Queen.pm 32.4 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
  gets $worker->batch_size() jobs from the analysis_job table, does its
29
  work, creates the next layer of analysis_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;
Leo Gordon's avatar
Leo Gordon committed
64
use Sys::Hostname;
65 66 67
use Bio::EnsEMBL::Utils::Argument;
use Bio::EnsEMBL::Utils::Exception;

Jessica Severin's avatar
Jessica Severin committed
68
use Bio::EnsEMBL::Hive::Worker;
69
use Bio::EnsEMBL::Hive::DBSQL::AnalysisCtrlRuleAdaptor;
Jessica Severin's avatar
Jessica Severin committed
70

Leo Gordon's avatar
Leo Gordon committed
71
use base ('Bio::EnsEMBL::DBSQL::BaseAdaptor');
72

Jessica Severin's avatar
Jessica Severin committed
73

74
############################
Jessica Severin's avatar
Jessica Severin committed
75
#
76
# PUBLIC API for Workers
Jessica Severin's avatar
Jessica Severin committed
77
#
78
############################
Jessica Severin's avatar
Jessica Severin committed
79 80 81

=head2 create_new_worker

82
  Arg [1]    : $analysis_id (optional)
Jessica Severin's avatar
Jessica Severin committed
83
  Example    :
84 85 86 87 88 89
  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.
               It creates an entry in the hive table, and returns a Worker object 
               based on that insert.  This guarantees that each worker registered
               in this queens hive is properly registered.
Jessica Severin's avatar
Jessica Severin committed
90 91 92 93 94 95 96
  Returntype : Bio::EnsEMBL::Hive::Worker
  Exceptions :
  Caller     :

=cut

sub create_new_worker {
97 98
  my ($self, @args) = @_;

Leo Gordon's avatar
Leo Gordon committed
99 100
  my ($rc_id, $analysis_id, $beekeeper ,$pid, $job, $no_write) =
     rearrange([qw(rc_id analysis_id beekeeper process_id job no_write) ], @args);
Jessica Severin's avatar
Jessica Severin committed
101

102
  my $analStatsDBA = $self->db->get_AnalysisStatsAdaptor;
103
  return undef unless($analStatsDBA);
104

105
  $analysis_id = $job->analysis_id if(defined($job));
106
  
107 108 109
  my $analysisStats;
  if($analysis_id) {
    $analysisStats = $analStatsDBA->fetch_by_analysis_id($analysis_id);
110
    $self->safe_synchronize_AnalysisStats($analysisStats);
111
    #return undef unless(($analysisStats->status ne 'BLOCKED') and ($analysisStats->num_required_workers > 0));
112
  } else {
Leo Gordon's avatar
Leo Gordon committed
113
    $analysisStats = $self->_pick_best_analysis_for_new_worker($rc_id);
114
  }
115
  return undef unless($analysisStats);
116

117 118
  unless($job) {
    #go into autonomous mode
119
    return undef if($self->get_hive_current_load() >= 1.1);
120
    
121 122
    $analStatsDBA->decrease_needed_workers($analysisStats->analysis_id);
    $analStatsDBA->increase_running_workers($analysisStats->analysis_id);
123 124 125 126 127 128 129 130 131 132
    $analysisStats->print_stats;
    
    if($analysisStats->status eq 'BLOCKED') {
      print("Analysis is BLOCKED, can't create workers\n");
      return undef;
    }
    if($analysisStats->status eq 'DONE') {
      print("Analysis is DONE, don't need to create workers\n");
      return undef;
    }
133
  }
134
  
Jessica Severin's avatar
Jessica Severin committed
135
  my $host = hostname;
136
  $pid = $$ unless($pid);
137
  $beekeeper = '' unless($beekeeper);
Jessica Severin's avatar
Jessica Severin committed
138

139 140 141
  my $sql = q{INSERT INTO hive 
              (born, last_check_in, process_id, analysis_id, beekeeper, host)
              VALUES (NOW(), NOW(), ?,?,?,?)};
Jessica Severin's avatar
Jessica Severin committed
142 143

  my $sth = $self->prepare($sql);
144
  $sth->execute($pid, $analysisStats->analysis_id, $beekeeper, $host);
Leo Gordon's avatar
Leo Gordon committed
145
  my $worker_id = $sth->{'mysql_insertid'};
Jessica Severin's avatar
Jessica Severin committed
146 147
  $sth->finish;

