Queen.pm 33.6 KB
Newer Older
Jessica Severin's avatar
Jessica Severin committed
1 2 3 4 5 6
#
# You may distribute this module under the same terms as perl itself

=pod 

=head1 NAME
7

8
  Bio::EnsEMBL::Hive::Queen
Jessica Severin's avatar
Jessica Severin committed
9 10

=head1 SYNOPSIS
11

12
  The Queen of the Hive based job control system
Jessica Severin's avatar
Jessica Severin committed
13 14

=head1 DESCRIPTION
15

16 17 18 19 20 21 22 23 24 25 26 27
  The Queen of the Hive based job control system is responsible to 'birthing' the
  correct number of workers of the right type so that they can find jobs to do.
  It will also free up jobs of Workers that died unexpectantly so that other workers
  can claim them to do.

  Hive based processing is a concept based on a more controlled version
  of an autonomous agent type system.  Each worker is not told what to do
  (like a centralized control system - like the current pipeline system)
  but rather queries a central database for jobs (give me jobs).

  Each worker is linked to an analysis_id, registers its self on creation
  into the Hive, creates a RunnableDB instance of the Analysis->module,
28 29
  gets $analysis->stats->batch_size jobs from the job table, does its work,
  creates the next layer of job entries by interfacing to
30 31 32
  the DataflowRuleAdaptor to determine the analyses it needs to pass its
  output data to and creates jobs on the next analysis database.
  It repeats this cycle until it has lived its lifetime or until there are no
33 34 35 36 37 38 39 40 41 42
  more jobs left.
  The lifetime limit is just a safety limit to prevent these from 'infecting'
  a system.

  The Queens job is to simply birth Workers of the correct analysis_id to get the
  work down.  The only other thing the Queen does is free up jobs that were
  claimed by Workers that died unexpectantly so that other workers can take
  over the work.

  The Beekeeper is in charge of interfacing between the Queen and a compute resource
43
  or 'compute farm'.  Its job is to query Queens if they need any workers and to
44 45 46
  send the requested number of workers to open machines via the runWorker.pl script.
  It is also responsible for interfacing with the Queen to identify worker which died
  unexpectantly.
Jessica Severin's avatar
Jessica Severin committed
47 48

=head1 CONTACT
49

50
  Please contact ehive-users@ebi.ac.uk mailing list with questions/suggestions.
Jessica Severin's avatar
Jessica Severin committed
51 52

=head1 APPENDIX
53

54 55
  The rest of the documentation details each of the object methods. 
  Internal methods are usually preceded with a _
Jessica Severin's avatar
Jessica Severin committed
56 57 58

=cut

59

Jessica Severin's avatar
Jessica Severin committed
60 61 62
package Bio::EnsEMBL::Hive::Queen;

use strict;
63
use POSIX;
64 65 66
use Bio::EnsEMBL::Utils::Argument;
use Bio::EnsEMBL::Utils::Exception;

67
use Bio::EnsEMBL::Hive::Utils ('destringify', 'dir_revhash');  # import 'destringify()' and 'dir_revhash()'
68
use Bio::EnsEMBL::Hive::AnalysisJob;
Jessica Severin's avatar
Jessica Severin committed
69 70
use Bio::EnsEMBL::Hive::Worker;

71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
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';
}
87

Jessica Severin's avatar
Jessica Severin committed
88

89
############################
Jessica Severin's avatar
Jessica Severin committed
90
#
91
# PUBLIC API for Workers
Jessica Severin's avatar
Jessica Severin committed
92
#
93
############################
Jessica Severin's avatar
Jessica Severin committed
94 95 96

=head2 create_new_worker

97
  Arg [1]    : $analysis_id (optional)
Jessica Severin's avatar
Jessica Severin committed
98
  Example    :
99 100 101
  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.
102
               It creates an entry in the worker table, and returns a Worker object 
103 104
               based on that insert.  This guarantees that each worker registered
               in this queens hive is properly registered.
Jessica Severin's avatar
Jessica Severin committed
105 106 107 108 109 110 111
  Returntype : Bio::EnsEMBL::Hive::Worker
  Exceptions :
  Caller     :

