Queen.pm 33.9 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
101
  my (  $meadow_type, $process_id, $exec_host,
        $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
105

 rearrange([qw(meadow_type process_id exec_host
        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
144
145
146
147
148
149
150
151
    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 {
            print "fetching job for job_id '$job_id'\n";
            if($job = $self->reset_and_fetch_job_by_dbID($job_id)) {
                $analysis_id = $job->analysis_id;
            } else {
                die "job_id '$job_id' could not be fetched from the database\n";
            }
        }
    }
152

153
  
154
  my $analysis_stats_adaptor = $self->db->get_AnalysisStatsAdaptor or return undef;
155
156
  my $analysisStats;
  if($analysis_id) {
157
    $analysisStats = $analysis_stats_adaptor->fetch_by_analysis_id($analysis_id);
158
    $self->safe_synchronize_AnalysisStats($analysisStats);
159
    #return undef unless(($analysisStats->status ne 'BLOCKED') and ($analysisStats->num_required_workers > 0));
160
  } else {
Leo Gordon's avatar
Leo Gordon committed
161
    $analysisStats = $self->_pick_best_analysis_for_new_worker($rc_id);
162
  }
163
  return undef unless($analysisStats);
164

165
166
  unless($job) {
    #go into autonomous mode
167
168
169
170
171

    if($self->get_hive_current_load() >= 1.1) {
        print "Hive is overloaded, can't create a worker\n";
        return;
    }
172
    if($analysisStats->status eq 'BLOCKED') {
173
174
      print "Analysis is BLOCKED, can't create workers\n";
      return;
175
176
    }
    if($analysisStats->status eq 'DONE') {
177
178
      print "Analysis is DONE, don't need to create workers\n";
      return;
179
    }
180

181
182
    $analysis_stats_adaptor->decrease_required_workers($analysisStats->analysis_id);

183
    $analysisStats->print_stats;
184
  }
185
  
186
187
188
189
190
191
192
193
194
195
    # 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);
  }

196
197
  my $sql = q{INSERT INTO worker 
              (born, last_check_in, meadow_type, process_id, host, analysis_id)
198
              VALUES (CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, ?,?,?,?)};
Jessica Severin's avatar
Jessica Severin committed
199
200

  my $sth = $self->prepare($sql);
201
  $sth->execute($meadow_type, $process_id, $exec_host, $analysisStats->analysis_id);
Leo Gordon's avatar
Leo Gordon committed
202
  my $worker_id = $self->dbc->db_handle->last_insert_id(undef, undef, 'worker', 'worker_id');
Jessica Severin's avatar
Jessica Severin committed
203
204
  $sth->finish;

205
  my $worker = $self->fetch_by_dbID($worker_id);
Jessica Severin's avatar
Jessica Severin committed
206
207
  $worker=undef unless($worker and $worker->analysis);

208
  if($worker and $analysisStats) {
209
    $analysisStats->update_status('WORKING');
Jessica Severin's avatar
Jessica Severin committed
210
  }
211
  
212
  $worker->_specific_job($job) if($job);
213
  $worker->execute_writes(0) if($no_write);
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239

    $worker->debug($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
240
241
242
243
244
245
246
  return $worker;
}

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

  return unless($worker);
Leo Gordon's avatar
Leo Gordon committed
247

Leo Gordon's avatar
bugfix  
Leo Gordon committed
248
249
  my $cod = $worker->cause_of_death();

250
  my $sql = "UPDATE worker SET died=CURRENT_TIMESTAMP, last_check_in=CURRENT_TIMESTAMP";
251
  $sql .= " ,status='DEAD'";
Jessica Severin's avatar
Jessica Severin committed
252
  $sql .= " ,work_done='" . $worker->work_done . "'";
Leo Gordon's avatar
bugfix  
Leo Gordon committed
253
  $sql .= " ,cause_of_death='$cod'";
254
  $sql .= " WHERE worker_id='" . $worker->dbID ."'";
Jessica Severin's avatar
Jessica Severin committed
255

Leo Gordon's avatar
bugfix  
Leo Gordon committed
256
  $self->dbc->do( $sql );
257

