Queen.pm 33 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 $worker->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
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
  my (  $meadow_type, $process_id, $exec_host,
        $rc_id, $logic_name, $analysis_id, $input_id, $job_id,
        $no_write, $debug, $worker_output_dir, $hive_output_dir, $batch_size, $job_limit, $life_span, $no_cleanup, $retry_throwing_jobs) =

 rearrange([qw(meadow_type process_id exec_host
        rc_id logic_name analysis_id input_id job_id
        no_write debug worker_output_dir hive_output_dir batch_size job_limit life_span no_cleanup retry_throwing_jobs) ], @args);

    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($job) {
    #go into autonomous mode
167
    return undef if($self->get_hive_current_load() >= 1.1);
168
    
169
170
    $analysis_stats_adaptor->decrease_needed_workers($analysisStats->analysis_id);
    $analysis_stats_adaptor->increase_running_workers($analysisStats->analysis_id);
171
172
173
174
175
176
177
178
179
180
    $analysisStats->print_stats;
    
    if($analysisStats->status eq 'BLOCKED') {
      print("Analysis is BLOCKED, can't create workers\n");
      return undef;
    }
    if($analysisStats->status eq 'DONE') {
      print("Analysis is DONE, don't need to create workers\n");
      return undef;
    }
181
  }
182
  
183
184
  my $sql = q{INSERT INTO worker 
              (born, last_check_in, meadow_type, process_id, host, analysis_id)
185
              VALUES (NOW(), NOW(), ?,?,?,?)};
Jessica Severin's avatar
Jessica Severin committed
186
187

  my $sth = $self->prepare($sql);
188
  $sth->execute($meadow_type, $process_id, $exec_host, $analysisStats->analysis_id);
Leo Gordon's avatar
Leo Gordon committed
189
  my $worker_id = $sth->{'mysql_insertid'};
Jessica Severin's avatar
Jessica Severin committed
190
191
  $sth->finish;

192
  my $worker = $self->fetch_by_dbID($worker_id);
Jessica Severin's avatar
Jessica Severin committed
193
194
  $worker=undef unless($worker and $worker->analysis);

195
  if($worker and $analysisStats) {
196
    $analysisStats->update_status('WORKING');
Jessica Severin's avatar
Jessica Severin committed
197
  }
198
  
199
  $worker->_specific_job($job) if($job);
200
  $worker->execute_writes(0) if($no_write);
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229

    $worker->debug($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($batch_size) {
      $worker->set_worker_batch_size($batch_size);
    }
    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
230
231
232
233
234
235
236
  return $worker;
}

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

  return unless($worker);
Leo Gordon's avatar
Leo Gordon committed
237

Leo Gordon's avatar
bugfix  
Leo Gordon committed
238
239
240
  my $cod = $worker->cause_of_death();

  unless ($cod eq 'HIVE_OVERLOAD') {
241
242
243
244
    ## HIVE_OVERLOAD occurs after a successful update of the analysis_stats teble. (c.f. Worker.pm)
    $worker->analysis->stats->adaptor->decrease_running_workers($worker->analysis->stats->analysis_id);
  }

245
  my $sql = "UPDATE worker SET died=now(), last_check_in=now()";
246
  $sql .= " ,status='DEAD'";
Jessica Severin's avatar
Jessica Severin committed
247
  $sql .= " ,work_done='" . $worker->work_done . "'";
Leo Gordon's avatar
bugfix  
Leo Gordon committed
248
  $sql .= " ,cause_of_death='$cod'";
249
  $sql .= " WHERE worker_id='" . $worker->dbID ."'";
Jessica Severin's avatar
Jessica Severin committed
250

Leo Gordon's avatar
bugfix  
Leo Gordon committed
251
  $self->dbc->do( $sql );
252

Leo Gordon's avatar
bugfix  
Leo Gordon committed
253
  if($cod eq 'NO_WORK') {
254
    $self->db->get_AnalysisStatsAdaptor->update_status($worker->analysis->dbID, 'ALL_CLAIMED');
255
  }
Leo Gordon's avatar
bugfix  
Leo Gordon committed
256
257
258
259
  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
260
    $self->db->get_AnalysisJobAdaptor->release_undone_jobs_from_worker($worker);
261
  }
262
  
