Queen.pm 37.9 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
  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, $compile_module_once) =
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 compile_module_once) ], @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
216
  $worker->init;

217
  if($worker and $analysisStats) {
218
    $analysisStats->update_status('WORKING');
Jessica Severin's avatar
Jessica Severin committed
219
  }
220
  
221
  $worker->_specific_job($job) if($job);
222
  $worker->execute_writes(0) if($no_write);
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247

    $worker->debug($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);
    }
248
249
250
    if(defined $compile_module_once) {
        $worker->compile_module_once($compile_module_once);
    }
251

252
    return $worker;
Jessica Severin's avatar
Jessica Severin committed
253
254
}

255

Jessica Severin's avatar
Jessica Severin committed
256
257
258
259
sub register_worker_death {
  my ($self, $worker) = @_;

  return unless($worker);
Leo Gordon's avatar
Leo Gordon committed
260

Leo Gordon's avatar
bugfix  
Leo Gordon committed
261
262
  my $cod = $worker->cause_of_death();

263
  my $sql = "UPDATE worker SET died=CURRENT_TIMESTAMP, last_check_in=CURRENT_TIMESTAMP";
264
  $sql .= " ,status='DEAD'";
Jessica Severin's avatar
Jessica Severin committed
265
  $sql .= " ,work_done='" . $worker->work_done . "'";
Leo Gordon's avatar
bugfix  
Leo Gordon committed
266
  $sql .= " ,cause_of_death='$cod'";
267
  $sql .= " WHERE worker_id='" . $worker->dbID ."'";
Jessica Severin's avatar
Jessica Severin committed
268

Leo Gordon's avatar
bugfix  
Leo Gordon committed
269
  $self->dbc->do( $sql );
270

271
272
273
274
  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
275
  if($cod eq 'NO_WORK') {
276
    $self->db->get_AnalysisStatsAdaptor->update_status($worker->analysis->dbID, 'ALL_CLAIMED');
277
  }
Leo Gordon's avatar
bugfix  
Leo Gordon committed
278
279
280
281
  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
282
    $self->db->get_AnalysisJobAdaptor->release_undone_jobs_from_worker($worker);
283
  }
284
  
285
  # re-sync the analysis_stats when a worker dies as part of dynamic sync system
286
  if($self->safe_synchronize_AnalysisStats($worker->analysis->stats)->status ne 'DONE') {
287
288
    # 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
289
    $self->db->get_AnalysisStatsAdaptor->increase_required_workers($worker->analysis->dbID);
290
291
  }

Jessica Severin's avatar
Jessica Severin committed
292
293
}

294
sub check_for_dead_workers {    # scans the whole Valley for lost Workers (but ignores unreachagle ones)
295
    my ($self, $valley, $check_buried_in_haste) = @_;
Leo Gordon's avatar
Leo Gordon committed
296

297
    warn "GarbageCollector:\tChecking for lost Workers...\n";
298

299
300
301
302
    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
303

304
    warn "GarbageCollector:\t[Queen:] we have ".scalar(@$queen_worker_list)." Workers alive.\n";
305

Leo Gordon's avatar
Leo Gordon committed
306
307
    foreach my $worker (@$queen_worker_list) {

308
309
310
        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
311
        } else {
312
            $worker_status_counts{$meadow_type}{'UNREACHABLE'}++;
313

314
            next;   # Worker is unreachable from this Valley
Leo Gordon's avatar
Leo Gordon committed
315
316
        }

317
318
319
320
321
        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'}++;
322

323
            $mt_and_pid_to_lost_worker{$meadow_type}{$process_id} = $worker;
324
        }
325
326
327
328
329
330
    }

        # 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";
    }
331

332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
    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);
            }
352
353
354
        }
    }