=cut

sub create_new_worker {
112
    my ($self, @args) = @_;
113

114
    my (  $meadow_type, $meadow_name, $process_id, $exec_host,
115
        $rc_id, $rc_name, $analysis_id, $logic_name, $job_id,
116
        $no_write, $debug, $worker_log_dir, $hive_log_dir, $job_limit, $life_span, $no_cleanup, $retry_throwing_jobs, $compile_module_once, $force) =
117

118 119 120
    rearrange([qw(meadow_type meadow_name process_id exec_host
            rc_id rc_name analysis_id logic_name job_id
            no_write debug worker_log_dir hive_log_dir job_limit life_span no_cleanup retry_throwing_jobs compile_module_once force) ], @args);
121

122 123
    if( scalar( grep {defined($_)} ($rc_id, $rc_name, $analysis_id, $logic_name, $job_id) ) > 1) {
        die "At most one of the options {-rc_id, -rc_name, -analysis_id, -logic_name, -job_id} can be set to pre-specialize a Worker";
Leo Gordon's avatar
Leo Gordon committed
124 125
    }

126 127
    my ($job, $analysis, $stats);
    my $analysis_stats_adaptor = $self->db->get_AnalysisStatsAdaptor;
Jessica Severin's avatar
Jessica Severin committed
128

129
    if($job_id or $analysis_id or $logic_name) {    # probably pre-specialized by running a runWorker.pl by hand
130

131
        if($job_id) {
132 133 134 135
            print "resetting and fetching job for job_id '$job_id'\n";

            my $job_adaptor = $self->db->get_AnalysisJobAdaptor;
            $job_adaptor->reset_job_by_dbID($job_id); 
136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168

            $job = $job_adaptor->fetch_by_dbID($job_id)
                or die "job with dbID='$job_id' could not be fetched from the database";

            $analysis_id = $job->analysis_id;
        }

        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";
        }

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

        unless($job or $force) {    # do we really need to run this analysis?
            if($self->get_hive_current_load() >= 1.1) {
                die "Hive is overloaded, can't create a worker";
            }
            if($stats->status eq 'BLOCKED') {
                die "Analysis is BLOCKED, can't create workers";
            }
            if($stats->num_required_workers <= 0) {
                die "Analysis does not require workers at the moment";
            }
            if($stats->status eq 'DONE') {
                die "Analysis is DONE, don't need to create workers";
169 170
            }
        }
171

172
    } else {    # probably scheduled by beekeeper.pl
173

174 175 176
        if($rc_name) {
            my $rc = $self->db->get_ResourceClassAdaptor->fetch_by_name($rc_name)
                or die "resource_class with name='$rc_name' could not be fetched from the database";
177

178 179
            $rc_id = $rc->dbID;
        }
180

181 182
        $stats = $self->specialize_new_worker($rc_id)
            or die "Scheduler failed to pick an analysis for the worker";
183

184 185
        print "Scheduler picked analysis with dbID=".$stats->analysis_id." for the worker\n";
    }
186

187
    my $sql = q{INSERT INTO worker (born, last_check_in, meadow_type, meadow_name, process_id, host, analysis_id)
188
              VALUES (CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, ?,?,?,?,?)};
Jessica Severin's avatar
Jessica Severin committed
189

190
    my $sth = $self->prepare($sql);
191 192 193
    $sth->execute($meadow_type, $meadow_name, $process_id, $exec_host, $stats->analysis_id);
    my $worker_id = $self->dbc->db_handle->last_insert_id(undef, undef, 'worker', 'worker_id')
        or die "Could not store worker";
194
    $sth->finish;
Jessica Severin's avatar
Jessica Severin committed
195

196 197 198 199 200 201 202 203 204 205 206 207 208 209
    unless($job) {  # count it as autonomous worker sharing the load of that analysis:
        $analysis_stats_adaptor->decrease_required_workers($stats->analysis_id);
    }