Leo Gordon's avatar
Leo Gordon committed
148
  my $worker = $self->fetch_by_worker_id($worker_id);
Jessica Severin's avatar
Jessica Severin committed
149 150
  $worker=undef unless($worker and $worker->analysis);

151
  if($worker and $analysisStats) {
152
    $analysisStats->update_status('WORKING');
Jessica Severin's avatar
Jessica Severin committed
153
  }
154 155 156 157
  
  $worker->_specific_job($job) if(defined($job));
  $worker->execute_writes(0) if($no_write);
  
Jessica Severin's avatar
Jessica Severin committed
158 159 160 161 162 163 164
  return $worker;
}

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

  return unless($worker);
Leo Gordon's avatar
Leo Gordon committed
165

Leo Gordon's avatar
bugfix  
Leo Gordon committed
166 167 168
  my $cod = $worker->cause_of_death();

  unless ($cod eq 'HIVE_OVERLOAD') {
169 170 171 172
    ## HIVE_OVERLOAD occurs after a successful update of the analysis_stats teble. (c.f. Worker.pm)
    $worker->analysis->stats->adaptor->decrease_running_workers($worker->analysis->stats->analysis_id);
  }

Jessica Severin's avatar
Jessica Severin committed
173
  my $sql = "UPDATE hive SET died=now(), last_check_in=now()";
174
  $sql .= " ,status='DEAD'";
Jessica Severin's avatar
Jessica Severin committed
175
  $sql .= " ,work_done='" . $worker->work_done . "'";
Leo Gordon's avatar
bugfix  
Leo Gordon committed
176
  $sql .= " ,cause_of_death='$cod'";
Leo Gordon's avatar
Leo Gordon committed
177
  $sql .= " WHERE worker_id='" . $worker->worker_id ."'";
Jessica Severin's avatar
Jessica Severin committed
178

Leo Gordon's avatar
bugfix  
Leo Gordon committed
179
  $self->dbc->do( $sql );
180

Leo Gordon's avatar
bugfix  
Leo Gordon committed
181
  if($cod eq 'NO_WORK') {
182
    $self->db->get_AnalysisStatsAdaptor->update_status($worker->analysis->dbID, 'ALL_CLAIMED');
183
  }
Leo Gordon's avatar
bugfix  
Leo Gordon committed
184 185 186 187
  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
188
    $self->db->get_AnalysisJobAdaptor->release_undone_jobs_from_worker($worker);
189
  }
190
  
191
  # re-sync the analysis_stats when a worker dies as part of dynamic sync system
192
  if($self->safe_synchronize_AnalysisStats($worker->analysis->stats)->status ne 'DONE') {
193 194
    # 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
195
    $self->db->get_AnalysisStatsAdaptor->increase_needed_workers($worker->analysis->dbID);
196 197
  }

Jessica Severin's avatar
Jessica Severin committed
198 199
}

Leo Gordon's avatar
Leo Gordon committed
200 201 202
sub check_for_dead_workers {
    my ($self, $meadow, $check_buried_in_haste) = @_;

203
    my $worker_status_hash    = $meadow->status_of_all_our_workers();
Leo Gordon's avatar
Leo Gordon committed
204 205 206 207 208
    my %worker_status_summary = ();
    my $queen_worker_list     = $self->fetch_overdue_workers(0);

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

209 210
    my %gc_wpid_to_worker = ();

Leo Gordon's avatar
Leo Gordon committed
211 212 213 214 215 216 217 218
    foreach my $worker (@$queen_worker_list) {
        next unless($meadow->responsible_for_worker($worker));

        my $worker_pid = $worker->process_id();
        if(my $status = $worker_status_hash->{$worker_pid}) { # can be RUN|PEND|xSUSP
            $worker_status_summary{$status}++;
        } else {
            $worker_status_summary{'AWOL'}++;
219

220
            $gc_wpid_to_worker{$worker_pid} = $worker;
Leo Gordon's avatar
Leo Gordon committed
221 222 223 224
        }
    }
    print "\t".join(', ', map { "$_:$worker_status_summary{$_}" } keys %worker_status_summary)."\n\n";