263
  # re-sync the analysis_stats when a worker dies as part of dynamic sync system
264
  if($self->safe_synchronize_AnalysisStats($worker->analysis->stats)->status ne 'DONE') {
265
266
    # 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
267
    $self->db->get_AnalysisStatsAdaptor->increase_needed_workers($worker->analysis->dbID);
268
269
  }

Jessica Severin's avatar
Jessica Severin committed
270
271
}

272
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
273
274
    my ($self, $meadow, $check_buried_in_haste) = @_;

275
    my $worker_status_hash    = $meadow->status_of_all_our_workers();
Leo Gordon's avatar
Leo Gordon committed
276
    my %worker_status_summary = ();
277
    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
278
279
280

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

281
282
    my %gc_wpid_to_worker = ();

Leo Gordon's avatar
Leo Gordon committed
283
284
285
    foreach my $worker (@$queen_worker_list) {
        next unless($meadow->responsible_for_worker($worker));

286
287
        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
288
289
290
            $worker_status_summary{$status}++;
        } else {
            $worker_status_summary{'AWOL'}++;
291

292
            $gc_wpid_to_worker{$process_id} = $worker;
Leo Gordon's avatar
Leo Gordon committed
293
294
295
296
        }
    }
    print "\t".join(', ', map { "$_:$worker_status_summary{$_}" } keys %worker_status_summary)."\n\n";

297
298
299
300
301
302
303
304
305
306
307
    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";
308
309
        while(my ($process_id, $worker) = each %gc_wpid_to_worker) {
            $worker->cause_of_death( $wpid_to_cod->{$process_id} || 'FATALITY');
310
311
312
313
            $self->register_worker_death($worker);
        }
    }

Leo Gordon's avatar
Leo Gordon committed
314
315
    if($check_buried_in_haste) {
        print "====== Checking for workers buried in haste... ";
316
        my $buried_in_haste_list = $self->fetch_all_dead_workers_with_jobs();
Leo Gordon's avatar
Leo Gordon committed
317
318
319
320
321
        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
322
                    $job_adaptor->release_undone_jobs_from_worker($worker);
Leo Gordon's avatar
Leo Gordon committed
323
324
325
326
327
328
329
                }
            }
        } else {
            print "none\n";
        }
    }
}
Jessica Severin's avatar
Jessica Severin committed
330

331
sub worker_check_in {
Jessica Severin's avatar
Jessica Severin committed
332
333
334
  my ($self, $worker) = @_;

  return unless($worker);
335
  my $sql = "UPDATE worker SET last_check_in=now()";
336
  $sql .= " ,work_done='" . $worker->work_done . "'";
337
  $sql .= " WHERE worker_id='" . $worker->dbID ."'";
Jessica Severin's avatar
Jessica Severin committed
338
339
340
341

  my $sth = $self->prepare($sql);
  $sth->execute();
  $sth->finish;
342
  
343
  $self->safe_synchronize_AnalysisStats($worker->analysis->stats);
Jessica Severin's avatar
Jessica Severin committed
344
345
346
}


347
=head2 reset_and_fetch_job_by_dbID
348

349
  Arg [1]: int $job_id
350
  Example: 
351
    my $job = $queen->reset_and_fetch_job_by_dbID($job_id);
352
  Description: 
353
    For the specified job_id it will fetch just that job, 
354
355
356
    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 
357
358
    same job simultaneously (use only for debugging).
  Returntype : 
359
    Bio::EnsEMBL::Hive::AnalysisJob object
360
  Exceptions :
361
  Caller     : beekeepers, runWorker.pl scripts
362
363
364

=cut

365
366
sub reset_and_fetch_job_by_dbID {
  my $self = shift;
367
  my $job_id = shift;
368
  
369
  my $jobDBA = $self->db->get_AnalysisJobAdaptor;
370
  $jobDBA->reset_job_by_dbID($job_id); 
371

372
  my $job = $jobDBA->fetch_by_dbID($job_id); 
373
374
  my $stats = $self->db->get_AnalysisStatsAdaptor->fetch_by_analysis_id($job->analysis_id);
  $self->synchronize_AnalysisStats($stats);
375
  
376
377
378
379
  return $job;
}


380
381
382
383
384
385
######################################
#
# Public API interface for beekeeper
#
######################################