        # 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() ) {
        $analysis_stats_adaptor->increase_running_workers($stats->analysis_id);
    }

210 211 212
    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;
Jessica Severin's avatar
Jessica Severin committed
213

214 215
            # Note: the following die-message will not reach the log files for circular reason!
        system("mkdir -p $worker_log_dir") && die "Could not create '$worker_log_dir' because: $!";
216

217 218 219 220 221 222
        my $sql_add_log = "UPDATE worker SET log_dir=? WHERE worker_id=?";
        my $sth_add_log = $self->prepare($sql_add_log);
        $sth_add_log->execute($worker_log_dir, $worker_id);
        $sth_add_log->finish;
    }

223 224
    my $worker = $self->fetch_by_dbID($worker_id)
        or die "Could not fetch worker with dbID=$worker_id";
225 226 227 228 229

    $worker->init;

    $worker->_specific_job($job) if($job);
    $worker->execute_writes(0) if($no_write);
230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246

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


    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);
    }
247 248 249
    if(defined $compile_module_once) {
        $worker->compile_module_once($compile_module_once);
    }
250

251
    $stats->update_status('WORKING');
252

253
    return $worker;
Jessica Severin's avatar
Jessica Severin committed
254 255
}

256

Jessica Severin's avatar
Jessica Severin committed
257 258 259 260
sub register_worker_death {
  my ($self, $worker) = @_;

  return unless($worker);
Leo Gordon's avatar
Leo Gordon committed
261

262
  my $cod = $worker->cause_of_death() || 'UNKNOWN';    # make sure we do not attempt to insert a void
Leo Gordon's avatar
bugfix  
Leo Gordon committed
263

264
  my $sql = "UPDATE worker SET died=CURRENT_TIMESTAMP, last_check_in=CURRENT_TIMESTAMP";
265
  $sql .= " ,status='DEAD'";
Jessica Severin's avatar
Jessica Severin committed
266
  $sql .= " ,work_done='" . $worker->work_done . "'";
Leo Gordon's avatar
bugfix  
Leo Gordon committed
267
  $sql .= " ,cause_of_death='$cod'";
268
  $sql .= " WHERE worker_id='" . $worker->dbID ."'";
Jessica Severin's avatar
Jessica Severin committed
269

Leo Gordon's avatar
bugfix  
Leo Gordon committed
270
  $self->dbc->do( $sql );
271

272 273 274 275
  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
276
  if($cod eq 'NO_WORK') {
277
    $self->db->get_AnalysisStatsAdaptor->update_status($worker->analysis->dbID, 'ALL_CLAIMED');
278
  }
279
  if($cod eq 'UNKNOWN'
Leo Gordon's avatar
bugfix  
Leo Gordon committed
280 281 282
  or $cod eq 'MEMLIMIT'
  or $cod eq 'RUNLIMIT'
  or $cod eq 'KILLED_BY_USER') {
Leo Gordon's avatar
Leo Gordon committed
283
    $self->db->get_AnalysisJobAdaptor->release_undone_jobs_from_worker($worker);
284
  }
285
  
286
  # re-sync the analysis_stats when a worker dies as part of dynamic sync system
287
  if($self->safe_synchronize_AnalysisStats($worker->analysis->stats)->status ne 'DONE') {
288 289
    # 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
290
    $self->db->get_AnalysisStatsAdaptor->increase_required_workers($worker->analysis->dbID);
291 292
  }

Jessica Severin's avatar
Jessica Severin committed
293 294
}

295

296
sub check_for_dead_workers {    # scans the whole Valley for lost Workers (but ignores unreachagle ones)
297
    my ($self, $valley, $check_buried_in_haste) = @_;
Leo Gordon's avatar
Leo Gordon committed
298

299
    warn "GarbageCollector:\tChecking for lost Workers...\n";
300

301 302 303 304
    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
305

306
    warn "GarbageCollector:\t[Queen:] we have ".scalar(@$queen_worker_list)." Workers alive.\n";
307