225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241
    if(my $total_lost = scalar(keys %gc_wpid_to_worker)) {
        warn "GarbageCollector: Discovered $total_lost lost workers\n";

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

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

Leo Gordon's avatar
Leo Gordon committed
242 243 244 245 246 247 248 249
    if($check_buried_in_haste) {
        print "====== Checking for workers buried in haste... ";
        my $buried_in_haste_list = $self->fetch_dead_workers_with_jobs();
        if(my $bih_number = scalar(@$buried_in_haste_list)) {
            print "$bih_number, reclaiming jobs.\n\n";
            if($bih_number) {
                my $job_adaptor = $self->db->get_AnalysisJobAdaptor();
                foreach my $worker (@$buried_in_haste_list) {
Leo Gordon's avatar
Leo Gordon committed
250
                    $job_adaptor->release_undone_jobs_from_worker($worker);
Leo Gordon's avatar
Leo Gordon committed
251 252 253 254 255 256 257
                }
            }
        } else {
            print "none\n";
        }
    }
}
Jessica Severin's avatar
Jessica Severin committed
258

259
sub worker_check_in {
Jessica Severin's avatar
Jessica Severin committed
260 261 262 263
  my ($self, $worker) = @_;

  return unless($worker);
  my $sql = "UPDATE hive SET last_check_in=now()";
264
  $sql .= " ,work_done='" . $worker->work_done . "'";
Leo Gordon's avatar
Leo Gordon committed
265
  $sql .= " WHERE worker_id='" . $worker->worker_id ."'";
Jessica Severin's avatar
Jessica Severin committed
266 267 268 269

  my $sth = $self->prepare($sql);
  $sth->execute();
  $sth->finish;
270
  
271
  $self->safe_synchronize_AnalysisStats($worker->analysis->stats);
Jessica Severin's avatar
Jessica Severin committed
272 273 274
}


275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300
=head2 worker_grab_jobs

  Arg [1]           : Bio::EnsEMBL::Hive::Worker object $worker
  Example: 
    my $jobs  = $queen->worker_grab_jobs();
  Description: 
    For the specified worker, it will search available jobs, 
    and using the workers requested batch_size, claim/fetch that
    number of jobs, and then return them.
  Returntype : 
    reference to array of Bio::EnsEMBL::Hive::AnalysisJob objects
  Exceptions :
  Caller     :

=cut

sub worker_grab_jobs {
  my $self            = shift;
  my $worker          = shift;
  
  my $jobDBA = $self->db->get_AnalysisJobAdaptor;
  my $claim = $jobDBA->claim_jobs_for_worker($worker);
  my $jobs = $jobDBA->fetch_by_claim_analysis($claim, $worker->analysis->dbID);
  return $jobs;
}

301
=head2 reset_and_fetch_job_by_dbID
302 303 304

  Arg [1]: int $analysis_job_id
  Example: 
305
    my $job = $queen->reset_and_fetch_job_by_dbID($analysis_job_id);
306 307
  Description: 
    For the specified analysis_job_id it will fetch just that job, 
308 309 310
    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 
311 312
    same job simultaneously (use only for debugging).
  Returntype : 
313
    Bio::EnsEMBL::Hive::AnalysisJob object
314
  Exceptions :
315
  Caller     : beekeepers, runWorker.pl scripts
316 317 318

=cut

319 320
sub reset_and_fetch_job_by_dbID {
  my $self = shift;
321
  my $analysis_job_id = shift;
322
  
323
  my $jobDBA = $self->db->get_AnalysisJobAdaptor;
324 325 326 327 328
  $jobDBA->reset_job_by_dbID($analysis_job_id); 

  my $job = $jobDBA->fetch_by_dbID($analysis_job_id); 
  my $stats = $self->db->get_AnalysisStatsAdaptor->fetch_by_analysis_id($job->analysis_id);
  $self->synchronize_AnalysisStats($stats);
329
  
330 331 332 333 334 335 336 337 338 339
  return $job;
}


sub worker_reclaim_job {
  my $self   = shift;
  my $worker = shift;
  my $job    = shift;

  return undef unless($job and $worker);
Leo Gordon's avatar
Leo Gordon committed
340
  $job->worker_id($worker->worker_id);
341
  $self->db->get_AnalysisJobAdaptor->reclaim_job($job);
342 343 344 345 346 347 348 349 350 351
  return $job;
}


