Queen.pm 30 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
use File::Path 'make_path';
65

66
use Bio::EnsEMBL::Utils::Argument ('rearrange');
67

68
use Bio::EnsEMBL::Hive::Utils ('destringify', 'dir_revhash');  # NB: needed by invisible code
69
use Bio::EnsEMBL::Hive::AnalysisJob;
Jessica Severin's avatar
Jessica Severin committed
70
71
use Bio::EnsEMBL::Hive::Worker;

72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
use base ('Bio::EnsEMBL::Hive::DBSQL::ObjectAdaptor');


sub default_table_name {
    return 'worker';
}


sub default_insertion_method {
    return 'INSERT';
}


sub object_class {
    return 'Bio::EnsEMBL::Hive::Worker';
}
88

Jessica Severin's avatar
Jessica Severin committed
89

90
############################
Jessica Severin's avatar
Jessica Severin committed
91
#
92
# PUBLIC API
Jessica Severin's avatar
Jessica Severin committed
93
#
94
############################
Jessica Severin's avatar
Jessica Severin committed
95

96

Jessica Severin's avatar
Jessica Severin committed
97
98
=head2 create_new_worker

99
100
101
102
  Description: Creates an entry in the worker table,
               populates some non-storable attributes
               and returns a Worker object based on that insert.
               This guarantees that each worker registered in this Queen's hive is properly registered.
Jessica Severin's avatar
Jessica Severin committed
103
  Returntype : Bio::EnsEMBL::Hive::Worker
104
  Caller     : runWorker.pl
Jessica Severin's avatar
Jessica Severin committed
105
106
107
108

=cut

sub create_new_worker {
109
    my ($self, @args) = @_;
110

111
    my ($meadow_type, $meadow_name, $process_id, $exec_host, $resource_class_id, $resource_class_name,
112
        $no_write, $debug, $worker_log_dir, $hive_log_dir, $job_limit, $life_span, $no_cleanup, $retry_throwing_jobs, $can_respecialize) =
113
114

    rearrange([qw(meadow_type meadow_name process_id exec_host resource_class_id resource_class_name
115
                no_write debug worker_log_dir hive_log_dir job_limit life_span no_cleanup retry_throwing_jobs can_respecialize) ], @args);
116

117
    if( defined($resource_class_name) ) {
118
119
120
121
122
123
124
125
126
127
128
129
        my $rc = $self->db->get_ResourceClassAdaptor->fetch_by_name($resource_class_name)
            or die "resource_class with name='$resource_class_name' could not be fetched from the database";

        $resource_class_id = $rc->dbID;
    }

    my $sql = q{INSERT INTO worker (born, last_check_in, meadow_type, meadow_name, host, process_id, resource_class_id)
              VALUES (CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, ?, ?, ?, ?, ?)};

    my $sth = $self->prepare($sql);
    $sth->execute($meadow_type, $meadow_name, $exec_host, $process_id, $resource_class_id);
    my $worker_id = $self->dbc->db_handle->last_insert_id(undef, undef, 'worker', 'worker_id')
130
        or die "Could not insert a new worker";
131
132
133
134
135
136
    $sth->finish;

    if($hive_log_dir or $worker_log_dir) {
        my $dir_revhash = dir_revhash($worker_id);
        $worker_log_dir ||= $hive_log_dir .'/'. ($dir_revhash ? "$dir_revhash/" : '') .'worker_id_'.$worker_id;

137
138
        eval {
            make_path( $worker_log_dir );
139
140
            1;
        } or die "Could not create '$worker_log_dir' directory : $@";
141
142
143
144
145
146
147
148
149
150
151

        my $sth_add_log = $self->prepare( "UPDATE worker SET log_dir=? WHERE worker_id=?" );
        $sth_add_log->execute($worker_log_dir, $worker_id);
        $sth_add_log->finish;
    }

    my $worker = $self->fetch_by_dbID($worker_id)
        or die "Could not fetch worker with dbID=$worker_id";

