src/PVE/HA/Manager.pm

   1 package PVE::HA::Manager;
   2
   3 use strict;
   4 use warnings;
   5 use Digest::MD5 qw(md5_base64);
   6
   7 use PVE::Tools;
   8 use PVE::HA::Tools ':exit_codes';
   9 use PVE::HA::NodeStatus;
  10
  11 ## Variable Name & Abbreviations Convention
  12 #
  13 # The HA stack has some variables it uses frequently and thus abbreviates it such that it may be
  14 # confusing for new readers. Here's a short list of the most common used.
  15 #
  16 # NOTE: variables should be assumed to be read only if not otherwise stated, only use the specific
  17 # methods to re-compute/read/alter them.
  18 #
  19 # - $haenv -> HA environment, the main interface to the simulator/test/real world
  20 # - $sid -> Service ID, unique identifier for a service, `type:vmid` is common
  21 #
  22 # - $ms -> Master/Manager Status, contains runtime info from the current active manager
  23 # - $ns -> Node Status, hash holding online/offline status about all nodes
  24 #
  25 # - $ss -> Service Status, hash holding the current state (last LRM cmd result, failed starts
  26 #          or migrates, maintenance fallback node, for *all* services ...
  27 # - $sd -> Service Data, the service status of a *single* service, iow. $ss->{$sid}
  28 #
  29 # - $sc -> Service Configuration, hash for all services including target state, group, ...
  30 # - $sd -> Configuration Data, the service config of a *single* service, iow. $sc->{$sid}
  31 #
  32 # Try to avoid adding new two letter (or similar over abbreviated) names, but also don't send
  33 # patches for changing above, as that set is mostly sensible and should be easy to remember once
  34 # spending a bit time in the HA code base.
  35
  36 sub new {
  37     my ($this, $haenv) = @_;
  38
  39     my $class = ref($this) || $this;
  40
  41     my $self = bless { haenv => $haenv }, $class;
  42
  43     my $old_ms = $haenv->read_manager_status();
  44
  45     # we only copy the state part of the manager which cannot be auto generated
  46
  47     $self->{ns} = PVE::HA::NodeStatus->new($haenv, $old_ms->{node_status} || {});
  48
  49     # fixme: use separate class  PVE::HA::ServiceStatus
  50     $self->{ss} = $old_ms->{service_status} || {};
  51
  52     $self->{ms} = { master_node => $haenv->nodename() };
  53
  54     return $self;
  55 }
  56
  57 sub cleanup {
  58     my ($self) = @_;
  59
  60     # todo: ?
  61 }
  62
  63 sub flush_master_status {
  64     my ($self) = @_;
  65
  66     my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
  67
  68     $ms->{node_status} = $ns->{status};
  69     $ms->{service_status} = $ss;
  70     $ms->{timestamp} = $haenv->get_time();
  71
  72     $haenv->write_manager_status($ms);
  73 }
  74
  75 sub get_service_group {
  76     my ($groups, $online_node_usage, $service_conf) = @_;
  77
  78     my $group = {};
  79     # add all online nodes to default group to allow try_next when no group set
  80     foreach my $node (keys %$online_node_usage) {
  81         $group->{nodes}->{$node} = 1;
  82     }
  83
  84     # overwrite default if service is bound to a specific group
  85     if (my $group_id = $service_conf->{group}) {
  86         $group = $groups->{ids}->{$group_id} if $groups->{ids}->{$group_id};
  87     }
  88
  89     return $group;
  90 }
  91
  92 # groups available nodes with their priority as group index
  93 sub get_node_priority_groups {
  94     my ($group, $online_node_usage) = @_;
  95
  96     my $pri_groups = {};
  97     my $group_members = {};
  98     foreach my $entry (keys %{$group->{nodes}}) {
  99         my ($node, $pri) = ($entry, 0);
 100         if ($entry =~ m/^(\S+):(\d+)$/) {
 101             ($node, $pri) = ($1, $2);
 102         }
 103         next if !defined($online_node_usage->{$node}); # offline
 104         $pri_groups->{$pri}->{$node} = 1;
 105         $group_members->{$node} = $pri;
 106     }
 107
 108     # add non-group members to unrestricted groups (priority -1)
 109     if (!$group->{restricted}) {
 110         my $pri = -1;
 111         foreach my $node (keys %$online_node_usage) {
 112             next if defined($group_members->{$node});
 113             $pri_groups->{$pri}->{$node} = 1;
 114             $group_members->{$node} = -1;
 115         }
 116     }
 117
 118     return ($pri_groups, $group_members);
 119 }
 120
 121 sub select_service_node {
 122     my ($groups, $online_node_usage, $service_conf, $current_node, $try_next, $tried_nodes, $maintenance_fallback) = @_;
 123
 124     my $group = get_service_group($groups, $online_node_usage, $service_conf);
 125
 126     my ($pri_groups, $group_members) = get_node_priority_groups($group, $online_node_usage);
 127
 128     my @pri_list = sort {$b <=> $a} keys %$pri_groups;
 129     return undef if !scalar(@pri_list);
 130
 131     # stay on current node if possible (avoids random migrations)
 132     if (!$try_next && $group->{nofailback} && defined($group_members->{$current_node})) {
 133         return $current_node;
 134     }
 135
 136     # select node from top priority node list