######################################
#
# Public API interface for beekeeper
#
######################################

352 353 354
    # Note: asking for Queen->fetch_overdue_workers(0) essentially means
    #       "fetch all workers known to the Queen not to be officially dead"
    #
355 356 357 358 359 360 361 362 363 364
sub fetch_overdue_workers {
  my ($self,$overdue_secs) = @_;

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

  my $constraint = "h.cause_of_death='' ".
                   "AND (UNIX_TIMESTAMP()-UNIX_TIMESTAMP(h.last_check_in))>$overdue_secs";
  return $self->_generic_fetch($constraint);
}

365 366
sub fetch_failed_workers {
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
367

368 369 370 371
  my $constraint = "h.cause_of_death='FATALITY' ";
  return $self->_generic_fetch($constraint);
}

Leo Gordon's avatar
Leo Gordon committed
372 373 374
sub fetch_dead_workers_with_jobs {
  my $self = shift;

Leo Gordon's avatar
Leo Gordon committed
375
  # select h.worker_id from hive h, analysis_job WHERE h.worker_id=analysis_job.worker_id AND h.cause_of_death!='' AND analysis_job.status not in ('DONE', 'READY','FAILED') group by h.worker_id
Leo Gordon's avatar
Leo Gordon committed
376

Leo Gordon's avatar
bugfix  
Leo Gordon committed
377
  my $constraint = "h.cause_of_death!='' ";
Leo Gordon's avatar
Leo Gordon committed
378
  my $join = [[['analysis_job', 'j'], " h.worker_id=j.worker_id AND j.status NOT IN ('DONE', 'READY', 'FAILED') GROUP BY h.worker_id"]];
Leo Gordon's avatar
Leo Gordon committed
379 380
  return $self->_generic_fetch($constraint, $join);
}
Jessica Severin's avatar
Jessica Severin committed
381

382 383
=head2 synchronize_hive

Leo Gordon's avatar
Leo Gordon committed
384
  Arg [1]    : $filter_analysis (optional)
385 386 387 388 389 390 391 392 393 394 395
  Example    : $queen->synchronize_hive();
  Description: Runs through all analyses in the system and synchronizes
              the analysis_stats summary with the states in the analysis_job 
              and hive tables.  Then follows by checking all the blocking rules
              and blocks/unblocks analyses as needed.
  Exceptions : none
  Caller     : general

=cut

sub synchronize_hive {
Leo Gordon's avatar
Leo Gordon committed
396
  my $self          = shift;
Leo Gordon's avatar
Leo Gordon committed
397
  my $filter_analysis = shift; # optional parameter
398

399
  my $start_time = time();
400

Leo Gordon's avatar
Leo Gordon committed
401
  my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
402

Leo Gordon's avatar
Leo Gordon committed
403
  print STDERR "\nSynchronizing the hive (".scalar(@$list_of_analyses)." analyses this time):\n";
Leo Gordon's avatar
Leo Gordon committed
404 405
  foreach my $analysis (@$list_of_analyses) {
    $self->synchronize_AnalysisStats($analysis->stats);
Leo Gordon's avatar
Leo Gordon committed
406
    print STDERR '.';
407
  }
Leo Gordon's avatar
Leo Gordon committed
408
  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
409

Leo Gordon's avatar
Leo Gordon committed
410
  print STDERR "Checking blocking control rules:\n";
Leo Gordon's avatar
Leo Gordon committed
411
  foreach my $analysis (@$list_of_analyses) {
412
    $self->check_blocking_control_rules_for_AnalysisStats($analysis->stats);
Leo Gordon's avatar
Leo Gordon committed
413
    print STDERR '.';
414
  }
Leo Gordon's avatar
Leo Gordon committed
415
  print STDERR "\n";
Leo Gordon's avatar
Leo Gordon committed
416

Leo Gordon's avatar
Leo Gordon committed
417
  print STDERR ''.((time() - $start_time))." seconds to synchronize_hive\n\n";
418
}
419

420

421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442
=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
443
                   ($stats->seconds_since_last_update < 3*60));
444 445 446 447