258
259
260
261
  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
262
  if($cod eq 'NO_WORK') {
263
    $self->db->get_AnalysisStatsAdaptor->update_status($worker->analysis->dbID, 'ALL_CLAIMED');
264
  }
Leo Gordon's avatar
bugfix  
Leo Gordon committed
265
266
267
268
  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
269
    $self->db->get_AnalysisJobAdaptor->release_undone_jobs_from_worker($worker);
270
  }
271
  
272
  # re-sync the analysis_stats when a worker dies as part of dynamic sync system
273
  if($self->safe_synchronize_AnalysisStats($worker->analysis->stats)->status ne 'DONE') {
274
275
    # 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
276
    $self->db->get_AnalysisStatsAdaptor->increase_required_workers($worker->analysis->dbID);
277
278
  }

Jessica Severin's avatar
Jessica Severin committed
279
280
}

281
sub check_for_dead_workers {    # a bit counter-intuitively only looks for current meadow's workers, not all of the dead workers.
Leo Gordon's avatar
Leo Gordon committed
282
283
    my ($self, $meadow, $check_buried_in_haste) = @_;

284
    my $worker_status_hash    = $meadow->status_of_all_our_workers();
Leo Gordon's avatar
Leo Gordon committed
285
    my %worker_status_summary = ();
286
    my $queen_worker_list     = $self->fetch_overdue_workers(0);    # maybe it should return a {meadow->worker_count} hash instead?
Leo Gordon's avatar
Leo Gordon committed
287
288
289

    print "====== Live workers according to    Queen:".scalar(@$queen_worker_list).", Meadow:".scalar(keys %$worker_status_hash)."\n";

290
291
    my %gc_wpid_to_worker = ();

Leo Gordon's avatar
Leo Gordon committed
292
293
294
    foreach my $worker (@$queen_worker_list) {
        next unless($meadow->responsible_for_worker($worker));

295
296
        my $process_id = $worker->process_id();
        if(my $status = $worker_status_hash->{$process_id}) { # can be RUN|PEND|xSUSP
Leo Gordon's avatar
Leo Gordon committed
297
298
299
            $worker_status_summary{$status}++;
        } else {
            $worker_status_summary{'AWOL'}++;
300

301
            $gc_wpid_to_worker{$process_id} = $worker;
Leo Gordon's avatar
Leo Gordon committed
302
303
304
305
        }
    }
    print "\t".join(', ', map { "$_:$worker_status_summary{$_}" } keys %worker_status_summary)."\n\n";

306
307
308
309
310
311
312
313
314
315
316
    if(my $total_lost = scalar(keys %gc_wpid_to_worker)) {
        warn "GarbageCollector: Discovered $total_lost lost workers\n";

        my $wpid_to_cod = {};
        if(UNIVERSAL::can($meadow, 'find_out_causes')) {
            $wpid_to_cod = $meadow->find_out_causes( keys %gc_wpid_to_worker );
            my $lost_with_known_cod = scalar(keys %$wpid_to_cod);
            warn "GarbageCollector: Found why $lost_with_known_cod of them died\n";
        }

        warn "GarbageCollector: Releasing the jobs\n";
317
318
        while(my ($process_id, $worker) = each %gc_wpid_to_worker) {
            $worker->cause_of_death( $wpid_to_cod->{$process_id} || 'FATALITY');
319
320
321
322
            $self->register_worker_death($worker);
        }
    }

Leo Gordon's avatar
Leo Gordon committed
323
324
    if($check_buried_in_haste) {
        print "====== Checking for workers buried in haste... ";
325
        my $buried_in_haste_list = $self->fetch_all_dead_workers_with_jobs();
Leo Gordon's avatar
Leo Gordon committed
326
327
328
329
330
        if(my $bih_number = scalar(@$buried_in_haste_list)) {
            print "$bih_number, reclaiming jobs.\n\n";
            if($bih_number) {
                my $job_adaptor = $self->db->get_AnalysisJobAdaptor();
                foreach my $worker (@$buried_in_haste_list) {
Leo Gordon's avatar
Leo Gordon committed
331
                    $job_adaptor->release_undone_jobs_from_worker($worker);
Leo Gordon's avatar
Leo Gordon committed
332
333
334
335
336
337
338
                }
            }
        } else {
            print "none\n";
        }
    }
}
Jessica Severin's avatar
Jessica Severin committed
339
340