    $worker->init;

152
153
    if(defined($job_limit)) {
      $worker->job_limiter($job_limit);
154
155
156
157
158
159
160
161
162
163
164
165
166
      $worker->life_span(0);
    }

    $worker->life_span($life_span * 60)                 if($life_span);

    $worker->execute_writes(0)                          if($no_write);

    $worker->perform_cleanup(0)                         if($no_cleanup);

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

    $worker->retry_throwing_jobs($retry_throwing_jobs)  if(defined $retry_throwing_jobs);

167
168
    $worker->can_respecialize($can_respecialize)        if(defined $can_respecialize);

169
170
171
172
173
    return $worker;
}


=head2 specialize_new_worker
174

175
176
  Description: If analysis_id or logic_name is specified it will try to specialize the Worker into this analysis.
               If not specified the Queen will analyze the hive and pick the most suitable analysis.
177
  Caller     : Bio::EnsEMBL::Hive::Worker
178
179
180
181
182
183
184
185

=cut

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

    my ($analysis_id, $logic_name, $job_id, $force) =
        rearrange([qw(analysis_id logic_name job_id force) ], @args);
186

187
188
    if( scalar( grep {defined($_)} ($analysis_id, $logic_name, $job_id) ) > 1) {
        die "At most one of the options {-analysis_id, -logic_name, -job_id} can be set to pre-specialize a Worker";
Leo Gordon's avatar
Leo Gordon committed
189
190
    }

191
    my ($analysis, $stats, $special_batch);
192
    my $analysis_stats_adaptor = $self->db->get_AnalysisStatsAdaptor;
Jessica Severin's avatar
Jessica Severin committed
193

194
    if($job_id or $analysis_id or $logic_name) {    # probably pre-specialized from command-line
195

196
        if($job_id) {
197
198
199
            print "resetting and fetching job for job_id '$job_id'\n";

            my $job_adaptor = $self->db->get_AnalysisJobAdaptor;
200

201
202
203
204
205
206
207
208
209
            my $job = $job_adaptor->fetch_by_dbID( $job_id )
                or die "Could not fetch job with dbID='$job_id'";
            my $job_status = $job->status();

            if($job_status =~/(CLAIMED|PRE_CLEANUP|FETCH_INPUT|RUN|WRITE_OUTPUT|POST_CLEANUP)/ ) {
                die "Job with dbID='$job_id' is already in progress, cannot run";   # FIXME: try GC first, then complain
            } elsif($job_status =~/(DONE|SEMAPHORED)/ and !$force) {
                die "Job with dbID='$job_id' is $job_status, please use -force 1 to override";
            }
210

211
212
213
214
215
216
217
218
            if(($job_status eq 'DONE') and $job->semaphored_job_id) {
                warn "Increasing the semaphore count of the dependent job";
                $job_adaptor->increase_semaphore_count_for_jobid( $job->semaphored_job_id );
            }

            my $worker_id = $worker->dbID;
            if($job = $job_adaptor->reset_or_grab_job_by_dbID($job_id, $worker_id)) {
                $special_batch = [ $job ];
219
220
                $analysis_id = $job->analysis_id;
            } else {
221
                die "Could not claim job with dbID='$job_id' for worker with dbID='$worker_id'";
222
            }
223
224
225
226
227
228
229
230
231
232
233
234
235
        }

        if($logic_name) {
            $analysis = $self->db->get_AnalysisAdaptor->fetch_by_logic_name($logic_name)
                or die "analysis with name='$logic_name' could not be fetched from the database";

            $analysis_id = $analysis->dbID;

        } elsif($analysis_id) {
            $analysis = $self->db->get_AnalysisAdaptor->fetch_by_dbID($analysis_id)
                or die "analysis with dbID='$analysis_id' could not be fetched from the database";
        }