Leo Gordon's avatar
Leo Gordon committed
308 309
    foreach my $worker (@$queen_worker_list) {

310 311 312
        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
313
        } else {
314
            $worker_status_counts{$meadow_type}{'UNREACHABLE'}++;
315

316
            next;   # Worker is unreachable from this Valley
Leo Gordon's avatar
Leo Gordon committed
317 318
        }

319 320 321 322 323
        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'}++;
324

325
            $mt_and_pid_to_lost_worker{$meadow_type}{$process_id} = $worker;
326
        }
327 328 329 330 331 332
    }

        # 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";
    }
333

334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350
    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) {
351
                $worker->cause_of_death( $wpid_to_cod->{$process_id} || 'UNKNOWN');
352 353
                $self->register_worker_death($worker);
            }
354 355 356
        }
    }

357
        # the following bit is completely Meadow-agnostic and only restores database integrity:
Leo Gordon's avatar
Leo Gordon committed
358
    if($check_buried_in_haste) {
359
        warn "GarbageCollector:\tChecking for Workers buried in haste...\n";
360
        my $buried_in_haste_list = $self->fetch_all_dead_workers_with_jobs();
Leo Gordon's avatar
Leo Gordon committed
361
        if(my $bih_number = scalar(@$buried_in_haste_list)) {
362
            warn "GarbageCollector:\tfound $bih_number jobs, reclaiming.\n\n";
Leo Gordon's avatar
Leo Gordon committed
363 364 365
            if($bih_number) {
                my $job_adaptor = $self->db->get_AnalysisJobAdaptor();
                foreach my $worker (@$buried_in_haste_list) {
Leo Gordon's avatar
Leo Gordon committed
366
                    $job_adaptor->release_undone_jobs_from_worker($worker);
Leo Gordon's avatar
Leo Gordon committed
367 368 369
                }
            }
        } else {
370
            warn "GarbageCollector:\tfound none\n";
Leo Gordon's avatar
Leo Gordon committed
371 372 373
        }
    }
}
Jessica Severin's avatar
Jessica Severin committed
374 375


376 377 378
    # 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
379

380
    $self->dbc->do("UPDATE worker SET last_check_in=CURRENT_TIMESTAMP, status='".$worker->status."', work_done='".$worker->work_done."' WHERE worker_id='".$worker->dbID."'");
Jessica Severin's avatar
Jessica Severin committed
381 382 383
}


384
=head2 reset_job_by_dbID_and_sync
385

386
  Arg [1]: int $job_id
387
  Example: 
388
    my $job = $queen->reset_job_by_dbID_and_sync($job_id);
389
  Description: 
390
    For the specified job_id it will fetch just that job, 
391 392 393
    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 
394
    same job simultaneously (use only for debugging).
395
  Returntype : none
396
  Exceptions :
397
  Caller     : beekeeper.pl
398 399 400

=cut

401 402
sub reset_job_by_dbID_and_sync {
    my ($self, $job_id) = @_;
403

404 405 406 407 408 409
    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);
410 411 412
}


413 414 415 416 417 418
######################################
#
# Public API interface for beekeeper
#
######################################

419

420 421 422
    # Note: asking for Queen->fetch_overdue_workers(0) essentially means
    #       "fetch all workers known to the Queen not to be officially dead"
    #
423
sub fetch_overdue_workers {
424
    my ($self,$overdue_secs) = @_;
425

426
    $overdue_secs = 3600 unless(defined($overdue_secs));
427

428
    my $constraint = "status!='DEAD' AND ".
429
                    ( ($self->dbc->driver eq 'sqlite')
430 431 432
                        ? "(strftime('%s','now')-strftime('%s',last_check_in))>$overdue_secs"
                        : "(UNIX_TIMESTAMP()-UNIX_TIMESTAMP(last_check_in))>$overdue_secs");
    return $self->fetch_all( $constraint );
433 434
}

