker->cause_of_death();

257
  my $sql = "UPDATE worker SET died=CURRENT_TIMESTAMP, last_check_in=CURRENT_TIMESTAMP";
258
  $sql .= " ,status='DEAD'";
Jessica Severin's avatar
Jessica Severin committed
259
  $sql .= " ,work_done='" . $worker->work_done . "'";
Leo Gordon's avatar
bugfix  
Leo Gordon committed
260
  $sql .= " ,cause_of_death='$cod'";
261
  $sql .= " WHERE worker_id='" . $worker->dbID ."'";
Jessica Severin's avatar
Jessica Severin committed
262

Leo Gordon's avatar
bugfix  
Leo Gordon committed
263
  $self->dbc->do( $sql );
264

265 266 267 268
  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
269
  if($cod eq 'NO_WORK') {
270
    $self->db->get_AnalysisStatsAdaptor->update_status($worker->analysis->dbID, 'ALL_CLAIMED');
271
  }
Leo Gordon's avatar
bugfix  
Leo Gordon committed
272 273 274 275
  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
276
    $self->db->get_AnalysisJobAdaptor->release_undone_jobs_from_worker($worker);
277
  }
278
  
279
  # re-sync the analysis_stats when a worker dies as part of dynamic sync system
280
  if($self->safe_synchronize_AnalysisStats($worker->analysis->stats)->status ne 'DONE') {
281 282
    # 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
283
    $self->db->get_AnalysisStatsAdaptor->increase_required_workers($worker->analysis->dbID);
284 285
  }

Jessica Severin's avatar
Jessica Severin committed
286 287
}

288
sub check_for_dead_workers {    # scans the whole Valley for lost Workers (but ignores unreachagle ones)
289
    my ($self, $valley, $check_buried_in_haste) = @_;
Leo Gordon's avatar
Leo Gordon committed
290

291
    warn "GarbageCollector:\tChecking for lost Workers...\n";
292

293 294 295 296
    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
297

298
    warn "GarbageCollector:\t[Queen:] we have ".scalar(@$queen_worker_list)." Workers alive.\n";
299

Leo Gordon's avatar
Leo Gordon committed
300 301
    foreach my $worker (@$queen_worker_list) {

302 303 304
        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
305
        } else {
306
            $worker_status_counts{$meadow_type}{'UNREACHABLE'}++;
307

308
            next;   # Worker is unreachable from this Valley
Leo Gordon's avatar
Leo Gordon committed
309 310
        }

311 312 313 314 315
        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'}++;
316

317
            $mt_and_pid_to_lost_worker{$meadow_type}{$process_id} = $worker;
318
        }
319 320 321 322 323 324
    }

        # 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";
    }
325

326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345
    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);
            }
346 347 348
        }
    }

349
        # the following bit is completely Meadow-agnostic and only restores database integrity:
Leo Gordon's avatar
Leo Gordon committed
350
    if($check_buried_in_haste) {
351
        warn "GarbageCollector:\tChecking for Workers buried in haste...\n";
352
        my $buried_in_haste_list = $self->fetch_all_dead_workers_with_jobs();
Leo Gordon's avatar
Leo Gordon committed
353
        if(my $bih_number = scalar(@$buried_in_haste_list)) {
354
            warn "GarbageCollector:\tfound $bih_number jobs, reclaiming.\n\n";
Leo Gordon's avatar
Leo Gordon committed
355 356 357
            if($bih_number) {
                my $job_adaptor = $self->db->get_AnalysisJobAdaptor();
                foreach my $worker (@$buried_in_haste_list) {
Leo Gordon's avatar
Leo Gordon committed
358
                    $job_adaptor->release_undone_jobs_from_worker($worker);
Leo Gordon's avatar
Leo Gordon committed
359 360 361
                }
            }
        } else {
362
            warn "GarbageCollector:\tfound none\n";
Leo Gordon's avatar
Leo Gordon committed
363 364 365
        }
    }
}
Jessica Severin's avatar
Jessica Severin committed
366 367


368 369 370
    # 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
371

372
    $self->dbc->do("UPDATE worker SET last_check_in=CURRENT_TIMESTAMP, status='".$worker->status."', work_done='".$worker->work_done."' WHERE worker_id='".$worker->dbID."'");
Jessica Severin's avatar
Jessica Severin committed
373 374 375
}