341
342
343
    # 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
344

345
    $self->dbc->do("UPDATE worker SET last_check_in=CURRENT_TIMESTAMP, status='".$worker->status."', work_done='".$worker->work_done."' WHERE worker_id='".$worker->dbID."'");
Jessica Severin's avatar
Jessica Severin committed
346
347
348
}


349
=head2 reset_and_fetch_job_by_dbID
350

351
  Arg [1]: int $job_id
352
  Example: 
353
    my $job = $queen->reset_and_fetch_job_by_dbID($job_id);
354
  Description: 
355
    For the specified job_id it will fetch just that job, 
356
357
358
    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 
359
360
    same job simultaneously (use only for debugging).
  Returntype : 
361
    Bio::EnsEMBL::Hive::AnalysisJob object
362
  Exceptions :
363
  Caller     : beekeepers, runWorker.pl scripts
364
365
366

=cut

367
368
sub reset_and_fetch_job_by_dbID {
  my $self = shift;
369
  my $job_id = shift;
370
  
371
  my $jobDBA = $self->db->get_AnalysisJobAdaptor;
372
  $jobDBA->reset_job_by_dbID($job_id); 
373

374
  my $job = $jobDBA->fetch_by_dbID($job_id); 
375
376
  my $stats = $self->db->get_AnalysisStatsAdaptor->fetch_by_analysis_id($job->analysis_id);
  $self->synchronize_AnalysisStats($stats);
377
  
378
379
380
381
  return $job;
}


382
383
384
385
386
387
######################################
#
# Public API interface for beekeeper
#
######################################

388
389
390
    # Note: asking for Queen->fetch_overdue_workers(0) essentially means
    #       "fetch all workers known to the Queen not to be officially dead"
    #
391
392
393
394
395
sub fetch_overdue_workers {
  my ($self,$overdue_secs) = @_;

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

396
397
398
399
  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");
400
401
402
  return $self->_generic_fetch($constraint);
}

403
404
sub fetch_failed_workers {
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
405

406
  my $constraint = "w.cause_of_death='FATALITY' ";
407
408
409
  return $self->_generic_fetch($constraint);
}

410
sub fetch_all_dead_workers_with_jobs {
Leo Gordon's avatar
Leo Gordon committed
411
412
  my $self = shift;

413
  # 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
414

415
416
  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
417
418
  return $self->_generic_fetch($constraint, $join);
}
Jessica Severin's avatar
Jessica Severin committed
419

420
421
=head2 synchronize_hive

Leo Gordon's avatar
Leo Gordon committed
422
  Arg [1]    : $filter_analysis (optional)
423
424
  Example    : $queen->synchronize_hive();
  Description: Runs through all analyses in the system and synchronizes
425
426
              the analysis_stats summary with the states in the job 
              and worker tables.  Then follows by checking all the blocking rules
427
428
429
430
431
432
433
              and blocks/unblocks analyses as needed.
  Exceptions : none
  Caller     : general

=cut

sub synchronize_hive {
Leo Gordon's avatar
Leo Gordon committed
434
  my $self          = shift;
Leo Gordon's avatar
Leo Gordon committed
435
  my $filter_analysis = shift; # optional parameter
436

437
  my $start_time = time();
438

Leo Gordon's avatar
Leo Gordon committed
439
  my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
440

Leo Gordon's avatar
Leo Gordon committed
441
  print STDERR "\nSynchronizing the hive (".scalar(@$list_of_analyses)." analyses this time):\n";
Leo Gordon's avatar
Leo Gordon committed
442
443
  foreach my $analysis (@$list_of_analyses) {
    $self->synchronize_AnalysisStats($analysis->stats);
Leo Gordon's avatar
Leo Gordon committed
444
    print STDERR '.';
445
  }
Leo Gordon's avatar
Leo Gordon committed
446
  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
447