236
237
238
        if( $worker->resource_class_id
        and $worker->resource_class_id != $analysis->resource_class_id) {
                die "resource_class of analysis ".$analysis->logic_name." is incompatible with this Worker's resource_class";
239
240
        }

241
242
243
        $stats = $analysis_stats_adaptor->fetch_by_analysis_id($analysis_id);
        $self->safe_synchronize_AnalysisStats($stats);

244
        unless($special_batch or $force) {    # do we really need to run this analysis?
245
            if($self->get_hive_current_load() >= 1.1) {
246
                $worker->cause_of_death('HIVE_OVERLOAD');
247
                die "Hive is overloaded, can't specialize a worker";
248
249
            }
            if($stats->status eq 'BLOCKED') {
250
                die "Analysis is BLOCKED, can't specialize a worker";
251
            }
252
            if($stats->num_required_workers <= 0) {
253
                die "Analysis doesn't require extra workers at the moment";
254
255
            }
            if($stats->status eq 'DONE') {
256
                die "Analysis is DONE, and doesn't require workers";
257
258
            }
        }
259
260
            # probably scheduled by beekeeper.pl:
    } elsif( $stats = $self->suggest_analysis_to_specialize_by_rc_id_meadow_type($worker->resource_class_id, $worker->meadow_type) ) {
261

262
        $worker->analysis( undef ); # make sure we reset anything that was there before
263
        $analysis_id = $stats->analysis_id;
264
265
266
    } else {
        $worker->cause_of_death('NO_ROLE');
        die "No analysis suitable for the worker was found\n";
267
    }
268

269
        # now set it in the $worker:
Jessica Severin's avatar
Jessica Severin committed
270

271
    $worker->analysis_id( $analysis_id );
Jessica Severin's avatar
Jessica Severin committed
272

273
274
275
276
    my $sth_update_analysis_id = $self->prepare( "UPDATE worker SET analysis_id=? WHERE worker_id=?" );
    $sth_update_analysis_id->execute($worker->analysis_id, $worker->dbID);
    $sth_update_analysis_id->finish;

277
278
    if($special_batch) {
        $worker->special_batch( $special_batch );
279
    } else {    # count it as autonomous worker sharing the load of that analysis:
280
281
282

        $stats->update_status('WORKING');

283
        $analysis_stats_adaptor->decrease_required_workers($worker->analysis_id);
284
285
286
287
288
289
290
291
292
    }

        # 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() ) {
293
        $analysis_stats_adaptor->increase_running_workers($worker->analysis_id);
294
    }
Jessica Severin's avatar
Jessica Severin committed
295
296
}

297

Jessica Severin's avatar
Jessica Severin committed
298
sub register_worker_death {
299
    my ($self, $worker, $self_burial) = @_;
300

301
302
303
304
    return unless($worker);

    my $cod = $worker->cause_of_death() || 'UNKNOWN';    # make sure we do not attempt to insert a void

305
                    # FIXME: make it possible to set the 'died' timestamp if we have detected it from logs:
306
    my $sql = qq{UPDATE worker SET died=CURRENT_TIMESTAMP
307
    } . ( $self_burial ? ',last_check_in=CURRENT_TIMESTAMP ' : '') . qq{
308
309
310
311
312
313
314
315
316
317
318
319
320
                    ,status='DEAD'
                    ,work_done='}. $worker->work_done . qq{'
                    ,cause_of_death='$cod'
                WHERE worker_id='}. $worker->dbID . qq{'};
    $self->dbc->do( $sql );

    if(my $analysis_id = $worker->analysis_id) {
        my $analysis_stats_adaptor = $self->db->get_AnalysisStatsAdaptor;

        unless( $self->db->hive_use_triggers() ) {
            $analysis_stats_adaptor->decrease_running_workers($worker->analysis_id);
        }

321
322
323
324
325
        unless( $cod eq 'NO_WORK'
            or  $cod eq 'JOB_LIMIT'
            or  $cod eq 'HIVE_OVERLOAD'
            or  $cod eq 'LIFESPAN'
        ) {
326
327
328
329
330
331
332
333
334
335
                $self->db->get_AnalysisJobAdaptor->release_undone_jobs_from_worker($worker);
        }