355
        # the following bit is completely Meadow-agnostic and only restores database integrity:
Leo Gordon's avatar
Leo Gordon committed
356
    if($check_buried_in_haste) {
357
        warn "GarbageCollector:\tChecking for Workers buried in haste...\n";
358
        my $buried_in_haste_list = $self->fetch_all_dead_workers_with_jobs();
Leo Gordon's avatar
Leo Gordon committed
359
        if(my $bih_number = scalar(@$buried_in_haste_list)) {
360
            warn "GarbageCollector:\tfound $bih_number jobs, reclaiming.\n\n";
Leo Gordon's avatar
Leo Gordon committed
361
362
363
            if($bih_number) {
                my $job_adaptor = $self->db->get_AnalysisJobAdaptor();
                foreach my $worker (@$buried_in_haste_list) {
Leo Gordon's avatar
Leo Gordon committed
364
                    $job_adaptor->release_undone_jobs_from_worker($worker);
Leo Gordon's avatar
Leo Gordon committed
365
366
367
                }
            }
        } else {
368
            warn "GarbageCollector:\tfound none\n";
Leo Gordon's avatar
Leo Gordon committed
369
370
371
        }
    }
}
Jessica Severin's avatar
Jessica Severin committed
372
373


374
375
376
    # 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
377

378
    $self->dbc->do("UPDATE worker SET last_check_in=CURRENT_TIMESTAMP, status='".$worker->status."', work_done='".$worker->work_done."' WHERE worker_id='".$worker->dbID."'");
Jessica Severin's avatar
Jessica Severin committed
379
380
381
}


382
=head2 reset_job_by_dbID_and_sync
383

384
  Arg [1]: int $job_id
385
  Example: 
386
    my $job = $queen->reset_job_by_dbID_and_sync($job_id);
387
  Description: 
388
    For the specified job_id it will fetch just that job, 
389
390
391
    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 
392
    same job simultaneously (use only for debugging).
393
  Returntype : none
394
  Exceptions :
395
  Caller     : beekeeper.pl
396
397
398

=cut

399
400
sub reset_job_by_dbID_and_sync {
    my ($self, $job_id) = @_;
401

402
403
404
405
406
407
    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);
408
409
410
}


411
412
413
414
415
416
######################################
#
# Public API interface for beekeeper
#
######################################

417
418
419
    # Note: asking for Queen->fetch_overdue_workers(0) essentially means
    #       "fetch all workers known to the Queen not to be officially dead"
    #
420
421
422
423
424
sub fetch_overdue_workers {
  my ($self,$overdue_secs) = @_;

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

425
426
427
428
  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");
429
430
431
  return $self->_generic_fetch($constraint);
}

432
433
sub fetch_failed_workers {
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
434

435
  my $constraint = "w.cause_of_death='FATALITY' ";
436
437
438
  return $self->_generic_fetch($constraint);
}

439
sub fetch_all_dead_workers_with_jobs {
Leo Gordon's avatar
Leo Gordon committed
440
441
  my $self = shift;

442
  # 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
443

444
445
  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
446
447
  return $self->_generic_fetch($constraint, $join);
}
Jessica Severin's avatar
Jessica Severin committed
448

449
450
=head2 synchronize_hive

Leo Gordon's avatar
Leo Gordon committed
451
  Arg [1]    : $filter_analysis (optional)
452
453
  Example    : $queen->synchronize_hive();
  Description: Runs through all analyses in the system and synchronizes
454
455
              the analysis_stats summary with the states in the job 
              and worker tables.  Then follows by checking all the blocking rules
456
457
458
459
460
461
462
              and blocks/unblocks analyses as needed.
  Exceptions : none
  Caller     : general

=cut

sub synchronize_hive {
Leo Gordon's avatar
Leo Gordon committed
463
  my $self          = shift;
Leo Gordon's avatar
Leo Gordon committed
464
  my $filter_analysis = shift; # optional parameter
465

466
  my $start_time = time();
467

Leo Gordon's avatar
Leo Gordon committed
468
  my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
469