Leo Gordon's avatar
Leo Gordon committed
448
  print STDERR "Checking blocking control rules:\n";
Leo Gordon's avatar
Leo Gordon committed
449
  foreach my $analysis (@$list_of_analyses) {
Leo Gordon's avatar
Leo Gordon committed
450
451
    my $open = $analysis->stats->check_blocking_control_rules();
    print STDERR ($open ? 'o' : 'x');
452
  }
Leo Gordon's avatar
Leo Gordon committed
453
  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
454

Leo Gordon's avatar
Leo Gordon committed
455
  print STDERR ''.((time() - $start_time))." seconds to synchronize_hive\n\n";
456
}
457

458

459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
=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
481
                   ($stats->seconds_since_last_update < 3*60));
482
483
484
485

  # 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=" . $stats->analysis_id;
486
  #print("$sql\n");
487
488
  my $row_count = $self->dbc->do($sql);  
  return $stats unless($row_count == 1);
489
  #printf("got sync_lock on analysis_stats(%d)\n", $stats->analysis_id);
490
491
492
  
  #OK have the lock, go and do the sync
  $self->synchronize_AnalysisStats($stats);
493

494
495
496
497
  return $stats;
}


498
=head2 synchronize_AnalysisStats
499

500
501
  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
  Example    : $self->synchronize($analysisStats);
502
  Description: Queries the job and worker tables to get summary counts
503
504
505
506
507
               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
508

509
=cut
510

511
sub synchronize_AnalysisStats {
512
  my $self = shift;
513
  my $analysisStats = shift;
514

515
  return $analysisStats unless($analysisStats);
516
  return $analysisStats unless($analysisStats->analysis_id);
517

518
  $analysisStats->refresh(); ## Need to get the new hive_capacity for dynamic analyses
519
520
  my $hive_capacity = $analysisStats->hive_capacity;

521
522
523
524
  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() );
525

526
527
                # adjust_stats_for_living_workers:
            if($hive_capacity > 0) {
528
                my $unfulfilled_capacity = $hive_capacity - $analysisStats->num_running_workers();
529

530
531
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
532
533
534
535
536
537
538
539
540
541
                }
            }
            $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);
542

543
      my $sql = "SELECT status, semaphore_count, count(*) FROM job ".
544
545
546
                "WHERE analysis_id=? GROUP BY status, semaphore_count";
      my $sth = $self->prepare($sql);
      $sth->execute($analysisStats->analysis_id);
547
548


549
550
551
552
      my $done_here       = 0;
      my $done_elsewhere  = 0;
      my $total_job_count = 0;
      while (my ($status, $semaphore_count, $job_count)=$sth->fetchrow_array()) {
553
    # print STDERR "$status: $job_count\n";
554

555
        $total_job_count += $job_count;
556
557
558
559
560
561
562
563

        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) {
564
                my $unfulfilled_capacity = $hive_capacity - $self->count_running_workers( $analysisStats->analysis_id() );
565

566
567
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
568
                }
569
            }
570
571
            $analysisStats->num_required_workers( $required_workers );

572
        } elsif($status eq 'DONE') {
573
            $done_here = $job_count;
574
        } elsif($status eq 'PASSED_ON') {
575
576
577
            $done_elsewhere = $job_count;
        } elsif ($status eq 'FAILED') {
            $analysisStats->failed_job_count($job_count);
578
        }
579
      } # /while
580
      $sth->finish;
581

582
583
      $analysisStats->total_job_count( $total_job_count );
      $analysisStats->done_job_count( $done_here + $done_elsewhere );
584
  } # /unless $self->{'_hive_use_triggers'}
585

586
  $analysisStats->check_blocking_control_rules();
587

588
589
590
  if($analysisStats->status ne 'BLOCKED') {
    $analysisStats->determine_status();
  }
591

592
  $analysisStats->update;  #update and release sync_lock
593

594
  return $analysisStats;
595
}
596
597


Leo Gordon's avatar
Leo Gordon committed
598
599
sub get_num_failed_analyses {
  my ($self, $analysis) = @_;
600
601

  my $statsDBA = $self->db->get_AnalysisStatsAdaptor;
Leo Gordon's avatar
Leo Gordon committed
602
  my $failed_analyses = $statsDBA->fetch_by_statuses(['FAILED']);
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
  if ($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);
}