            # re-sync the analysis_stats when a worker dies as part of dynamic sync system
        if($self->safe_synchronize_AnalysisStats($worker->analysis->stats)->status ne 'DONE') {
            # 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
            $analysis_stats_adaptor->increase_required_workers($worker->analysis_id);
        }
    }
Jessica Severin's avatar
Jessica Severin committed
336
337
}

338

339
sub check_for_dead_workers {    # scans the whole Valley for lost Workers (but ignores unreachagle ones)
340
    my ($self, $valley, $check_buried_in_haste) = @_;
Leo Gordon's avatar
Leo Gordon committed
341

342
    warn "GarbageCollector:\tChecking for lost Workers...\n";
343

344
345
    my $last_few_seconds            = 5;    # FIXME: It is probably a good idea to expose this parameter for easier tuning.
    my $queen_overdue_workers       = $self->fetch_overdue_workers( $last_few_seconds );    # check the workers we have not seen active during the $last_few_seconds
346
347
348
    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
349

350
    warn "GarbageCollector:\t[Queen:] out of ".scalar(@$queen_overdue_workers)." Workers that haven't checked in during the last $last_few_seconds seconds...\n";
351

352
    foreach my $worker (@$queen_overdue_workers) {
Leo Gordon's avatar
Leo Gordon committed
353

354
355
356
        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
357
        } else {
358
            $worker_status_counts{$meadow_type}{'UNREACHABLE'}++;
359

360
            next;   # Worker is unreachable from this Valley
Leo Gordon's avatar
Leo Gordon committed
361
362
        }

363
364
365
366
367
        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'}++;
368

369
            $mt_and_pid_to_lost_worker{$meadow_type}{$process_id} = $worker;
370
        }
371
372
373
374
375
376
    }

        # 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";
    }
377

378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
    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) {
395
                $worker->cause_of_death( $wpid_to_cod->{$process_id} || 'UNKNOWN');
396
397
                $self->register_worker_death($worker);
            }
398
399
400
        }
    }

401
        # the following bit is completely Meadow-agnostic and only restores database integrity:
Leo Gordon's avatar
Leo Gordon committed
402
    if($check_buried_in_haste) {
403
        warn "GarbageCollector:\tChecking for Workers buried in haste...\n";
404
        my $buried_in_haste_list = $self->fetch_all_dead_workers_with_jobs();
Leo Gordon's avatar
Leo Gordon committed
405
        if(my $bih_number = scalar(@$buried_in_haste_list)) {
406
            warn "GarbageCollector:\tfound $bih_number jobs, reclaiming.\n\n";
Leo Gordon's avatar
Leo Gordon committed
407
408
409
            if($bih_number) {
                my $job_adaptor = $self->db->get_AnalysisJobAdaptor();
                foreach my $worker (@$buried_in_haste_list) {
Leo Gordon's avatar
Leo Gordon committed
410
                    $job_adaptor->release_undone_jobs_from_worker($worker);
Leo Gordon's avatar
Leo Gordon committed
411
412
413
                }
            }
        } else {
414
            warn "GarbageCollector:\tfound none\n";
Leo Gordon's avatar
Leo Gordon committed
415
416
417
        }
    }
}
Jessica Severin's avatar
Jessica Severin committed
418
419


420
421
422
    # 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
423

424
    $self->dbc->do("UPDATE worker SET last_check_in=CURRENT_TIMESTAMP, status='".$worker->status."', work_done='".$worker->work_done."' WHERE worker_id='".$worker->dbID."'");
Jessica Severin's avatar
Jessica Severin committed
425
426
427
}