376
=head2 reset_job_by_dbID_and_sync
377

378
  Arg [1]: int $job_id
379
  Example: 
380
    my $job = $queen->reset_job_by_dbID_and_sync($job_id);
381
  Description: 
382
    For the specified job_id it will fetch just that job, 
383 384 385
    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 
386
    same job simultaneously (use only for debugging).
387
  Returntype : none
388
  Exceptions :
389
  Caller     : beekeeper.pl
390 391 392

=cut

393 394
sub reset_job_by_dbID_and_sync {
    my ($self, $job_id) = @_;
395

396 397 398 399 400 401
    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);
402 403 404
}


405 406 407 408 409 410
######################################
#
# Public API interface for beekeeper
#
######################################

411 412 413
    # Note: asking for Queen->fetch_overdue_workers(0) essentially means
    #       "fetch all workers known to the Queen not to be officially dead"
    #
414 415 416 417 418
sub fetch_overdue_workers {
  my ($self,$overdue_secs) = @_;

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

419 420 421 422
  my $constraint = "w.cause_of_death='' AND ".
                    ( ($self->dbc->driver eq 'sqlite')
                        ? "(strftime('%s','now')-strftime('%s',w.last_check_in))>$overdue_secs"
                        : "(UNIX_TIMESTAMP()-UNIX_TIMESTAMP(w.last_check_in))>$overdue_secs");
423 424 425
  return $self->_generic_fetch($constraint);
}

426 427
sub fetch_failed_workers {
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
428

429
  my $constraint = "w.cause_of_death='FATALITY' ";
430 431 432
  return $self->_generic_fetch($constraint);
}

433
sub fetch_all_dead_workers_with_jobs {
Leo Gordon's avatar
Leo Gordon committed
434 435
  my $self = shift;

436
  # SELECT w.* FROM worker h, job j WHERE w.worker_id=j.worker_id AND w.cause_of_death!='' AND j.status NOT IN ('DONE', 'READY','FAILED', 'PASSED_ON') GROUP BY w.worker_id
Leo Gordon's avatar
Leo Gordon committed
437

438 439
  my $constraint = "w.cause_of_death!='' ";
  my $join = [[['job', 'j'], " w.worker_id=j.worker_id AND j.status NOT IN ('DONE', 'READY', 'FAILED', 'PASSED_ON') GROUP BY w.worker_id"]];
Leo Gordon's avatar
Leo Gordon committed
440 441
  return $self->_generic_fetch($constraint, $join);
}
Jessica Severin's avatar
Jessica Severin committed
442

443 444
=head2 synchronize_hive

Leo Gordon's avatar
Leo Gordon committed
445
  Arg [1]    : $filter_analysis (optional)
446 447
  Example    : $queen->synchronize_hive();
  Description: Runs through all analyses in the system and synchronizes
448 449
              the analysis_stats summary with the states in the job 
              and worker tables.  Then follows by checking all the blocking rules
450 451 452 453 454 455 456
              and blocks/unblocks analyses as needed.
  Exceptions : none
  Caller     : general

=cut

sub synchronize_hive {
Leo Gordon's avatar
Leo Gordon committed
457
  my $self          = shift;
Leo Gordon's avatar
Leo Gordon committed
458
  my $filter_analysis = shift; # optional parameter
459

460
  my $start_time = time();
461

Leo Gordon's avatar
Leo Gordon committed
462
  my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
463

Leo Gordon's avatar
Leo Gordon committed
464
  print STDERR "\nSynchronizing the hive (".scalar(@$list_of_analyses)." analyses this time):\n";
Leo Gordon's avatar
Leo Gordon committed
465 466
  foreach my $analysis (@$list_of_analyses) {
    $self->synchronize_AnalysisStats($analysis->stats);
467
    print STDERR ( ($analysis->stats()->status eq 'BLOCKED') ? 'x' : 'o');
468
  }
Leo Gordon's avatar
Leo Gordon committed
469
  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
470

471 472 473 474 475 476
#  print STDERR "Checking blocking control rules:\n";
#  foreach my $analysis (@$list_of_analyses) {
#    my $open = $analysis->stats->check_blocking_control_rules();
#    print STDERR ($open ? 'o' : 'x');
#  }
#  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
477

Leo Gordon's avatar
Leo Gordon committed
478
  print STDERR ''.((time() - $start_time))." seconds to synchronize_hive\n\n";
479
}
480