386
387
388
    # Note: asking for Queen->fetch_overdue_workers(0) essentially means
    #       "fetch all workers known to the Queen not to be officially dead"
    #
389
390
391
392
393
sub fetch_overdue_workers {
  my ($self,$overdue_secs) = @_;

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

394
395
  my $constraint = "w.cause_of_death='' ".
                   "AND (UNIX_TIMESTAMP()-UNIX_TIMESTAMP(w.last_check_in))>$overdue_secs";
396
397
398
  return $self->_generic_fetch($constraint);
}

399
400
sub fetch_failed_workers {
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
401

402
  my $constraint = "w.cause_of_death='FATALITY' ";
403
404
405
  return $self->_generic_fetch($constraint);
}

406
sub fetch_all_dead_workers_with_jobs {
Leo Gordon's avatar
Leo Gordon committed
407
408
  my $self = shift;

409
  # 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
410

411
412
  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
413
414
  return $self->_generic_fetch($constraint, $join);
}
Jessica Severin's avatar
Jessica Severin committed
415

416
417
=head2 synchronize_hive

Leo Gordon's avatar
Leo Gordon committed
418
  Arg [1]    : $filter_analysis (optional)
419
420
  Example    : $queen->synchronize_hive();
  Description: Runs through all analyses in the system and synchronizes
421
422
              the analysis_stats summary with the states in the job 
              and worker tables.  Then follows by checking all the blocking rules
423
424
425
426
427
428
429
              and blocks/unblocks analyses as needed.
  Exceptions : none
  Caller     : general

=cut

sub synchronize_hive {
Leo Gordon's avatar
Leo Gordon committed
430
  my $self          = shift;
Leo Gordon's avatar
Leo Gordon committed
431
  my $filter_analysis = shift; # optional parameter
432

433
  my $start_time = time();
434

Leo Gordon's avatar
Leo Gordon committed
435
  my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
436

Leo Gordon's avatar
Leo Gordon committed
437
  print STDERR "\nSynchronizing the hive (".scalar(@$list_of_analyses)." analyses this time):\n";
Leo Gordon's avatar
Leo Gordon committed
438
439
  foreach my $analysis (@$list_of_analyses) {
    $self->synchronize_AnalysisStats($analysis->stats);
Leo Gordon's avatar
Leo Gordon committed
440
    print STDERR '.';
441
  }
Leo Gordon's avatar
Leo Gordon committed
442
  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
443

Leo Gordon's avatar
Leo Gordon committed
444
  print STDERR "Checking blocking control rules:\n";
Leo Gordon's avatar
Leo Gordon committed
445
  foreach my $analysis (@$list_of_analyses) {
Leo Gordon's avatar
Leo Gordon committed
446
447
    my $open = $analysis->stats->check_blocking_control_rules();
    print STDERR ($open ? 'o' : 'x');
448
  }
Leo Gordon's avatar
Leo Gordon committed
449
  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
450

Leo Gordon's avatar
Leo Gordon committed
451
  print STDERR ''.((time() - $start_time))." seconds to synchronize_hive\n\n";
452
}
453

454

455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
=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
477
                   ($stats->seconds_since_last_update < 3*60));
478
479
480
481

  # 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=" . $stats->analysis_id;
482
  #print("$sql\n");
483
484
  my $row_count = $self->dbc->do($sql);  
  return $stats unless($row_count == 1);
485
  #printf("got sync_lock on analysis_stats(%d)\n", $stats->analysis_id);
486
487
488
  
  #OK have the lock, go and do the sync
  $self->synchronize_AnalysisStats($stats);
489

490
491
492
493
  return $stats;
}


494
=head2 synchronize_AnalysisStats
495

496
497
  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
  Example    : $self->synchronize($analysisStats);
498
  Description: Queries the job and worker tables to get summary counts
499
500
501
502
503
               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
504

505
=cut
506

507
sub synchronize_AnalysisStats {
508
  my $self = shift;
509
  my $analysisStats = shift;
510

511
  return $analysisStats unless($analysisStats);
512
  return $analysisStats unless($analysisStats->analysis_id);
513
  
514
  $analysisStats->refresh(); ## Need to get the new hive_capacity for dynamic analyses
515
516
517
518
519
520
  $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);