 137
 138     my $top_pri = $pri_list[0];
 139
 140     # try to avoid nodes where the service failed already if we want to relocate
 141     if ($try_next) {
 142         foreach my $node (@$tried_nodes) {
 143             delete $pri_groups->{$top_pri}->{$node};
 144         }
 145     }
 146
 147     my @nodes = sort {
 148         $online_node_usage->{$a} <=> $online_node_usage->{$b} || $a cmp $b
 149     } keys %{$pri_groups->{$top_pri}};
 150
 151     my $found;
 152     my $found_maintenance_fallback;
 153     for (my $i = scalar(@nodes) - 1; $i >= 0; $i--) {
 154         my $node = $nodes[$i];
 155         if ($node eq $current_node) {
 156             $found = $i;
 157         }
 158         if (defined($maintenance_fallback) && $node eq $maintenance_fallback) {
 159             $found_maintenance_fallback = $i;
 160         }
 161     }
 162
 163     if (defined($found_maintenance_fallback)) {
 164         return $nodes[$found_maintenance_fallback];
 165     }
 166
 167     if ($try_next) {
 168         if (defined($found) && ($found < (scalar(@nodes) - 1))) {
 169             return $nodes[$found + 1];
 170         } else {
 171             return $nodes[0];
 172         }
 173     } elsif (defined($found)) {
 174         return $nodes[$found];
 175     } else {
 176         return $nodes[0];
 177     }
 178 }
 179
 180 my $uid_counter = 0;
 181
 182 sub compute_new_uuid {
 183     my ($state) = @_;
 184
 185     $uid_counter++;
 186     return md5_base64($state . $$ . time() . $uid_counter);
 187 }
 188
 189 my $valid_service_states = {
 190     stopped => 1,
 191     request_stop => 1,
 192     started => 1,
 193     fence => 1,
 194     recovery => 1,
 195     migrate => 1,
 196     relocate => 1,
 197     freeze => 1,
 198     error => 1,
 199 };
 200
 201 sub recompute_online_node_usage {
 202     my ($self) = @_;
 203
 204     my $online_node_usage = {};
 205
 206     my $online_nodes = $self->{ns}->list_online_nodes();
 207
 208     foreach my $node (@$online_nodes) {
 209         $online_node_usage->{$node} = 0;
 210     }
 211
 212     foreach my $sid (keys %{$self->{ss}}) {
 213         my $sd = $self->{ss}->{$sid};
 214         my $state = $sd->{state};
 215         my $target = $sd->{target}; # optional
 216         if (defined($online_node_usage->{$sd->{node}})) {
 217             if (
 218                 $state eq 'started' || $state eq 'request_stop' || $state eq 'fence' ||
 219                 $state eq 'freeze' || $state eq 'error' || $state eq 'recovery'
 220             ) {
 221                 $online_node_usage->{$sd->{node}}++;
 222             } elsif (($state eq 'migrate') || ($state eq 'relocate')) {
 223                 # count it for both, source and target as load is put on both
 224                 $online_node_usage->{$sd->{node}}++;
 225                 $online_node_usage->{$target}++;
 226             } elsif ($state eq 'stopped') {
 227                 # do nothing
 228             } else {
 229                 die "should not be reached (sid = '$sid', state = '$state')";
 230             }
 231         } elsif (defined($target) && defined($online_node_usage->{$target})) {
 232             if ($state eq 'migrate' || $state eq 'relocate') {
 233                 # to correctly track maintenance modi and also consider the target as used for the
 234                 # case a node dies, as we cannot really know if the to-be-aborted incoming migration
 235                 # has already cleaned up all used resources
 236                 $online_node_usage->{$target}++;
 237             }
 238         }
 239     }
 240
 241     $self->{online_node_usage} = $online_node_usage;
 242 }
 243
 244 my $change_service_state = sub {
 245     my ($self, $sid, $new_state, %params) = @_;
 246
 247     my ($haenv, $ss) = ($self->{haenv}, $self->{ss});
 248
 249     my $sd = $ss->{$sid} || die "no such service '$sid";
 250
 251     my $old_state = $sd->{state};
 252     my $old_node = $sd->{node};
 253     my $old_failed_nodes = $sd->{failed_nodes};
 254     my $old_maintenance_node = $sd->{maintenance_node};
 255
 256     die "no state change" if $old_state eq $new_state; # just to be sure
 257
 258     die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state};
 259
 260     foreach my $k (keys %$sd) { delete $sd->{$k}; };
 261
 262     $sd->{state} = $new_state;
 263     $sd->{node} = $old_node;
 264     $sd->{failed_nodes} = $old_failed_nodes if defined($old_failed_nodes);
 265     $sd->{maintenance_node} = $old_maintenance_node if defined($old_maintenance_node);
 266
 267     my $text_state = '';
 268     foreach my $k (sort keys %params) {
 269         my $v = $params{$k};
 270         $text_state .= ", " if $text_state;
 271         $text_state .= "$k = $v";
 272         $sd->{$k} = $v;
 273     }
 274
 275     $self->recompute_online_node_usage();
 276
 277     $sd->{uid} = compute_new_uuid($new_state);
 278
 279     $text_state = "  ($text_state)" if $text_state;
 280     $haenv->log('info', "service '$sid': state changed from '${old_state}'" .