481

482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503
=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
504
                   ($stats->seconds_since_last_update < 3*60));
505 506 507 508

  # 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=" . $stats->analysis_id;
509
  #print("$sql\n");
510
  my $row_count = $self->dbc->do($sql);  
511
  return $stats unless($row_count == 1);        # return the un-updated status if locked
512
  #printf("got sync_lock on analysis_stats(%d)\n", $stats->analysis_id);
513
  
514
      # since we managed to obtain the lock, let's go and perform the sync:
515
  $self->synchronize_AnalysisStats($stats);
516

517 518 519 520
  return $stats;
}


521
=head2 synchronize_AnalysisStats
522

523 524
  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
  Example    : $self->synchronize($analysisStats);
525
  Description: Queries the job and worker tables to get summary counts
526 527 528 529 530
               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
531

532
=cut
533

534
sub synchronize_AnalysisStats {
535
  my $self = shift;
536
  my $analysisStats = shift;
537

538
  return $analysisStats unless($analysisStats);
539
  return $analysisStats unless($analysisStats->analysis_id);
540

541
  $analysisStats->refresh(); ## Need to get the new hive_capacity for dynamic analyses
542 543
  my $hive_capacity = $analysisStats->hive_capacity;

544 545 546 547
  if($self->db->hive_use_triggers()) {

            my $job_count = $analysisStats->unclaimed_job_count();
            my $required_workers = POSIX::ceil( $job_count / $analysisStats->get_or_estimate_batch_size() );
548

549 550
                # adjust_stats_for_living_workers:
            if($hive_capacity > 0) {
551
                my $unfulfilled_capacity = $hive_capacity - $analysisStats->num_running_workers();
552

553 554
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
555 556 557 558 559 560 561 562 563 564
                }
            }
            $analysisStats->num_required_workers( $required_workers );

  } else {
      $analysisStats->total_job_count(0);
      $analysisStats->unclaimed_job_count(0);
      $analysisStats->done_job_count(0);
      $analysisStats->failed_job_count(0);
      $analysisStats->num_required_workers(0);
565

566
      my $sql = "SELECT status, semaphore_count, count(*) FROM job ".
567 568 569
                "WHERE analysis_id=? GROUP BY status, semaphore_count";
      my $sth = $self->prepare($sql);
      $sth->execute($analysisStats->analysis_id);
570 571


572 573 574 575
      my $done_here       = 0;
      my $done_elsewhere  = 0;
      my $total_job_count = 0;
      while (my ($status, $semaphore_count, $job_count)=$sth->fetchrow_array()) {
576
    # print STDERR "$status: $job_count\n";
577

578
        $total_job_count += $job_count;
579 580 581 582 583 584 585 586

        if(($status eq 'READY') and ($semaphore_count<=0)) {
            $analysisStats->unclaimed_job_count($job_count);

            my $required_workers = POSIX::ceil( $job_count / $analysisStats->get_or_estimate_batch_size() );

                # adjust_stats_for_living_workers:
            if($hive_capacity > 0) {
587
                my $unfulfilled_capacity = $hive_capacity - $self->count_running_workers( $analysisStats->analysis_id() );
588

589 590
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
591
                }
592
            }
593 594
            $analysisStats->num_required_workers( $required_workers );

595
        } elsif($status eq 'DONE') {
596
            $done_here = $job_count;
597
        } elsif($status eq 'PASSED_ON') {
598 599 600
            $done_elsewhere = $job_count;
        } elsif ($status eq 'FAILED') {
            $analysisStats->failed_job_count($job_count);
601
        }
602
      } # /while
603
      $sth->finish;
604

605 606
      $analysisStats->total_job_count( $total_job_count );
      $analysisStats->done_job_count( $done_here + $done_elsewhere );
607
  } # /unless $self->{'_hive_use_triggers'}
608

609
  $analysisStats->check_blocking_control_rules();
610

611 612 613
  if($analysisStats->status ne 'BLOCKED') {
    $analysisStats->determine_status();
  }
614

615
  # $analysisStats->sync_lock(0); ## do we perhaps need it here?
616
  $analysisStats->update;  #update and release sync_lock
617