  # 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=" . $stats->analysis_id;
448
  #print("$sql\n");
449 450
  my $row_count = $self->dbc->do($sql);  
  return $stats unless($row_count == 1);
451
  #printf("got sync_lock on analysis_stats(%d)\n", $stats->analysis_id);
452 453 454
  
  #OK have the lock, go and do the sync
  $self->synchronize_AnalysisStats($stats);
455

456 457 458 459
  return $stats;
}


460
=head2 synchronize_AnalysisStats
461

462 463 464 465 466 467 468 469
  Arg [1]    : Bio::EnsEMBL::Hive::AnalysisStats object
  Example    : $self->synchronize($analysisStats);
  Description: Queries the analysis_job and hive tables to get summary counts
               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
470

471
=cut
472

473
sub synchronize_AnalysisStats {
474
  my $self = shift;
475
  my $analysisStats = shift;
476

477
  return $analysisStats unless($analysisStats);
478
  return $analysisStats unless($analysisStats->analysis_id);
479
  
480
  $analysisStats->refresh(); ## Need to get the new hive_capacity for dynamic analyses
481 482 483 484 485 486
  $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);

487 488
  my $sql = "SELECT status, count(*), semaphore_count FROM analysis_job ".
            "WHERE analysis_id=? GROUP BY status, semaphore_count";
489 490 491
  my $sth = $self->prepare($sql);
  $sth->execute($analysisStats->analysis_id);

492 493
  my $hive_capacity = $analysisStats->hive_capacity;

494
  while (my ($status, $count, $semaphore_count)=$sth->fetchrow_array()) {
495
# print STDERR "$status - $count\n";
496 497 498 499

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

500
    if(($status eq 'READY') and ($semaphore_count<=0)) {
501
      $analysisStats->unclaimed_job_count($count);
502 503 504 505 506 507 508 509 510
      my $numWorkers;
      if($analysisStats->batch_size > 0) {
        $numWorkers = POSIX::ceil($count / $analysisStats->batch_size);
      } else {
        my $job_msec = $analysisStats->avg_msec_per_job;
        $job_msec = 100 if($job_msec>0 and $job_msec<100);
        $numWorkers = POSIX::ceil(($count * $job_msec) / (3*60*1000)); 
        # guess num needed workers by total jobs / (num jobs a worker could do in 3 minutes)
      }
511
      $numWorkers=$count if($numWorkers==0);
512 513 514 515 516
      if($analysisStats->hive_capacity>0 and $numWorkers > $analysisStats->hive_capacity) {
        $numWorkers=$analysisStats->hive_capacity;
      }
      $analysisStats->num_required_workers($numWorkers);
    }
517 518
    if ($status eq 'DONE') { $analysisStats->done_job_count($count); }
    if ($status eq 'FAILED') { $analysisStats->failed_job_count($count); }
519 520
  }
  $sth->finish;
521

522
  $self->check_blocking_control_rules_for_AnalysisStats($analysisStats);
523

524 525 526
  if($analysisStats->status ne 'BLOCKED') {
    $analysisStats->determine_status();
  }
527

528 529 530 531
  #
  # adjust_stats_for_living_workers
  #
  
532
  if($analysisStats->hive_capacity > 0) {
533
    my $liveCount = $analysisStats->get_running_worker_count();
534 535 536 537 538 539 540 541

    my $numWorkers = $analysisStats->num_required_workers;

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

    $analysisStats->num_required_workers($numWorkers);
542
  }
543

544
  $analysisStats->update;  #update and release sync_lock
545

546
  return $analysisStats;
547
}
548 549


550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572
sub check_blocking_control_rules_for_AnalysisStats
{
  my $self = shift;
  my $stats = shift;
  
  return unless($stats);

  #print("check ctrl on analysis ");  $stats->print_stats;
  my $ctrlRules = $self->db->get_AnalysisCtrlRuleAdaptor->
                  fetch_by_ctrled_analysis_id($stats->analysis_id);
  my $allRulesDone = 1;
  if(scalar @$ctrlRules > 0) {
    #print("HAS blocking_ctrl_rules to check\n");
    foreach my $ctrlrule (@{$ctrlRules}) {
      #use this method because the condition_analysis objects can be
      #network distributed to a different database so use it's adaptor to get
      #the AnalysisStats object
      #$ctrlrule->print_rule;
      my $condAnalysis = $ctrlrule->condition_analysis;
      my $condStats = $condAnalysis->stats if($condAnalysis);
      $allRulesDone = 0 unless($condStats and $condStats->status eq 'DONE');
      #print("  "); $condStats->print_stats;
    }
573