428
=head2 reset_job_by_dbID_and_sync
429

430
  Arg [1]: int $job_id
431
  Example: 
432
    my $job = $queen->reset_job_by_dbID_and_sync($job_id);
433
  Description: 
434
    For the specified job_id it will fetch just that job, 
435
436
437
    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 
438
    same job simultaneously (use only for debugging).
439
  Returntype : none
440
  Exceptions :
441
  Caller     : beekeeper.pl
442
443
444

=cut

445
446
sub reset_job_by_dbID_and_sync {
    my ($self, $job_id) = @_;
447

448
    my $job_adaptor = $self->db->get_AnalysisJobAdaptor;
449
    my $job = $job_adaptor->reset_or_grab_job_by_dbID($job_id); 
450
451
452

    my $stats = $self->db->get_AnalysisStatsAdaptor->fetch_by_analysis_id($job->analysis_id);
    $self->synchronize_AnalysisStats($stats);
453
454
455
}


456
457
458
459
460
461
######################################
#
# Public API interface for beekeeper
#
######################################

462

463
464
465
    # Note: asking for Queen->fetch_overdue_workers(0) essentially means
    #       "fetch all workers known to the Queen not to be officially dead"
    #
466
sub fetch_overdue_workers {
467
    my ($self,$overdue_secs) = @_;
468

469
    $overdue_secs = 3600 unless(defined($overdue_secs));
470

471
472
473
474
475
476
    my $constraint = "status!='DEAD' AND ".{
            'mysql'     =>  "(UNIX_TIMESTAMP()-UNIX_TIMESTAMP(last_check_in)) > $overdue_secs",
            'sqlite'    =>  "(strftime('%s','now')-strftime('%s',last_check_in)) > $overdue_secs",
            'pgsql'     =>  "EXTRACT(EPOCH FROM CURRENT_TIMESTAMP - last_check_in) > $overdue_secs",
        }->{ $self->dbc->driver };

477
    return $self->fetch_all( $constraint );
478
479
}

480
481
482

sub fetch_all_dead_workers_with_jobs {
    my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
483

484
    return $self->fetch_all( "JOIN job j USING(worker_id) WHERE worker.status='DEAD' AND j.status NOT IN ('DONE', 'READY', 'FAILED', 'PASSED_ON') GROUP BY worker_id" );
Leo Gordon's avatar
Leo Gordon committed
485
}
Jessica Severin's avatar
Jessica Severin committed
486

487

488
489
=head2 synchronize_hive

Leo Gordon's avatar
Leo Gordon committed
490
  Arg [1]    : $filter_analysis (optional)
491
492
  Example    : $queen->synchronize_hive();
  Description: Runs through all analyses in the system and synchronizes
493
494
              the analysis_stats summary with the states in the job 
              and worker tables.  Then follows by checking all the blocking rules
495
496
497
498
499
500
501
              and blocks/unblocks analyses as needed.
  Exceptions : none
  Caller     : general

=cut

sub synchronize_hive {
Leo Gordon's avatar
Leo Gordon committed
502
  my $self          = shift;
Leo Gordon's avatar
Leo Gordon committed
503
  my $filter_analysis = shift; # optional parameter
504

505
  my $start_time = time();
506

Leo Gordon's avatar
Leo Gordon committed
507
  my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
508

Leo Gordon's avatar
Leo Gordon committed
509
  print STDERR "\nSynchronizing the hive (".scalar(@$list_of_analyses)." analyses this time):\n";
Leo Gordon's avatar
Leo Gordon committed
510
511
  foreach my $analysis (@$list_of_analyses) {
    $self->synchronize_AnalysisStats($analysis->stats);
512
    print STDERR ( ($analysis->stats()->status eq 'BLOCKED') ? 'x' : 'o');
513
  }
Leo Gordon's avatar
Leo Gordon committed
514
  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
515

Leo Gordon's avatar
Leo Gordon committed
516
  print STDERR ''.((time() - $start_time))." seconds to synchronize_hive\n\n";
517
}
518