435 436 437

sub fetch_all_dead_workers_with_jobs {
    my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
438

439
    return $self->fetch_all( "JOIN job j USING(worker_id) WHERE j.status NOT IN ('DONE', 'READY', 'FAILED', 'PASSED_ON') GROUP BY worker_id" );
Leo Gordon's avatar
Leo Gordon committed
440
}
Jessica Severin's avatar
Jessica Severin committed
441

442

443 444
=head2 synchronize_hive

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

=cut

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

460
  my $start_time = time();
461

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

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

Leo Gordon's avatar
Leo Gordon committed
471
  print STDERR ''.((time() - $start_time))." seconds to synchronize_hive\n\n";
472
}
473

474

475 476 477
=head2 safe_synchronize_AnalysisStats

  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
478
  Example    : $self->safe_synchronize_AnalysisStats($stats);
479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495
  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->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
496
                   ($stats->seconds_since_last_update < 3*60));
497 498 499 500

  # 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=" . $stats->analysis_id;
501
  #print("$sql\n");
502
  my $row_count = $self->dbc->do($sql);  
503
  return $stats unless($row_count == 1);        # return the un-updated status if locked
504
  #printf("got sync_lock on analysis_stats(%d)\n", $stats->analysis_id);
505
  
506
      # since we managed to obtain the lock, let's go and perform the sync:
507
  $self->synchronize_AnalysisStats($stats);
508

509 510 511 512
  return $stats;
}


513
=head2 synchronize_AnalysisStats
514

515 516
  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
  Example    : $self->synchronize($analysisStats);
517
  Description: Queries the job and worker tables to get summary counts
518 519 520 521 522
               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
523

524
=cut
525

526
sub synchronize_AnalysisStats {
527
  my $self = shift;
528
  my $analysisStats = shift;
529

530
  return $analysisStats unless($analysisStats);
531
  return $analysisStats unless($analysisStats->analysis_id);
532

533
  $analysisStats->refresh(); ## Need to get the new hive_capacity for dynamic analyses
534 535
  my $hive_capacity = $analysisStats->hive_capacity;

536 537
  if($self->db->hive_use_triggers()) {

538 539
            my $job_count = $analysisStats->ready_job_count();
            my $required_workers = $hive_capacity && POSIX::ceil( $job_count / $analysisStats->get_or_estimate_batch_size() );
540

541 542
                # adjust_stats_for_living_workers:
            if($hive_capacity > 0) {
543
                my $unfulfilled_capacity = $hive_capacity - $analysisStats->num_running_workers();
544

545 546
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
547 548 549 550 551 552
                }
            }
            $analysisStats->num_required_workers( $required_workers );

  } else {
      $analysisStats->total_job_count(0);
553 554
      $analysisStats->semaphored_job_count(0);
      $analysisStats->ready_job_count(0);
555 556 557
      $analysisStats->done_job_count(0);
      $analysisStats->failed_job_count(0);
      $analysisStats->num_required_workers(0);
558

559
            # ask for analysis_id to force MySQL to use existing index on (analysis_id, status)
560
      my $sql = "SELECT analysis_id, status, count(*) FROM job WHERE analysis_id=? GROUP BY analysis_id, status";
561 562
      my $sth = $self->prepare($sql);
      $sth->execute($analysisStats->analysis_id);
563

564 565 566
      my $done_here       = 0;
      my $done_elsewhere  = 0;
      my $total_job_count = 0;
567
      while (my ($dummy_analysis_id, $status, $job_count)=$sth->fetchrow_array()) {
568
    # print STDERR "$status: $job_count\n";
569

570
        $total_job_count += $job_count;
571

572
        if($status eq 'READY') {
573
            $analysisStats->ready_job_count($job_count);
574

575
            my $required_workers = $hive_capacity && POSIX::ceil( $job_count / $analysisStats->get_or_estimate_batch_size() );
576 577 578

                # adjust_stats_for_living_workers:
            if($hive_capacity > 0) {
579
                my $unfulfilled_capacity = $hive_capacity - $self->count_running_workers( $analysisStats->analysis_id() );
580