521
  my $sql = "SELECT status, count(*), semaphore_count FROM job ".
522
            "WHERE analysis_id=? GROUP BY status, semaphore_count";
523
524
525
  my $sth = $self->prepare($sql);
  $sth->execute($analysisStats->analysis_id);

526
527
  my $hive_capacity = $analysisStats->hive_capacity;

528
529
  my $done_here      = 0;
  my $done_elsewhere = 0;
530
  while (my ($status, $count, $semaphore_count)=$sth->fetchrow_array()) {
531
# print STDERR "$status - $count\n";
532
533
534
535

    my $total = $analysisStats->total_job_count();
    $analysisStats->total_job_count($total + $count);

536
    if(($status eq 'READY') and ($semaphore_count<=0)) {
537
      $analysisStats->unclaimed_job_count($count);
538
539
540
541
542
543
544
545
546
      my $numWorkers;
      if($analysisStats->batch_size > 0) {
        $numWorkers = POSIX::ceil($count / $analysisStats->batch_size);
      } else {
        my $job_msec = $analysisStats->avg_msec_per_job;
        $job_msec = 100 if($job_msec>0 and $job_msec<100);
        $numWorkers = POSIX::ceil(($count * $job_msec) / (3*60*1000)); 
        # guess num needed workers by total jobs / (num jobs a worker could do in 3 minutes)
      }
547
      $numWorkers=$count if($numWorkers==0);
548
549
550
551
      if($analysisStats->hive_capacity>0 and $numWorkers > $analysisStats->hive_capacity) {
        $numWorkers=$analysisStats->hive_capacity;
      }
      $analysisStats->num_required_workers($numWorkers);
552
553
554
555
556
557
    } elsif($status eq 'DONE' and $semaphore_count<=0) {
        $done_here = $count;
    } elsif($status eq 'PASSED_ON' and $semaphore_count<=0) {
        $done_elsewhere = $count;
    } elsif ($status eq 'FAILED') {
        $analysisStats->failed_job_count($count);
558
559
560
    }
  }
  $sth->finish;
561

562
563
  $analysisStats->done_job_count($done_here + $done_elsewhere);

564
  $analysisStats->check_blocking_control_rules();
565

566
567
568
  if($analysisStats->status ne 'BLOCKED') {
    $analysisStats->determine_status();
  }
569

570
571
572
573
  #
  # adjust_stats_for_living_workers
  #
  
574
  if($analysisStats->hive_capacity > 0) {
575
    my $liveCount = $analysisStats->get_running_worker_count();
576
577
578
579
580
581
582
583

    my $numWorkers = $analysisStats->num_required_workers;

    my $capacityAdjust = ($numWorkers + $liveCount) - $analysisStats->hive_capacity;
    $numWorkers -= $capacityAdjust if($capacityAdjust > 0);
    $numWorkers=0 if($numWorkers<0);

    $analysisStats->num_required_workers($numWorkers);
584
  }
585

586
  $analysisStats->update;  #update and release sync_lock
587

588
  return $analysisStats;
589
}
590
591


Leo Gordon's avatar
Leo Gordon committed
592
593
sub get_num_failed_analyses {
  my ($self, $analysis) = @_;
594
595

  my $statsDBA = $self->db->get_AnalysisStatsAdaptor;
Leo Gordon's avatar
Leo Gordon committed
596
  my $failed_analyses = $statsDBA->fetch_by_statuses(['FAILED']);
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
  if ($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);
}


619
620
sub get_hive_current_load {
  my $self = shift;
621
622
623
  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";
624
625
626
627
628
  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
629
  print("current hive load = $load\n");
630
631
632
633
  return $load;
}


634
635
sub get_num_running_workers {
  my $self = shift;
636
  my $sql = "SELECT count(*) FROM worker WHERE cause_of_death =''";
637
638
639
640
641
  my $sth = $self->prepare($sql);
  $sth->execute();
  (my $runningCount)=$sth->fetchrow_array();
  $sth->finish;
  $runningCount=0 unless($runningCount);
Leo Gordon's avatar
Leo Gordon committed
642
  print("current hive num_running_workers = $runningCount\n");
643
644
645
  return $runningCount;
}

646
647
648
sub enter_status {
  my ($self, $worker, $status) = @_;

649
  $self->dbc->do("UPDATE worker SET status = '$status' WHERE worker_id = ".$worker->dbID);
650
651
}