519

520
521
522
=head2 safe_synchronize_AnalysisStats

  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
523
  Example    : $self->safe_synchronize_AnalysisStats($stats);
524
525
526
527
528
529
530
531
532
  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 {
533
534
535
536
537
538
539
540
541
542
    my ($self, $stats) = @_;

    my $max_refresh_attempts = 5;
    while($stats->sync_lock and $max_refresh_attempts--) {   # another Worker/Beekeeper is synching this analysis right now
        sleep(1);
        $stats->refresh();  # just try to avoid collision
    }

    return $stats if($stats->status eq 'DONE');
    return $stats if(($stats->status eq 'WORKING') and
543
                   ($stats->seconds_since_last_update < 3*60));
544

545
546
547
548
549
        # 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;
    my $row_count = $self->dbc->do($sql);  
    return $stats unless($row_count == 1);        # return the un-updated status if locked
550
  
551
552
        # if we managed to obtain the lock, let's go and perform the sync:
    $self->synchronize_AnalysisStats($stats);
553

554
    return $stats;
555
556
557
}


558
=head2 synchronize_AnalysisStats
559

560
561
  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
  Example    : $self->synchronize($analysisStats);
562
  Description: Queries the job and worker tables to get summary counts
563
564
565
566
567
               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
568

569
=cut
570

571
sub synchronize_AnalysisStats {
572
  my $self = shift;
573
  my $analysisStats = shift;
574

575
  return $analysisStats unless($analysisStats);
576
  return $analysisStats unless($analysisStats->analysis_id);
577

578
  $analysisStats->refresh(); ## Need to get the new hive_capacity for dynamic analyses
579

580

581
  unless($self->db->hive_use_triggers()) {
582
      $analysisStats->total_job_count(0);
583
584
      $analysisStats->semaphored_job_count(0);
      $analysisStats->ready_job_count(0);
585
586
      $analysisStats->done_job_count(0);
      $analysisStats->failed_job_count(0);
587

588
            # ask for analysis_id to force MySQL to use existing index on (analysis_id, status)
589
      my $sql = "SELECT analysis_id, status, count(*) FROM job WHERE analysis_id=? GROUP BY analysis_id, status";
590
591
      my $sth = $self->prepare($sql);
      $sth->execute($analysisStats->analysis_id);
592

593
594
595
      my $done_here       = 0;
      my $done_elsewhere  = 0;
      my $total_job_count = 0;
596
      while (my ($dummy_analysis_id, $status, $job_count)=$sth->fetchrow_array()) {
597
    # print STDERR "$status: $job_count\n";
598

599
        $total_job_count += $job_count;
600

601
        if($status eq 'READY') {
602
603
604
            $analysisStats->ready_job_count($job_count);
        } elsif($status eq 'SEMAPHORED') {
            $analysisStats->semaphored_job_count($job_count);
605
        } elsif($status eq 'DONE') {
606
            $done_here = $job_count;
607
        } elsif($status eq 'PASSED_ON') {
608
609
610
            $done_elsewhere = $job_count;
        } elsif ($status eq 'FAILED') {
            $analysisStats->failed_job_count($job_count);
611
        }
612
      } # /while
613
      $sth->finish;
614

615
616
      $analysisStats->total_job_count( $total_job_count );
      $analysisStats->done_job_count( $done_here + $done_elsewhere );
617
  } # unless($self->db->hive_use_triggers())
618

619

620
621
622
623
624
        # compute the number of total required workers for this analysis (taking into account the jobs that are already running)
    my $analysis              = $analysisStats->get_analysis();
    my $scheduling_allowed    =  ( !defined( $analysisStats->hive_capacity ) or $analysisStats->hive_capacity )
                              && ( !defined( $analysis->analysis_capacity  ) or $analysis->analysis_capacity  );
    my $required_workers    = $scheduling_allowed
625
                            && POSIX::ceil( $analysisStats->ready_job_count() / $analysisStats->get_or_estimate_batch_size() );
626
627
628
629
630
631
632
633
    $analysisStats->num_required_workers( $required_workers );