581 582
                if($unfulfilled_capacity < $required_workers ) {
                    $required_workers = (0 < $unfulfilled_capacity) ? $unfulfilled_capacity : 0;
583
                }
584
            }
585 586
            $analysisStats->num_required_workers( $required_workers );

587 588
        } elsif($status eq 'SEMAPHORED') {
            $analysisStats->semaphored_job_count($job_count);
589
        } elsif($status eq 'DONE') {
590
            $done_here = $job_count;
591
        } elsif($status eq 'PASSED_ON') {
592 593 594
            $done_elsewhere = $job_count;
        } elsif ($status eq 'FAILED') {
            $analysisStats->failed_job_count($job_count);
595
        }
596
      } # /while
597
      $sth->finish;
598

599 600
      $analysisStats->total_job_count( $total_job_count );
      $analysisStats->done_job_count( $done_here + $done_elsewhere );
601
  } # /unless $self->{'_hive_use_triggers'}
602

603
  $analysisStats->check_blocking_control_rules();
604

605 606 607
  if($analysisStats->status ne 'BLOCKED') {
    $analysisStats->determine_status();
  }
608

609
  # $analysisStats->sync_lock(0); ## do we perhaps need it here?
610
  $analysisStats->update;  #update and release sync_lock
611

612
  return $analysisStats;
613
}
614 615


616 617 618 619 620 621 622 623 624 625 626 627 628 629
=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
630
sub get_num_failed_analyses {
631
    my ($self, $filter_analysis) = @_;
632

633 634 635 636 637 638 639 640 641 642 643 644
    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;
        }
    }
645

646
    return $filter_analysis ? $filter_analysis_failed : scalar(@$failed_analyses);
647 648 649
}


650
sub get_hive_current_load {
651 652 653
    my $self = shift;
    my $sql = qq{
        SELECT sum(1/hive_capacity)
654 655 656 657
        FROM worker w
        JOIN analysis_stats USING(analysis_id)
        WHERE w.status!='DEAD'
        AND hive_capacity>0
658 659 660 661 662 663
    };
    my $sth = $self->prepare($sql);
    $sth->execute();
    my ($load)=$sth->fetchrow_array();
    $sth->finish;
    return ($load || 0);
664 665 666
}


667 668 669
sub count_running_workers {
    my ($self, $analysis_id) = @_;

670 671 672 673 674
    my $sql = qq{
            SELECT count(*)
            FROM worker
            WHERE status!='DEAD'
        } . ($analysis_id ? " AND analysis_id='$analysis_id'" : '');
675 676 677 678 679 680 681

    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::Hive::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
sub schedule_workers {
699
  my ($self, $filter_analysis, $orig_pending_by_rc_name, $available_submit_limit) = @_;
700

701 702 703
  my @suitable_analyses         = $filter_analysis
                                    ? ( $filter_analysis->stats )
                                    : @{ $self->db->get_AnalysisStatsAdaptor->fetch_all_by_suitability_rc_id() };
704

705 706 707 708
  my $analysis_id2rc_id         = $self->db->get_AnalysisAdaptor->fetch_HASHED_FROM_analysis_id_TO_resource_class_id();
  my $rc_id2name                = $self->db->get_ResourceClassAdaptor->fetch_HASHED_FROM_resource_class_id_TO_name();
        # combined mapping
  my %analysis_id2rc_name       = map { $_ => $rc_id2name->{ $analysis_id2rc_id->{ $_ }} } keys %$analysis_id2rc_id;
709

710
  return {} unless(@suitable_analyses);
711

712
  my %pending_by_rc_name        = %{ $orig_pending_by_rc_name || {} };  # NB: make our own copy to be able to modify it
713
  my $total_workers_to_run      = 0;
714
  my %workers_to_run_by_rc_name = ();
715
  my $available_load            = 1.0 - $self->get_hive_current_load();
716