652
653
=head2 get_num_needed_workers

654
  Arg[1]     : Bio::EnsEMBL::Analysis object (optional)
655
656
657
658
  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
659
               If Arg[1] is defined, does it only for the given analysis.
660
  Exceptions : none
661
  Caller     : beekeepers and other external processes
662
663
664

=cut

665
sub get_num_needed_workers {
Leo Gordon's avatar
Leo Gordon committed
666
  my ($self, $filter_analysis) = @_;
667

668
  my $statsDBA = $self->db->get_AnalysisStatsAdaptor;
Leo Gordon's avatar
Leo Gordon committed
669
670
671
  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);
672

Leo Gordon's avatar
Leo Gordon committed
673
  return 0 unless(@all_analyses);
674

Leo Gordon's avatar
Leo Gordon committed
675
  my $available_load = 1.0 - $self->get_hive_current_load();
676

Leo Gordon's avatar
Leo Gordon committed
677
  return 0 if($available_load <=0.0);
678

Leo Gordon's avatar
Leo Gordon committed
679
680
  my $total_workers = 0;
  my %rc2workers = ();
681

Leo Gordon's avatar
Leo Gordon committed
682
683
  foreach my $analysis_stats (@all_analyses) {
    next if (defined $filter_analysis && $filter_analysis->dbID != $analysis_stats->analysis_id);
684

Leo Gordon's avatar
Leo Gordon committed
685
        #digging deeper under the surface so need to sync
686
687
688
689
690
691
692
    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);

Leo Gordon's avatar
Leo Gordon committed
693
        # FIXME: the following call sometimes returns a stale number greater than the number of workers actually needed for an analysis; resync fixes it
Leo Gordon's avatar
Leo Gordon committed
694
    my $workers_this_analysis = $analysis_stats->num_required_workers;
695

Leo Gordon's avatar
Leo Gordon committed
696
697
698
699
700
701
702
703
    if($analysis_stats->hive_capacity > 0) {   # if there is a limit, use it for cut-off
        my $limit_workers_this_analysis = int($available_load * $analysis_stats->hive_capacity);

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

        $available_load -= 1.0*$workers_this_analysis/$analysis_stats->hive_capacity;
704
    }
Leo Gordon's avatar
Leo Gordon committed
705
706
707
708
709
710
    $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);
711
  }
712

Leo Gordon's avatar
Leo Gordon committed
713
714
715
716
  printf("need a total of $total_workers workers (availLoad=%1.5f)\n", $available_load);
  return ($total_workers, \%rc2workers);
}

Leo Gordon's avatar
Leo Gordon committed
717
sub get_needed_workers_resync_if_necessary {
Leo Gordon's avatar
Leo Gordon committed
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
    my ($self, $meadow, $analysis) = @_;

    my $load                     = $self->get_hive_current_load();
    my $running_count            = $self->get_num_running_workers();
    my ($needed_count, $rc_hash) = $self->get_num_needed_workers($analysis);

    if($load==0 and $needed_count==0 and $running_count==0) {
        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
733
    return ($needed_count, $rc_hash);
734
735
}

Leo Gordon's avatar
Leo Gordon committed
736
sub get_remaining_jobs_show_hive_progress {
737
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
738
  my $sql = "SELECT sum(done_job_count), sum(failed_job_count), sum(total_job_count), ".
739
740
            "sum(unclaimed_job_count * analysis_stats.avg_msec_per_job)/1000/60/60 ".
            "FROM analysis_stats";
741
742
  my $sth = $self->prepare($sql);
  $sth->execute();
743
  my ($done, $failed, $total, $cpuhrs) = $sth->fetchrow_array();
744
  $sth->finish;
745
746
747
748
749
750
751

  $done   ||= 0;
  $failed ||= 0;
  $total  ||= 0;
  my $completed = $total
    ? ((100.0 * ($done+$failed))/$total)
    : 0.0;
752
753
754
  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
755
  return $remaining;
756
757
}

758

Leo Gordon's avatar
Leo Gordon committed
759
sub print_analysis_status {
Leo Gordon's avatar
Leo Gordon committed
760
    my ($self, $filter_analysis) = @_;
761

Leo Gordon's avatar
Leo Gordon committed
762
    my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
763
764
765
    foreach my $analysis (sort {$a->dbID <=> $b->dbID} @$list_of_analyses) {
        $analysis->stats->print_stats($self->{'verbose_stats'});
    }
766
767
768
}