618
  return $analysisStats;
619
}
620 621


Leo Gordon's avatar
Leo Gordon committed
622 623
sub get_num_failed_analyses {
  my ($self, $analysis) = @_;
624 625

  my $statsDBA = $self->db->get_AnalysisStatsAdaptor;
Leo Gordon's avatar
Leo Gordon committed
626
  my $failed_analyses = $statsDBA->fetch_by_statuses(['FAILED']);
627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648
  if ($analysis) {
    foreach my $this_failed_analysis (@$failed_analyses) {
      if ($this_failed_analysis->analysis_id == $analysis->dbID) {
        print "#########################################################\n",
            " Too many jobs failed for analysis ".$analysis->logic_name.". FAIL!!\n",
            "#########################################################\n\n";
        return 1;
      }
    }
    return 0;
  }

  if (@$failed_analyses) {
    print "##################################################\n",
        " Too many failed jobs. FAIL!!\n",
        "##################################################\n";
  }

  return scalar(@$failed_analyses);
}


649 650
sub get_hive_current_load {
  my $self = shift;
651 652 653
  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";
654 655 656 657 658
  my $sth = $self->prepare($sql);
  $sth->execute();
  (my $load)=$sth->fetchrow_array();
  $sth->finish;
  $load=0 unless($load);
Leo Gordon's avatar
Leo Gordon committed
659
  print("current hive load = $load\n");
660 661 662 663
  return $load;
}


664 665 666 667 668 669 670 671 672 673 674 675
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;
676 677
}

678

679
=head2 schedule_workers
680

681
  Arg[1]     : Bio::EnsEMBL::Analysis object (optional)
682
  Example    : $count = $queen->schedule_workers();
683 684 685
  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
686
               If Arg[1] is defined, does it only for the given analysis.
687
  Exceptions : none
688
  Caller     : beekeepers and other external processes
689 690 691

=cut

692 693
sub schedule_workers {
  my ($self, $filter_analysis, $orig_pending_by_rc_id, $available_submit_limit) = @_;
694

695 696 697 698
  my $statsDBA                      = $self->db->get_AnalysisStatsAdaptor;
  my $clearly_needed_analyses       = $statsDBA->fetch_by_needed_workers(undef);
  my $potentially_needed_analyses   = $statsDBA->fetch_by_statuses(['LOADING', 'BLOCKED', 'ALL_CLAIMED']);
  my @all_analyses                  = (@$clearly_needed_analyses, @$potentially_needed_analyses);
699

700
  return {} unless(@all_analyses);
701

702 703 704 705
  my %pending_by_rc_id          = %{ $orig_pending_by_rc_id || {} };
  my $total_workers_to_run      = 0;
  my %workers_to_run_by_rc_id   = ();
  my $available_load            = 1.0 - $self->get_hive_current_load();
706

Leo Gordon's avatar
Leo Gordon committed
707
  foreach my $analysis_stats (@all_analyses) {
708 709
    last if ($available_load <= 0.0);
    last if (defined($available_submit_limit) and !$available_submit_limit);
Leo Gordon's avatar
Leo Gordon committed
710
    next if (defined $filter_analysis && $filter_analysis->dbID != $analysis_stats->analysis_id);
711

Leo Gordon's avatar
Leo Gordon committed
712
        #digging deeper under the surface so need to sync
713
    if(($analysis_stats->status eq 'LOADING') or ($analysis_stats->status eq 'BLOCKED') or ($analysis_stats->status eq 'ALL_CLAIMED')) {
714 715 716 717 718 719
      $self->synchronize_AnalysisStats($analysis_stats);
    }

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

720
        # 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
721
    my $workers_this_analysis = $analysis_stats->num_required_workers;
722

723 724 725 726 727 728
    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
729

730 731 732 733 734
    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
735 736
        }

737
        $available_load -= 1.0*$workers_this_analysis/$hive_capacity;
738
    }
Leo Gordon's avatar
Leo Gordon committed
739

740
    my $curr_rc_id = $analysis_stats->resource_class_id;
741 742 743 744 745 746 747
    if($pending_by_rc_id{ $curr_rc_id }) {                              # per-rc_id capping by pending processes, if available
        my $pending_this_analysis = ($pending_by_rc_id{ $curr_rc_id } < $workers_this_analysis) ? $pending_by_rc_id{ $curr_rc_id } : $workers_this_analysis;

        $workers_this_analysis              -= $pending_this_analysis;
        $pending_by_rc_id{ $curr_rc_id }    -= $pending_this_analysis;
    }