625
626
sub get_hive_current_load {
  my $self = shift;
627
628
629
  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";
630
631
632
633
634
  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
635
  print("current hive load = $load\n");
636
637
638
639
  return $load;
}


640
641
642
643
644
645
646
647
648
649
650
651
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;
652
653
}

654

655
656
=head2 get_num_needed_workers

657
  Arg[1]     : Bio::EnsEMBL::Analysis object (optional)
658
659
660
661
  Example    : $count = $queen->get_num_needed_workers();
  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
662
               If Arg[1] is defined, does it only for the given analysis.
663
  Exceptions : none
664
  Caller     : beekeepers and other external processes
665
666
667

=cut

668
sub get_num_needed_workers {
Leo Gordon's avatar
Leo Gordon committed
669
  my ($self, $filter_analysis) = @_;
670

671
  my $statsDBA = $self->db->get_AnalysisStatsAdaptor;
Leo Gordon's avatar
Leo Gordon committed
672
673
674
  my $clearly_needed_analyses     = $statsDBA->fetch_by_needed_workers(undef,$self->{maximise_concurrency});
  my $potentially_needed_analyses = $statsDBA->fetch_by_statuses(['LOADING', 'BLOCKED']);
  my @all_analyses = (@$clearly_needed_analyses, @$potentially_needed_analyses);
675

Leo Gordon's avatar
Leo Gordon committed
676
  return 0 unless(@all_analyses);
677

Leo Gordon's avatar
Leo Gordon committed
678
  my $available_load = 1.0 - $self->get_hive_current_load();
679

Leo Gordon's avatar
Leo Gordon committed
680
  return 0 if($available_load <=0.0);
681

Leo Gordon's avatar
Leo Gordon committed
682
683
  my $total_workers = 0;
  my %rc2workers = ();
684

Leo Gordon's avatar
Leo Gordon committed
685
686
  foreach my $analysis_stats (@all_analyses) {
    next if (defined $filter_analysis && $filter_analysis->dbID != $analysis_stats->analysis_id);
687

Leo Gordon's avatar
Leo Gordon committed
688
        #digging deeper under the surface so need to sync
689
690
691
692
693
694
695
    if(($analysis_stats->status eq 'LOADING') or ($analysis_stats->status eq 'BLOCKED')) {
      $self->synchronize_AnalysisStats($analysis_stats);
    }

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

696
        # 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
697
    my $workers_this_analysis = $analysis_stats->num_required_workers;
698

699
700
    if((my $hive_capacity = $analysis_stats->hive_capacity) > 0) {   # if there is a limit, use it for cut-off
        my $limit_workers_this_analysis = int($available_load * $hive_capacity);
Leo Gordon's avatar
Leo Gordon committed
701
702
703
704
705

        if($workers_this_analysis > $limit_workers_this_analysis) {
            $workers_this_analysis = $limit_workers_this_analysis;
        }

706
        $available_load -= 1.0*$workers_this_analysis/$hive_capacity;
707
    }
Leo Gordon's avatar
Leo Gordon committed
708
709
710
711
712
713
    $total_workers += $workers_this_analysis;
    $rc2workers{$analysis_stats->rc_id} += $workers_this_analysis;
    $analysis_stats->print_stats();
    printf("  (%1.3f remaining-hive-load) use %3d workers of analysis_id=%d\n", $available_load, $workers_this_analysis, $analysis_stats->analysis_id);

    last if($available_load <= 0.0);
714
  }
715

Leo Gordon's avatar
Leo Gordon committed
716
717
718
719
  printf("need a total of $total_workers workers (availLoad=%1.5f)\n", $available_load);
  return ($total_workers, \%rc2workers);
}

720

Leo Gordon's avatar
Leo Gordon committed
721
sub get_needed_workers_resync_if_necessary {
Leo Gordon's avatar
Leo Gordon committed
722
723
724
725
    my ($self, $meadow, $analysis) = @_;

    my ($needed_count, $rc_hash) = $self->get_num_needed_workers($analysis);