Leo Gordon's avatar
Leo Gordon committed
769
sub print_running_worker_status {
770
  my $self = shift;
771

Leo Gordon's avatar
Leo Gordon committed
772
  print "====== Live workers according to Queen:\n";
773
774
775
  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
776
777

  my $total = 0;
778
779
780
781
  my $sth = $self->prepare($sql);
  $sth->execute();
  while((my $logic_name, my $count)=$sth->fetchrow_array()) {
    printf("%20s : %d workers\n", $logic_name, $count);
782
    $total += $count;
783
  }
784
  printf("  %d total workers\n", $total);
Leo Gordon's avatar
Leo Gordon committed
785
  print "===========================\n";
786
  $sth->finish;
787
}
788

789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
=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

sub monitor
{
  my $self = shift;
  my $sql = qq{
      INSERT INTO monitor
      SELECT
          now(),
          count(*),
          sum(work_done/TIME_TO_SEC(TIMEDIFF(now(),born))),
          sum(work_done/TIME_TO_SEC(TIMEDIFF(now(),born)))/count(*),
          group_concat(DISTINCT logic_name)
811
      FROM worker left join analysis USING (analysis_id)
812
813
814
815
816
817
      WHERE cause_of_death = ""};
      
  my $sth = $self->prepare($sql);
  $sth->execute();
}

Leo Gordon's avatar
Leo Gordon committed
818
819
820
821
822
823
824
825
826
827
828
829
830
831
=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}) {
832
        $worker->cause_of_death( 'FATALITY' );  # well, maybe we could have investigated further...
Leo Gordon's avatar
Leo Gordon committed
833
834
835
836
        $self->register_worker_death($worker);
    }
}

837

Jessica Severin's avatar
Jessica Severin committed
838
839
840
841
842
#
# INTERNAL METHODS
#
###################

843
sub _pick_best_analysis_for_new_worker {
Leo Gordon's avatar
Leo Gordon committed
844
845
  my $self  = shift;
  my $rc_id = shift;    # this parameter will need to percolate very deep
846
847
848
849

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

Leo Gordon's avatar
Leo Gordon committed
850
  my ($stats) = @{$statsDBA->fetch_by_needed_workers(1,$self->{maximise_concurrency}, $rc_id)};
851
852
853
854
855
  if($stats) {
    #synchronize and double check that it can be run
    $self->safe_synchronize_AnalysisStats($stats);
    return $stats if(($stats->status ne 'BLOCKED') and ($stats->num_required_workers > 0));
  }
856
857

  # ok so no analyses 'need' workers.
858
859
860
  if ($self->get_num_failed_analyses()) {
    return undef;
  }
861
862
863
864
  # 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
865
  my $stats_list = $statsDBA->fetch_by_statuses(['LOADING', 'BLOCKED']);
866
  foreach $stats (@$stats_list) {
867
    $self->safe_synchronize_AnalysisStats($stats);
868

Leo Gordon's avatar
Leo Gordon committed
869
    return $stats if(($stats->status ne 'BLOCKED') and ($stats->num_required_workers > 0) and (!defined($rc_id) or ($stats->rc_id == $rc_id)));
870
871
  }

Leo Gordon's avatar
Leo Gordon committed
872
  ($stats) = @{$statsDBA->fetch_by_needed_workers(1,$self->{maximise_concurrency}, $rc_id)};
873
874
875
876
877
878
  return $stats if($stats);

  return undef;
}


879
=head2 fetch_by_dbID
Jessica Severin's avatar
Jessica Severin committed
880
881
882

  Arg [1]    : int $id
               the unique database identifier for the feature to be obtained
883
  Example    : $feat = $queen->fetch_by_dbID(1234);
Jessica Severin's avatar
Jessica Severin committed
884
885
886
  Description: Returns the feature created from the database defined by the
               the id $id.
  Returntype : Bio::EnsEMBL::Hive::Worker
887
  Exceptions : thrown if $worker_id is not defined
Jessica Severin's avatar
Jessica Severin committed
888
889
890
891
  Caller     : general

=cut

892
893
sub fetch_by_dbID {
  my ($self, $worker_id) = @_;
Jessica Severin's avatar
Jessica Severin committed
894