    $analysisStats->check_blocking_control_rules();

    if($analysisStats->status ne 'BLOCKED') {
        $analysisStats->determine_status();
    }
634

635
636
    # $analysisStats->sync_lock(0); ## do we perhaps need it here?
    $analysisStats->update;  #update and release sync_lock
637

638
    return $analysisStats;
639
}
640
641


642
643
644
645
646
647
648
649
650
651
652
653
654
655
=head2 get_num_failed_analyses

  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object (optional)
  Example    : if( $self->get_num_failed_analyses( $my_analysis )) { do_something; }
  Example    : my $num_failed_analyses = $self->get_num_failed_analyses();
  Description: Reports all failed analyses and returns
                either the number of total failed (if no $filter_analysis was provided)
                or 1/0, depending on whether $filter_analysis failed or not.
  Returntype : int
  Exceptions : none
  Caller     : general

=cut

Leo Gordon's avatar
Leo Gordon committed
656
sub get_num_failed_analyses {
657
    my ($self, $filter_analysis) = @_;
658

659
660
661
662
663
664
665
666
667
668
669
670
    my $failed_analyses = $self->db->get_AnalysisAdaptor->fetch_all_failed_analyses();

    my $filter_analysis_failed = 0;

    foreach my $failed_analysis (@$failed_analyses) {
        print "\t##########################################################\n";
        print "\t# Too many jobs in analysis '".$failed_analysis->logic_name."' FAILED #\n";
        print "\t##########################################################\n\n";
        if($filter_analysis and ($filter_analysis->dbID == $failed_analysis)) {
            $filter_analysis_failed = 1;
        }
    }
671

672
    return $filter_analysis ? $filter_analysis_failed : scalar(@$failed_analyses);
673
674
675
}


676
sub get_hive_current_load {
677
678
679
    my $self = shift;
    my $sql = qq{
        SELECT sum(1/hive_capacity)
680
681
682
        FROM worker w
        JOIN analysis_stats USING(analysis_id)
        WHERE w.status!='DEAD'
683
        AND hive_capacity IS NOT NULL
684
        AND hive_capacity>0
685
686
687
688
689
690
    };
    my $sth = $self->prepare($sql);
    $sth->execute();
    my ($load)=$sth->fetchrow_array();
    $sth->finish;
    return ($load || 0);
691
692
693
}


694
695
696
sub count_running_workers {
    my ($self, $analysis_id) = @_;

697
698
    return $self->count_all( "status!='DEAD'".($analysis_id ? " AND analysis_id=$analysis_id" : '') );
}
699
700


701
702
703
704
sub get_workers_rank {
    my ($self, $worker) = @_;

    return $self->count_all( "status!='DEAD' AND analysis_id=".$worker->analysis_id." AND worker_id<".$worker->dbID );
705
706
}

707

Leo Gordon's avatar
Leo Gordon committed
708
sub get_remaining_jobs_show_hive_progress {
709
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
710
  my $sql = "SELECT sum(done_job_count), sum(failed_job_count), sum(total_job_count), ".
711
            "sum(ready_job_count * analysis_stats.avg_msec_per_job)/1000/60/60 ".
712
            "FROM analysis_stats";
713
714
  my $sth = $self->prepare($sql);
  $sth->execute();
715
  my ($done, $failed, $total, $cpuhrs) = $sth->fetchrow_array();
716
  $sth->finish;
717
718
719
720
721
722
723

  $done   ||= 0;
  $failed ||= 0;
  $total  ||= 0;
  my $completed = $total
    ? ((100.0 * ($done+$failed))/$total)
    : 0.0;
724
725
726
  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
727
  return $remaining;
728
729
}

730

Leo Gordon's avatar
Leo Gordon committed
731
sub print_analysis_status {
Leo Gordon's avatar
Leo Gordon committed
732
    my ($self, $filter_analysis) = @_;
733