726
    unless( $needed_count or $self->get_hive_current_load() or $self->count_running_workers() ) {
Leo Gordon's avatar
Leo Gordon committed
727
728
729
730
731
732
733
734
        print "*** nothing is running and nothing to do (according to analysis_stats) => perform a hard resync\n" ;

        $self->synchronize_hive($analysis);
        $self->check_for_dead_workers($meadow, 1);

        ($needed_count, $rc_hash) = $self->get_num_needed_workers($analysis);
    }

Leo Gordon's avatar
Leo Gordon committed
735
    return ($needed_count, $rc_hash);
736
737
}

738

Leo Gordon's avatar
Leo Gordon committed
739
sub get_remaining_jobs_show_hive_progress {
740
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
741
  my $sql = "SELECT sum(done_job_count), sum(failed_job_count), sum(total_job_count), ".
742
743
            "sum(unclaimed_job_count * analysis_stats.avg_msec_per_job)/1000/60/60 ".
            "FROM analysis_stats";
744
745
  my $sth = $self->prepare($sql);
  $sth->execute();
746
  my ($done, $failed, $total, $cpuhrs) = $sth->fetchrow_array();
747
  $sth->finish;
748
749
750
751
752
753
754

  $done   ||= 0;
  $failed ||= 0;
  $total  ||= 0;
  my $completed = $total
    ? ((100.0 * ($done+$failed))/$total)
    : 0.0;
755
756
757
  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
758
  return $remaining;
759
760
}

761

Leo Gordon's avatar
Leo Gordon committed
762
sub print_analysis_status {
Leo Gordon's avatar
Leo Gordon committed
763
    my ($self, $filter_analysis) = @_;
764

Leo Gordon's avatar
Leo Gordon committed
765
    my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
766
767
768
    foreach my $analysis (sort {$a->dbID <=> $b->dbID} @$list_of_analyses) {
        $analysis->stats->print_stats($self->{'verbose_stats'});
    }
769
770
771
}


Leo Gordon's avatar
Leo Gordon committed
772
sub print_running_worker_status {
773
  my $self = shift;
774

Leo Gordon's avatar
Leo Gordon committed
775
  print "====== Live workers according to Queen:\n";
776
777
778
  my $sql = "SELECT logic_name, count(*) FROM worker, analysis ".
            "WHERE worker.analysis_id=analysis.analysis_id AND worker.cause_of_death='' ".
            "GROUP BY worker.analysis_id";
Leo Gordon's avatar
Leo Gordon committed
779

780
  my $total_workers = 0;
781
782
  my $sth = $self->prepare($sql);
  $sth->execute();
783
784
785
  while((my $logic_name, my $worker_count)=$sth->fetchrow_array()) {
    printf("%20s : %d workers\n", $logic_name, $worker_count);
    $total_workers += $worker_count;
786
  }
787
  printf("  %d total workers\n", $total_workers);
Leo Gordon's avatar
Leo Gordon committed
788
  print "===========================\n";
789
  $sth->finish;
790
}
791

792
793
794
795
796
797
798
799
800
801
802
=head2 monitor

  Arg[1]     : --none--
  Example    : $queen->monitor();
  Description: Monitors current throughput and store the result in the monitor
               table
  Exceptions : none
  Caller     : beekeepers and other external processes

=cut

Leo Gordon's avatar
Leo Gordon committed
803
sub monitor {
804
805
806
807
  my $self = shift;
  my $sql = qq{
      INSERT INTO monitor
      SELECT
808
          CURRENT_TIMESTAMP,
809
          count(*),
Leo Gordon's avatar
Leo Gordon committed
810
811
812
813
814
815
  }. ( ($self->dbc->driver eq 'sqlite')
        ? qq{ sum(work_done/(strftime('%s','now')-strftime('%s',born))),
              sum(work_done/(strftime('%s','now')-strftime('%s',born)))/count(*), }
        : qq{ sum(work_done/(UNIX_TIMESTAMP()-UNIX_TIMESTAMP(born))),
              sum(work_done/(UNIX_TIMESTAMP()-UNIX_TIMESTAMP(born)))/count(*), }
  ). qq{
816
          group_concat(DISTINCT logic_name)
817
      FROM worker left join analysis USING (analysis_id)
818
819
      WHERE cause_of_death = ''
  };
820
821
822
823
824
      
  my $sth = $self->prepare($sql);
  $sth->execute();
}

Leo Gordon's avatar
Leo Gordon committed
825
826
827
828
829
830
831
832
833
834
835
836
837
838
=head2 register_all_workers_dead