574 575 576
    if($allRulesDone) {
      if($stats->status eq 'BLOCKED') {
        #print("  UNBLOCK analysis : all conditions met\n");
577
        $stats->update_status('LOADING'); #trigger sync
578 579 580 581
      }
    } else {
      #print("  RE-BLOCK analysis : some conditions failed\n");
      $stats->update_status('BLOCKED');
582 583
    }
  }
584 585 586
}


Leo Gordon's avatar
Leo Gordon committed
587 588
sub get_num_failed_analyses {
  my ($self, $analysis) = @_;
589 590

  my $statsDBA = $self->db->get_AnalysisStatsAdaptor;
Leo Gordon's avatar
Leo Gordon committed
591
  my $failed_analyses = $statsDBA->fetch_by_statuses(['FAILED']);
592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613
  if ($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);
}


614 615
sub get_hive_current_load {
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
616
  my $sql = "SELECT sum(1/analysis_stats.hive_capacity) FROM hive, analysis_stats ".
617 618
            "WHERE hive.analysis_id=analysis_stats.analysis_id and cause_of_death ='' ".
            "AND analysis_stats.hive_capacity>0";
619 620 621 622 623
  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
624
  print("current hive load = $load\n");
625 626 627 628
  return $load;
}


629 630 631 632 633 634 635 636
sub get_num_running_workers {
  my $self = shift;
  my $sql = "SELECT count(*) FROM hive WHERE cause_of_death =''";
  my $sth = $self->prepare($sql);
  $sth->execute();
  (my $runningCount)=$sth->fetchrow_array();
  $sth->finish;
  $runningCount=0 unless($runningCount);
Leo Gordon's avatar
Leo Gordon committed
637
  print("current hive num_running_workers = $runningCount\n");
638 639 640
  return $runningCount;
}

641 642 643
sub enter_status {
  my ($self, $worker, $status) = @_;

Leo Gordon's avatar
Leo Gordon committed
644
  $self->dbc->do("UPDATE hive SET status = '$status' WHERE worker_id = ".$worker->worker_id);
645 646
}

647 648
=head2 get_num_needed_workers

649
  Arg[1]     : Bio::EnsEMBL::Analysis object (optional)
650 651 652 653
  Example    : $count = $queen->get_num_needed_workers();
  Description: Runs through the analyses in the system which are waiting
               for workers to be created for them.  Calculates the maximum
               number of workers needed to fill the current needs of the system
654
               If Arg[1] is defined, does it only for the given analysis.
655
  Exceptions : none
656
  Caller     : beekeepers and other external processes
657 658 659

=cut

660
sub get_num_needed_workers {
Leo Gordon's avatar
Leo Gordon committed
661
  my ($self, $filter_analysis) = @_;
662

663
  my $statsDBA = $self->db->get_AnalysisStatsAdaptor;
Leo Gordon's avatar
Leo Gordon committed
664 665 666
  my $clearly_needed_analyses     = $statsDBA->fetch_by_needed_workers(undef,$self->{maximise_concurrency});
  my $potentially_needed_analyses = $statsDBA->fetch_by_statuses(['LOADING', 'BLOCKED']);
  my @all_analyses = (@$clearly_needed_analyses, @$potentially_needed_analyses);
667

Leo Gordon's avatar
Leo Gordon committed
668
  return 0 unless(@all_analyses);
669

Leo Gordon's avatar
Leo Gordon committed
670
  my $available_load = 1.0 - $self->get_hive_current_load();
671

Leo Gordon's avatar
Leo Gordon committed
672
  return 0 if($available_load <=0.0);
673

Leo Gordon's avatar
Leo Gordon committed
674 675
  my $total_workers = 0;
  my %rc2workers = ();
676

677 678
# FIXME: I have a feeling sometimes this method returns an incorrect value. Please doublecheck.