748 749
    next unless($workers_this_analysis);    # do not autovivify the hash by a zero

750 751 752
    $total_workers_to_run += $workers_this_analysis;
    $workers_to_run_by_rc_id{ $curr_rc_id } += $workers_this_analysis;
    $analysis_stats->print_stats();
753
    printf("Scheduler suggests adding %d more workers of resource_class_id=%d for analysis_id=%d [%1.3f hive_load remaining]\n", $workers_this_analysis, $curr_rc_id, $analysis_stats->analysis_id, $available_load);
754
  }
755

756
  printf("Scheduler suggests adding a total of %d workers [%1.5f hive_load remaining]\n", $total_workers_to_run, $available_load);
757
  return \%workers_to_run_by_rc_id;
Leo Gordon's avatar
Leo Gordon committed
758 759
}

760

761
sub schedule_workers_resync_if_necessary {
762 763 764
    my ($self, $valley, $analysis) = @_;

    my $meadow              = $valley->get_current_meadow();
Leo Gordon's avatar
Leo Gordon committed
765

766 767 768
    my $pending_by_rc_id    = ($meadow->can('count_pending_workers_by_rc_id') and $meadow->config_get('PendingAdjust')) ? $meadow->count_pending_workers_by_rc_id() : {};
    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
769

770 771 772 773
    my $available_submit_limit = ($submit_limit and $meadow_limit)
                                    ? (($submit_limit<$meadow_limit) ? $submit_limit : $meadow_limit)
                                    : (defined($submit_limit) ? $submit_limit : $meadow_limit);

774
    my $workers_to_run_by_rc_id = $self->schedule_workers($analysis, $pending_by_rc_id, $available_submit_limit);
775

776
    unless( keys %$workers_to_run_by_rc_id or $self->get_hive_current_load() or $self->count_running_workers() ) {
Leo Gordon's avatar
Leo Gordon committed
777 778
        print "*** nothing is running and nothing to do (according to analysis_stats) => perform a hard resync\n" ;

779
        $self->check_for_dead_workers($valley, 1);
780
        $self->synchronize_hive($analysis);
Leo Gordon's avatar
Leo Gordon committed
781

782
        $workers_to_run_by_rc_id = $self->schedule_workers($analysis, $pending_by_rc_id, $available_submit_limit);
Leo Gordon's avatar
Leo Gordon committed
783 784
    }

785
    return $workers_to_run_by_rc_id;
786 787
}

788

Leo Gordon's avatar
Leo Gordon committed
789
sub get_remaining_jobs_show_hive_progress {
790
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
791
  my $sql = "SELECT sum(done_job_count), sum(failed_job_count), sum(total_job_count), ".
792 793
            "sum(unclaimed_job_count * analysis_stats.avg_msec_per_job)/1000/60/60 ".
            "FROM analysis_stats";
794 795
  my $sth = $self->prepare($sql);
  $sth->execute();
796
  my ($done, $failed, $total, $cpuhrs) = $sth->fetchrow_array();
797
  $sth->finish;
798 799 800 801 802 803 804

  $done   ||= 0;
  $failed ||= 0;
  $total  ||= 0;
  my $completed = $total
    ? ((100.0 * ($done+$failed))/$total)
    : 0.0;
805 806 807
  my $remaining = $total - $done - $failed;
  printf("hive %1.3f%% complete (< %1.3f CPU_hrs) (%d todo + %d done + %d failed = %d total)\n", 
          $completed, $cpuhrs, $remaining, $done, $failed, $total);
Leo Gordon's avatar
Leo Gordon committed
808
  return $remaining;
809 810
}

811

Leo Gordon's avatar
Leo Gordon committed
812
sub print_analysis_status {
Leo Gordon's avatar
Leo Gordon committed
813
    my ($self, $filter_analysis) = @_;
814

Leo Gordon's avatar
Leo Gordon committed
815
    my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
816 817 818
    foreach my $analysis (sort {$a->dbID <=> $b->dbID} @$list_of_analyses) {
        $analysis->stats->print_stats($self->{'verbose_stats'});
    }
819 820 821
}


Leo Gordon's avatar
Leo Gordon committed
822
sub print_running_worker_counts {
823
  my $self = shift;
824

Leo Gordon's avatar
Leo Gordon committed
825
  print "\n===== Stats of live Workers according to the Queen: ======\n";
Leo Gordon's avatar