  Example    : $queen->register_all_workers_dead();
  Description: Registers all workers dead
  Exceptions : none
  Caller     : beekeepers and other external processes

=cut

sub register_all_workers_dead {
    my $self = shift;

    my $overdueWorkers = $self->fetch_overdue_workers(0);
    foreach my $worker (@{$overdueWorkers}) {
839
        $worker->cause_of_death( 'FATALITY' );  # well, maybe we could have investigated further...
Leo Gordon's avatar
Leo Gordon committed
840
841
842
843
        $self->register_worker_death($worker);
    }
}

844

Jessica Severin's avatar
Jessica Severin committed
845
846
847
848
849
#
# INTERNAL METHODS
#
###################

850
sub _pick_best_analysis_for_new_worker {
Leo Gordon's avatar
Leo Gordon committed
851
852
  my $self  = shift;
  my $rc_id = shift;    # this parameter will need to percolate very deep
853
854
855
856

  my $statsDBA = $self->db->get_AnalysisStatsAdaptor;
  return undef unless($statsDBA);

Leo Gordon's avatar
Leo Gordon committed
857
  my ($stats) = @{$statsDBA->fetch_by_needed_workers(1,$self->{maximise_concurrency}, $rc_id)};
858
859
860
  if($stats) {
    #synchronize and double check that it can be run
    $self->safe_synchronize_AnalysisStats($stats);
861
    return $stats if(($stats->status ne 'BLOCKED') and ($stats->num_required_workers > 0) and (!defined($rc_id) or ($stats->rc_id == $rc_id)));
862
  }
863

864
  # ok so no analyses 'need' workers with the given $rc_id.
865
866
867
  if ($self->get_num_failed_analyses()) {
    return undef;
  }
868
869
870
871
  # see if anything needs an update, in case there are
  # hidden jobs that haven't made it into the summary stats

  print("QUEEN: no obvious needed workers, need to dig deeper\n");
Leo Gordon's avatar
Leo Gordon committed
872
  my $stats_list = $statsDBA->fetch_by_statuses(['LOADING', 'BLOCKED']);
873
  foreach $stats (@$stats_list) {
874
    $self->safe_synchronize_AnalysisStats($stats);
875

Leo Gordon's avatar
Leo Gordon committed
876
    return $stats if(($stats->status ne 'BLOCKED') and ($stats->num_required_workers > 0) and (!defined($rc_id) or ($stats->rc_id == $rc_id)));
877
878
  }

879
    # does the following really ever help?
Leo Gordon's avatar
Leo Gordon committed
880
  ($stats) = @{$statsDBA->fetch_by_needed_workers(1,$self->{maximise_concurrency}, $rc_id)};
881
  return $stats;
882
883
884
}


885
=head2 fetch_by_dbID
Jessica Severin's avatar
Jessica Severin committed
886
887
888

  Arg [1]    : int $id
               the unique database identifier for the feature to be obtained
889
  Example    : $feat = $queen->fetch_by_dbID(1234);
Jessica Severin's avatar
Jessica Severin committed
890
891
892
  Description: Returns the feature created from the database defined by the
               the id $id.
  Returntype : Bio::EnsEMBL::Hive::Worker
893
  Exceptions : thrown if $worker_id is not defined
Jessica Severin's avatar
Jessica Severin committed
894
895
896
897
  Caller     : general

=cut

898
899
sub fetch_by_dbID {
  my ($self, $worker_id) = @_;
Jessica Severin's avatar
Jessica Severin committed
900

901
  unless(defined $worker_id) {
902
    throw("fetch_by_dbID must have an id");
Jessica Severin's avatar
Jessica Severin committed
903
904
  }

905
  my ($obj) = @{$self->_generic_fetch( "w.worker_id = $worker_id" ) };
Jessica Severin's avatar
Jessica Severin committed
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949