Leo Gordon's avatar
Leo Gordon committed
679 680
  foreach my $analysis_stats (@all_analyses) {
    next if (defined $filter_analysis && $filter_analysis->dbID != $analysis_stats->analysis_id);
681

Leo Gordon's avatar
Leo Gordon committed
682
        #digging deeper under the surface so need to sync
683 684 685 686 687 688 689
    if(($analysis_stats->status eq 'LOADING') or ($analysis_stats->status eq 'BLOCKED')) {
      $self->synchronize_AnalysisStats($analysis_stats);
    }

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

Leo Gordon's avatar
Leo Gordon committed
690
    my $workers_this_analysis = $analysis_stats->num_required_workers;
691

Leo Gordon's avatar
Leo Gordon committed
692 693 694 695 696 697 698 699
    if($analysis_stats->hive_capacity > 0) {   # if there is a limit, use it for cut-off
        my $limit_workers_this_analysis = int($available_load * $analysis_stats->hive_capacity);

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

        $available_load -= 1.0*$workers_this_analysis/$analysis_stats->hive_capacity;
700
    }
Leo Gordon's avatar
Leo Gordon committed
701 702 703 704 705 706
    $total_workers += $workers_this_analysis;
    $rc2workers{$analysis_stats->rc_id} += $workers_this_analysis;
    $analysis_stats->print_stats();
    printf("  (%1.3f remaining-hive-load) use %3d workers of analysis_id=%d\n", $available_load, $workers_this_analysis, $analysis_stats->analysis_id);

    last if($available_load <= 0.0);
707
  }
708

Leo Gordon's avatar
Leo Gordon committed
709 710 711 712
  printf("need a total of $total_workers workers (availLoad=%1.5f)\n", $available_load);
  return ($total_workers, \%rc2workers);
}

Leo Gordon's avatar
Leo Gordon committed
713
sub get_needed_workers_resync_if_necessary {
Leo Gordon's avatar
Leo Gordon committed
714 715 716 717 718 719 720 721 722 723 724 725 726 727 728
    my ($self, $meadow, $analysis) = @_;

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

    if($load==0 and $needed_count==0 and $running_count==0) {
        print "*** nothing is running and nothing to do (according to analysis_stats) => perform a hard resync\n" ;

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

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

Leo Gordon's avatar
Leo Gordon committed
729
    return ($needed_count, $rc_hash);
730 731
}

Leo Gordon's avatar
Leo Gordon committed
732
sub get_remaining_jobs_show_hive_progress {
733
  my $self = shift;
Leo Gordon's avatar
Leo Gordon committed
734
  my $sql = "SELECT sum(done_job_count), sum(failed_job_count), sum(total_job_count), ".
735 736
            "sum(unclaimed_job_count * analysis_stats.avg_msec_per_job)/1000/60/60 ".
            "FROM analysis_stats";
737 738
  my $sth = $self->prepare($sql);
  $sth->execute();
739
  my ($done, $failed, $total, $cpuhrs) = $sth->fetchrow_array();
740
  $sth->finish;
741 742 743 744 745 746 747

  $done   ||= 0;
  $failed ||= 0;
  $total  ||= 0;
  my $completed = $total
    ? ((100.0 * ($done+$failed))/$total)
    : 0.0;
748 749 750
  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
751
  return $remaining;
752 753
}

754 755 756 757 758 759 760 761
## Can't see where this method is used.
#
#sub print_hive_status {
#    my ($self, $filter_analysis) = @_;
#
#    $self->print_analysis_status($filter_analysis);
#    $self->print_running_worker_status;
#}
762 763


Leo Gordon's avatar
Leo Gordon committed
764
sub print_analysis_status {
Leo Gordon's avatar
Leo Gordon committed
765
    my ($self, $filter_analysis) = @_;
766

Leo Gordon's avatar
Leo Gordon committed
767
    my $list_of_analyses = $filter_analysis ? [$filter_analysis] : $self->db->get_AnalysisAdaptor->fetch_all;
Leo Gordon's avatar
Leo Gordon committed
768 769 770
    foreach my $analysis (sort {$a->dbID <=> $b->dbID} @$list_of_analyses) {
        $analysis->stats->print_stats($self->{'verbose_stats'});
    }
771 772 773
}


Leo Gordon's avatar
Leo Gordon committed
774
sub print_running_worker_status {
775
  my $self = shift;
776