Leo Gordon's avatar
Leo Gordon committed
470
  print STDERR "\nSynchronizing the hive (".scalar(@$list_of_analyses)." analyses this time):\n";
Leo Gordon's avatar
Leo Gordon committed
471
472
  foreach my $analysis (@$list_of_analyses) {
    $self->synchronize_AnalysisStats($analysis->stats);
473
    print STDERR ( ($analysis->stats()->status eq 'BLOCKED') ? 'x' : 'o');
474
  }
Leo Gordon's avatar
Leo Gordon committed
475
  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
476

477
478
479
480
481
482
#  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
483

Leo Gordon's avatar
Leo Gordon committed
484
  print STDERR ''.((time() - $start_time))." seconds to synchronize_hive\n\n";
485
}
486

487

488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
=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
510
                   ($stats->seconds_since_last_update < 3*60));
511
512
513
514

  # 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=" . $stats->analysis_id;
515
  #print("$sql\n");
516
  my $row_count = $self->dbc->do($sql);  
517
  return $stats unless($row_count == 1);        # return the un-updated status if locked
518
  #printf("got sync_lock on analysis_stats(%d)\n", $stats->analysis_id);
519
  
520
      # since we managed to obtain the lock, let's go and perform the sync:
521
  $self->synchronize_AnalysisStats($stats);
522

523
524
525
526
  return $stats;
}


527
=head2 synchronize_AnalysisStats
528

529
530
  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
  Example    : $self->synchronize($analysisStats);
531
  Description: Queries the job and worker tables to get summary counts
532
533
534
535
536
               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
537

538
=cut
539

540
sub synchronize_AnalysisStats {
541
  my $self = shift;
542
  my $analysisStats = shift;
543

544
  return $analysisStats unless($analysisStats);
545
  return $analysisStats unless($analysisStats->analysis_id);
546

547
  $analysisStats->refresh(); ## Need to get the new hive_capacity for dynamic analyses
548
549
  my $hive_capacity = $analysisStats->hive_capacity;

550
551
552
553
  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() );
554

555
556
                # adjust_stats_for_living_workers:
            if($hive_capacity > 0) {
557
                my $unfulfilled_capacity = $hive_capacity - $analysisStats->num_running_workers();
558

559
560
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
561
562
563
564
565
566
567
568
569
570
                }
            }
            $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);
571

572
      my $sql = "SELECT status, semaphore_count, count(*) FROM job ".
573
574
575
                "WHERE analysis_id=? GROUP BY status, semaphore_count";
      my $sth = $self->prepare($sql);
      $sth->execute($analysisStats->analysis_id);
576
577


578
579
580
581
      my $done_here       = 0;
      my $done_elsewhere  = 0;
      my $total_job_count = 0;
      while (my ($status, $semaphore_count, $job_count)=$sth->fetchrow_array()) {
582
    # print STDERR "$status: $job_count\n";
583

584
        $total_job_count += $job_count;
585
586
587
588
589
590
591
592

        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) {
593
                my $unfulfilled_capacity = $hive_capacity - $self->count_running_workers( $analysisStats->analysis_id() );
594

595
596
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
597
                }
598
            }
599
600
            $analysisStats->num_required_workers( $required_workers );

601
        } elsif($status eq 'DONE') {
602
            $done_here = $job_count;
603
        } elsif($status eq 'PASSED_ON') {
604
605
606
            $done_elsewhere = $job_count;
        } elsif ($status eq 'FAILED') {
            $analysisStats->failed_job_count($job_count);
607
        }
608
      } # /while
609
      $sth->finish;
610

611
612
      $analysisStats->total_job_count( $total_job_count );
      $analysisStats->done_job_count( $done_here + $done_elsewhere );
613
  } # /unless $self->{'_hive_use_triggers'}
614

615
  $analysisStats->check_blocking_control_rules();
616

617
618
619
  if($analysisStats->status ne 'BLOCKED') {
    $analysisStats->determine_status();
  }
620

621
  # $analysisStats->sync_lock(0); ## do we perhaps need it here?
622
  $analysisStats->update;  #update and release sync_lock
623