717

718
  foreach my $analysis_stats (@suitable_analyses) {
719 720
    last if ($available_load <= 0.0);
    last if (defined($available_submit_limit) and !$available_submit_limit);
721

Leo Gordon's avatar
Leo Gordon committed
722
        #digging deeper under the surface so need to sync
723
    if(($analysis_stats->status eq 'LOADING') or ($analysis_stats->status eq 'BLOCKED') or ($analysis_stats->status eq 'ALL_CLAIMED')) {
724 725 726 727 728 729
      $self->synchronize_AnalysisStats($analysis_stats);
    }

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

730
        # 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
731
    my $workers_this_analysis = $analysis_stats->num_required_workers;
732

733 734 735 736 737 738
    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
739

740 741 742 743 744
    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
745 746
        }

747
        $available_load -= 1.0*$workers_this_analysis/$hive_capacity;
748
    }
Leo Gordon's avatar
Leo Gordon committed
749

750
    my $curr_rc_name    = $analysis_id2rc_name{ $analysis_stats->analysis_id };
751

752 753 754 755 756 757
    if($pending_by_rc_name{ $curr_rc_name }) {                              # per-rc_name capping by pending processes, if available
        my $pending_this_analysis = ($pending_by_rc_name{ $curr_rc_name } < $workers_this_analysis) ? $pending_by_rc_name{ $curr_rc_name } : $workers_this_analysis;

        print "Scheduler detected $pending_this_analysis pending workers with resource_class_name=$curr_rc_name\n";
        $workers_this_analysis                  -= $pending_this_analysis;
        $pending_by_rc_name{ $curr_rc_name }    -= $pending_this_analysis;
758 759
    }

760 761
    next unless($workers_this_analysis);    # do not autovivify the hash by a zero

762
    $total_workers_to_run += $workers_this_analysis;
763
    $workers_to_run_by_rc_name{ $curr_rc_name } += $workers_this_analysis;
764
    $analysis_stats->print_stats();
765
    printf("Scheduler suggests adding %d more workers of resource_class_name=%s for analysis_id=%d [%1.3f hive_load remaining]\n", $workers_this_analysis, $curr_rc_name, $analysis_stats->analysis_id, $available_load);
766
  }
767

768
  printf("Scheduler suggests adding a total of %d workers [%1.5f hive_load remaining]\n", $total_workers_to_run, $available_load);
769
  return \%workers_to_run_by_rc_name;
Leo Gordon's avatar
Leo Gordon committed
770 771
}

772

773
sub schedule_workers_resync_if_necessary {
774 775 776
    my ($self, $valley, $analysis) = @_;

    my $meadow              = $valley->get_current_meadow();
Leo Gordon's avatar
Leo Gordon committed
777

778
    my $pending_by_rc_name  = ($meadow->can('count_pending_workers_by_rc_name') and $meadow->config_get('PendingAdjust')) ? $meadow->count_pending_workers_by_rc_name() : {};
779 780
    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
781

782 783 784 785
    my $available_submit_limit = ($submit_limit and $meadow_limit)
                                    ? (($submit_limit<$meadow_limit) ? $submit_limit : $meadow_limit)
                                    : (defined($submit_limit) ? $submit_limit : $meadow_limit);

786
    my $workers_to_run_by_rc_name = $self->schedule_workers($analysis, $pending_by_rc_name, $available_submit_limit);
787

788
    unless( keys %$workers_to_run_by_rc_name or $self->get_hive_current_load() or $self->count_running_workers() ) {
789
        print "\nScheduler: nothing is running and nothing to do (according to analysis_stats) => executing garbage collection and sync\n" ;
Leo Gordon's avatar
Leo Gordon committed
790

791
        $self->check_for_dead_workers($valley, 1);
792
        $self->synchronize_hive($analysis);
Leo Gordon's avatar
Leo Gordon committed
793

794
        $workers_to_run_by_rc_name = $self->schedule_workers($analysis, $pending_by_rc_name, $available_submit_limit);
Leo Gordon's avatar
Leo Gordon committed