895
  unless(defined $worker_id) {
896
    throw("fetch_by_dbID must have an id");
Jessica Severin's avatar
Jessica Severin committed
897
898
  }

899
  my ($obj) = @{$self->_generic_fetch( "w.worker_id = $worker_id" ) };
Jessica Severin's avatar
Jessica Severin committed
900
901
902
903
904
905
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
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
  return $obj;
}

=head2 _generic_fetch

  Arg [1]    : (optional) string $constraint
               An SQL query constraint (i.e. part of the WHERE clause)
  Arg [2]    : (optional) string $logic_name
               the logic_name of the analysis of the features to obtain
  Example    : $fts = $a->_generic_fetch('contig_id in (1234, 1235)', 'Swall');
  Description: Performs a database fetch and returns feature objects in
               contig coordinates.
  Returntype : listref of Bio::EnsEMBL::SeqFeature in contig coordinates
  Exceptions : none
  Caller     : BaseFeatureAdaptor, ProxyDnaAlignFeatureAdaptor::_generic_fetch

=cut

sub _generic_fetch {
  my ($self, $constraint, $join) = @_;

  my @tables = $self->_tables;
  my $columns = join(', ', $self->_columns());

  if ($join) {
    foreach my $single_join (@{$join}) {
      my ($tablename, $condition, $extra_columns) = @{$single_join};
      if ($tablename && $condition) {
        push @tables, $tablename;

        if($constraint) {
          $constraint .= " AND $condition";
        } else {
          $constraint = " $condition";
        }
      }
      if ($extra_columns) {
        $columns .= ", " . join(', ', @{$extra_columns});
      }
    }
  }

  #construct a nice table string like 'table1 t1, table2 t2'
  my $tablenames = join(', ', map({ join(' ', @$_) } @tables));

  my $sql = "SELECT $columns FROM $tablenames";

  my $default_where = $self->_default_where_clause;
  my $final_clause = $self->_final_clause;

  #append a where clause if it was defined
  if($constraint) {
    $sql .= " WHERE $constraint ";
    if($default_where) {
      $sql .= " AND $default_where ";
    }
  } elsif($default_where) {
    $sql .= " WHERE $default_where ";
  }

  #append additional clauses which may have been defined
  $sql .= " $final_clause";

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

#  print STDERR $sql,"\n";

  return $self->_objs_from_sth($sth);
}

sub _tables {
  my $self = shift;

974
  return (['worker', 'w']);
Jessica Severin's avatar
Jessica Severin committed
975
976
977
978
979
}

sub _columns {
  my $self = shift;

980
981
982
983
984
985
986
987
988
989
990
  return qw (w.worker_id
             w.analysis_id
             w.meadow_type
             w.host
             w.process_id
             w.work_done
             w.status
             w.born
             w.last_check_in
             w.died
             w.cause_of_death
Jessica Severin's avatar
Jessica Severin committed
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
            );
}

sub _objs_from_sth {
  my ($self, $sth) = @_;

  my %column;
  $sth->bind_columns( \( @column{ @{$sth->{NAME_lc} } } ));

  my @workers = ();

  while ($sth->fetch()) {
    my $worker = new Bio::EnsEMBL::Hive::Worker;
    $worker->init;

1006
1007
    $worker->dbID($column{'worker_id'});
    $worker->meadow_type($column{'meadow_type'});
Jessica Severin's avatar
Jessica Severin committed
1008
1009
1010
    $worker->host($column{'host'});
    $worker->process_id($column{'process_id'});
    $worker->work_done($column{'work_done'});
1011
    $worker->status($column{'status'});
Jessica Severin's avatar
Jessica Severin committed
1012
1013
1014
1015
    $worker->born($column{'born'});
    $worker->last_check_in($column{'last_check_in'});
    $worker->died($column{'died'});
    $worker->cause_of_death($column{'cause_of_death'});
1016
    $worker->queen($self);
Jessica Severin's avatar
Jessica Severin committed
1017
1018
    $worker->db($self->db);

1019
1020
    if($column{'analysis_id'} and $self->db->get_AnalysisAdaptor) {
      $worker->analysis($self->db->get_AnalysisAdaptor->fetch_by_dbID($column{'analysis_id'}));
Jessica Severin's avatar
Jessica Severin committed
1021