624
  return $analysisStats;
625
}
626
627


Leo Gordon's avatar
Leo Gordon committed
628
629
sub get_num_failed_analyses {
  my ($self, $analysis) = @_;
630
631

  my $statsDBA = $self->db->get_AnalysisStatsAdaptor;
Leo Gordon's avatar
Leo Gordon committed
632
  my $failed_analyses = $statsDBA->fetch_by_statuses(['FAILED']);
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
  if ($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);
}


655
656
sub get_hive_current_load {
  my $self = shift;
657
658
659
  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";
660
661
662
663
664
  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
665
  print("current hive load = $load\n");
666
667
668
669
  return $load;
}


670
671
672
673
674
675
676
677
678
679
680
681
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;
682
683
}

684

685
=head2 schedule_workers
686

687
  Arg[1]     : Bio::EnsEMBL::Analysis object (optional)
688
  Example    : $count = $queen->schedule_workers();
689
690
691
  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
692
               If Arg[1] is defined, does it only for the given analysis.
693
  Exceptions : none
694
  Caller     : beekeepers and other external processes
695
696
697

=cut

698
699
sub schedule_workers {
  my ($self, $filter_analysis, $orig_pending_by_rc_id, $available_submit_limit) = @_;
700

701
  my $statsDBA                      = $self->db->get_AnalysisStatsAdaptor;
702
  my $clearly_needed_analyses       = $statsDBA->fetch_by_needed_workers();
703
704
  my $potentially_needed_analyses   = $statsDBA->fetch_by_statuses(['LOADING', 'BLOCKED', 'ALL_CLAIMED']);
  my @all_analyses                  = (@$clearly_needed_analyses, @$potentially_needed_analyses);
705

706
  return {} unless(@all_analyses);
707

708
709
710
711
  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();
712

713

Leo Gordon's avatar
Leo Gordon committed
714
  foreach my $analysis_stats (@all_analyses) {
715
716
    last if ($available_load <= 0.0);
    last if (defined($available_submit_limit) and !$available_submit_limit);
Leo Gordon's avatar
Leo Gordon committed
717
    next if (defined $filter_analysis && $filter_analysis->dbID != $analysis_stats->analysis_id);
718

Leo Gordon's avatar
Leo Gordon committed
719
        #digging deeper under the surface so need to sync
720
    if(($analysis_stats->status eq 'LOADING') or ($analysis_stats->status eq 'BLOCKED') or ($analysis_stats->status eq 'ALL_CLAIMED')) {
721
722
723
724
725
726
      $self->synchronize_AnalysisStats($analysis_stats);
    }

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

727
        # 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
728
    my $workers_this_analysis = $analysis_stats->num_required_workers;
729

730
731
732
733
734
735
    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
736

737
738
739
740
741
    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
742
743
        }

744
        $available_load -= 1.0*$workers_this_analysis/$hive_capacity;
745
    }
Leo Gordon's avatar
Leo Gordon committed
746

747
    my $curr_rc_id = $analysis_stats->resource_class_id;
748
749
750
    if($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;

751
        print "Scheduler detected $pending_this_analysis pending workers with rc_id=$curr_rc_id\n";
752
753
754
755
        $workers_this_analysis              -= $pending_this_analysis;
        $pending_by_rc_id{ $curr_rc_id }    -= $pending_this_analysis;
    }

756
757
    next unless($workers_this_analysis);    # do not autovivify the hash by a zero

758
759
760
    $total_workers_to_run += $workers_this_analysis;
    $workers_to_run_by_rc_id{ $curr_rc_id } += $workers_this_analysis;
    $analysis_stats->print_stats();
761
    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);
762
  }
763

764
  printf("Scheduler suggests adding a total of %d workers [%1.5f hive_load remaining]\n", $total_workers_to_run, $available_load);
765
  return \%workers_to_run_by_rc_id;
Leo Gordon's avatar
Leo Gordon committed
766
767
}