Leo Gordon's avatar
Leo Gordon committed
734
    my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
735
    foreach my $analysis (sort {$a->dbID <=> $b->dbID} @$list_of_analyses) {
736
        $analysis->stats->print_stats();
Leo Gordon's avatar
Leo Gordon committed
737
    }
738
739
740
}


Leo Gordon's avatar
Leo Gordon committed
741
sub print_running_worker_counts {
742
    my $self = shift;
743

744
745
746
    my $sql = qq{
        SELECT logic_name, count(*)
        FROM worker w
747
        JOIN analysis_base a USING(analysis_id)
748
        WHERE w.status!='DEAD'
749
        GROUP BY a.analysis_id
750
    };
Leo Gordon's avatar
Leo Gordon committed
751

752
753
754
755
756
757
758
759
760
761
762
    my $total_workers = 0;
    my $sth = $self->prepare($sql);
    $sth->execute();

    print "\n===== Stats of live Workers according to the Queen: ======\n";
    while((my $logic_name, my $worker_count)=$sth->fetchrow_array()) {
        printf("%30s : %d workers\n", $logic_name, $worker_count);
        $total_workers += $worker_count;
    }
    $sth->finish;
    printf("%30s : %d workers\n\n", '======= TOTAL =======', $total_workers);
763
}
764

765

766
767
768
769
770
771
772
773
774
775
776
=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
777
sub monitor {
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
    my $self = shift;
    my $sql = qq{
        INSERT INTO monitor
        SELECT CURRENT_TIMESTAMP, count(*),
    } . {
        'mysql'     =>  qq{ sum(work_done/(UNIX_TIMESTAMP()-UNIX_TIMESTAMP(born))),
                            sum(work_done/(UNIX_TIMESTAMP()-UNIX_TIMESTAMP(born)))/count(*), },
        'sqlite'    =>  qq{ sum(work_done/(strftime('%s','now')-strftime('%s',born))),
                            sum(work_done/(strftime('%s','now')-strftime('%s',born)))/count(*), },
        'pgsql'     =>  qq{ sum(work_done/(EXTRACT(EPOCH FROM CURRENT_TIMESTAMP - born))),
                            sum(work_done/(EXTRACT(EPOCH FROM CURRENT_TIMESTAMP - born)))/count(*), },
    }->{ $self->dbc->driver }. qq{
        group_concat(DISTINCT logic_name)
        FROM worker w
        LEFT JOIN analysis_base USING (analysis_id)
        WHERE w.status!='DEAD'
    };
795
796
797
798
799
      
  my $sth = $self->prepare($sql);
  $sth->execute();
}

800

Leo Gordon's avatar
Leo Gordon committed
801
802
803
804
805
806
807
808
809
810
811
812
=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;

813
814
    my $all_workers_considered_alive = $self->fetch_all( "status!='DEAD'" );
    foreach my $worker (@{$all_workers_considered_alive}) {
815
        $worker->cause_of_death( 'UNKNOWN' );  # well, maybe we could have investigated further...
Leo Gordon's avatar
Leo Gordon committed
816
817
818
819
        $self->register_worker_death($worker);
    }
}

820

821
sub suggest_analysis_to_specialize_by_rc_id_meadow_type {
822
    my $self                = shift;
823
    my $rc_id               = shift;
824
    my $meadow_type         = shift;
825

826
    my @suitable_analyses = @{ $self->db->get_AnalysisStatsAdaptor->fetch_all_by_suitability_rc_id_meadow_type( $rc_id, $meadow_type ) };
827

828
    foreach my $stats (@suitable_analyses) {
829

830
            #synchronize and double check that it can be run:
831
        $self->safe_synchronize_AnalysisStats($stats);
832
        return $stats if( ($stats->status ne 'BLOCKED') and ($stats->status ne 'SYNCHING') and ($stats->num_required_workers > 0) );
833
    }
834

835
    return undef;
836
837
838
}


Jessica Severin's avatar
Jessica Severin committed
839
1;