Leo Gordon's avatar
Leo Gordon committed
777
  print "====== Live workers according to Queen:\n";
778 779 780
  my $sql = "select logic_name, count(*) from hive, analysis ".
            "where hive.analysis_id=analysis.analysis_id and hive.cause_of_death='' ".
            "group by hive.analysis_id";
Leo Gordon's avatar
Leo Gordon committed
781 782

  my $total = 0;
783 784 785 786
  my $sth = $self->prepare($sql);
  $sth->execute();
  while((my $logic_name, my $count)=$sth->fetchrow_array()) {
    printf("%20s : %d workers\n", $logic_name, $count);
787
    $total += $count;
788
  }
789
  printf("  %d total workers\n", $total);
Leo Gordon's avatar
Leo Gordon committed
790
  print "===========================\n";
791
  $sth->finish;
792
}
793

794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822
=head2 monitor

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

=cut

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

Leo Gordon's avatar
Leo Gordon committed
823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840
=head2 register_all_workers_dead

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

=cut

sub register_all_workers_dead {
    my $self = shift;

    my $overdueWorkers = $self->fetch_overdue_workers(0);
    foreach my $worker (@{$overdueWorkers}) {
        $self->register_worker_death($worker);
    }
}

841

Jessica Severin's avatar
Jessica Severin committed
842 843 844 845 846
#
# INTERNAL METHODS
#
###################

847
sub _pick_best_analysis_for_new_worker {
Leo Gordon's avatar
Leo Gordon committed
848 849
  my $self  = shift;
  my $rc_id = shift;    # this parameter will need to percolate very deep
850 851 852 853

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

Leo Gordon's avatar
Leo Gordon committed
854
  my ($stats) = @{$statsDBA->fetch_by_needed_workers(1,$self->{maximise_concurrency}, $rc_id)};
855 856 857 858 859
  if($stats) {
    #synchronize and double check that it can be run
    $self->safe_synchronize_AnalysisStats($stats);
    return $stats if(($stats->status ne 'BLOCKED') and ($stats->num_required_workers > 0));
  }
860 861

  # ok so no analyses 'need' workers.
862 863 864
  if ($self->get_num_failed_analyses()) {
    return undef;
  }
865 866 867 868
  # see if anything needs an update, in case there are
  # hidden jobs that haven't made it into the summary stats

  print("QUEEN: no obvious needed workers, need to dig deeper\n");
Leo Gordon's avatar
Leo Gordon committed
869
  my $stats_list = $statsDBA->fetch_by_statuses(['LOADING', 'BLOCKED']);
870
  foreach $stats (@$stats_list) {
871
    $self->safe_synchronize_AnalysisStats($stats);
872

Leo Gordon's avatar
Leo Gordon committed
873
    return $stats if(($stats->status ne 'BLOCKED') and ($stats->num_required_workers > 0) and (!defined($rc_id) or ($stats->rc_id == $rc_id)));
874 875
  }

Leo Gordon's avatar
Leo Gordon committed
876
  ($stats) = @{$statsDBA->fetch_by_needed_workers(1,$self->{maximise_concurrency}, $rc_id)};
877 878 879 880 881 882
  return $stats if($stats);

  return undef;
}


Leo Gordon's avatar
Leo Gordon committed
883
=head2 fetch_by_worker_id
Jessica Severin's avatar
Jessica Severin committed
884 885 886

  Arg [1]    : int $id
               the unique database identifier for the feature to be obtained
887
  Example    : $feat = $queen->fetch_by_dbID(1234);
Jessica Severin's avatar
Jessica Severin committed
888 889 890 891 892 893 894 895
  Description: Returns the feature created from the database defined by the
               the id $id.
  Returntype : Bio::EnsEMBL::Hive::Worker
  Exceptions : thrown if $id is not defined
  Caller     : general

=cut

Leo Gordon's avatar
Leo Gordon committed
896
sub fetch_by_worker_id {
Jessica Severin's avatar
Jessica Severin committed
897 898 899
  my ($self,$id) = @_;

  unless(defined $id) {
900
    throw("fetch_by_dbID must have an id");
Jessica Severin's avatar
Jessica Severin committed
901 902 903 904
  }

  my @tabs = $self->_tables;

Leo Gordon's avatar
Leo Gordon committed
905
  my $constraint = "h.worker_id = $id";
Jessica Severin's avatar
Jessica Severin committed
906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980