768

769
sub schedule_workers_resync_if_necessary {
770
771
772
    my ($self, $valley, $analysis) = @_;

    my $meadow              = $valley->get_current_meadow();
Leo Gordon's avatar
Leo Gordon committed
773

774
775
776
    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
777

778
779
780
781
    my $available_submit_limit = ($submit_limit and $meadow_limit)
                                    ? (($submit_limit<$meadow_limit) ? $submit_limit : $meadow_limit)
                                    : (defined($submit_limit) ? $submit_limit : $meadow_limit);

782
    my $workers_to_run_by_rc_id = $self->schedule_workers($analysis, $pending_by_rc_id, $available_submit_limit);
783

784
    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
785
786
        print "*** nothing is running and nothing to do (according to analysis_stats) => perform a hard resync\n" ;

787
        $self->check_for_dead_workers($valley, 1);
788
        $self->synchronize_hive($analysis);
Leo Gordon's avatar
Leo Gordon committed
789

790
        $workers_to_run_by_rc_id = $self->schedule_workers($analysis, $pending_by_rc_id, $available_submit_limit);
Leo Gordon's avatar
Leo Gordon committed
791
792
    }

793
    return $workers_to_run_by_rc_id;
794
795
}

796

Leo Gordon's avatar
Leo Gordon committed
797
sub get_remaining_jobs_show_hive_progress {
798
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
799
  my $sql = "SELECT sum(done_job_count), sum(failed_job_count), sum(total_job_count), ".
800
801
            "sum(unclaimed_job_count * analysis_stats.avg_msec_per_job)/1000/60/60 ".
            "FROM analysis_stats";
802
803
  my $sth = $self->prepare($sql);
  $sth->execute();
804
  my ($done, $failed, $total, $cpuhrs) = $sth->fetchrow_array();
805
  $sth->finish;
806
807
808
809
810
811
812

  $done   ||= 0;
  $failed ||= 0;
  $total  ||= 0;
  my $completed = $total
    ? ((100.0 * ($done+$failed))/$total)
    : 0.0;
813
814
815
  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
816
  return $remaining;
817
818
}

819

Leo Gordon's avatar
Leo Gordon committed
820
sub print_analysis_status {
Leo Gordon's avatar
Leo Gordon committed
821
    my ($self, $filter_analysis) = @_;
822

Leo Gordon's avatar
Leo Gordon committed
823
    my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
824
825
826
    foreach my $analysis (sort {$a->dbID <=> $b->dbID} @$list_of_analyses) {
        $analysis->stats->print_stats($self->{'verbose_stats'});
    }
827
828
829
}


Leo Gordon's avatar
Leo Gordon committed
830
sub print_running_worker_counts {
831
  my $self = shift;
832

Leo Gordon's avatar
Leo Gordon committed
833
  print "\n===== Stats of live Workers according to the Queen: ======\n";
834
835
836
  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
837

838
  my $total_workers = 0;
839
840
  my $sth = $self->prepare($sql);
  $sth->execute();
841
  while((my $logic_name, my $worker_count)=$sth->fetchrow_array()) {
Leo Gordon's avatar
Leo Gordon committed
842
    printf("%30s : %d workers\n", $logic_name, $worker_count);
843
    $total_workers += $worker_count;
844
  }
845
  $sth->finish;
Leo Gordon's avatar
Leo Gordon committed
846
  printf("%30s : %d workers\n\n", '======= TOTAL =======', $total_workers);
847
}
848

849
850
851
852
853
854
855
856
857
858
859
=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
860
sub monitor {
861
862
863
864
  my $self = shift;
  my $sql = qq{
      INSERT INTO monitor
      SELECT
865
          CURRENT_TIMESTAMP,
866
          count(*),
Leo Gordon's avatar
Leo Gordon committed
867
868
869
870
871
872
  }. ( ($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{
873
          group_concat(DISTINCT logic_name)
874
      FROM worker left join analysis USING (analysis_id)
875