 281                 " to '${new_state}'$text_state");
 282 };
 283
 284 # clean up a possible bad state from a recovered service to allow its start
 285 my $fence_recovery_cleanup = sub {
 286     my ($self, $sid, $fenced_node) = @_;
 287
 288     my $haenv = $self->{haenv};
 289
 290     my (undef, $type, $id) = $haenv->parse_sid($sid);
 291     my $plugin = PVE::HA::Resources->lookup($type);
 292
 293     # should not happen
 294     die "unknown resource type '$type'" if !$plugin;
 295
 296     # locks may block recovery, cleanup those which are safe to remove after fencing,
 297     # i.e., after the original node was reset and thus all it's state
 298     my $removable_locks = [
 299         'backup',
 300         'mounted',
 301         'migrate',
 302         'clone',
 303         'rollback',
 304         'snapshot',
 305         'snapshot-delete',
 306         'suspending',
 307         'suspended',
 308     ];
 309     if (my $removed_lock = $plugin->remove_locks($haenv, $id, $removable_locks, $fenced_node)) {
 310         $haenv->log('warning', "removed leftover lock '$removed_lock' from recovered " .
 311                     "service '$sid' to allow its start.");
 312     }
 313 };
 314
 315 # read LRM status for all nodes
 316 sub read_lrm_status {
 317     my ($self) = @_;
 318
 319     my $nodes = $self->{ns}->list_nodes();
 320     my $haenv = $self->{haenv};
 321
 322     my $results = {};
 323     my $modes = {};
 324     foreach my $node (@$nodes) {
 325         my $lrm_status = $haenv->read_lrm_status($node);
 326         $modes->{$node} = $lrm_status->{mode} || 'active';
 327         foreach my $uid (keys %{$lrm_status->{results}}) {
 328             next if $results->{$uid}; # should not happen
 329             $results->{$uid} = $lrm_status->{results}->{$uid};
 330         }
 331     }
 332
 333     return ($results, $modes);
 334 }
 335
 336 # read new crm commands and save them into crm master status
 337 sub update_crm_commands {
 338     my ($self) = @_;
 339
 340     my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
 341
 342     my $cmdlist = $haenv->read_crm_commands();
 343
 344     foreach my $cmd (split(/\n/, $cmdlist)) {
 345         chomp $cmd;
 346
 347         if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) {
 348             my ($task, $sid, $node) = ($1, $2, $3);
 349             if (my $sd = $ss->{$sid}) {
 350                 if (!$ns->node_is_online($node)) {
 351                     $haenv->log('err', "crm command error - node not online: $cmd");
 352                 } else {
 353                     if ($node eq $sd->{node}) {
 354                         $haenv->log('info', "ignore crm command - service already on target node: $cmd");
 355                     } else {
 356                         $haenv->log('info', "got crm command: $cmd");
 357                         $ss->{$sid}->{cmd} = [ $task, $node ];
 358                     }
 359                 }
 360             } else {
 361                 $haenv->log('err', "crm command error - no such service: $cmd");
 362             }
 363
 364         } elsif ($cmd =~ m/^stop\s+(\S+)\s+(\S+)$/) {
 365             my ($sid, $timeout) = ($1, $2);
 366             if (my $sd = $ss->{$sid}) {
 367                 $haenv->log('info', "got crm command: $cmd");
 368                 $ss->{$sid}->{cmd} = [ 'stop', $timeout ];
 369             } else {
 370                 $haenv->log('err', "crm command error - no such service: $cmd");
 371             }
 372         } else {
 373             $haenv->log('err', "unable to parse crm command: $cmd");
 374         }
 375     }
 376
 377 }
 378
 379 sub manage {
 380     my ($self) = @_;
 381
 382     my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
 383
 384     my ($node_info) = $haenv->get_node_info();
 385     my ($lrm_results, $lrm_modes) = $self->read_lrm_status();
 386
 387     $ns->update($node_info, $lrm_modes);
 388
 389     if (!$ns->node_is_operational($haenv->nodename())) {
 390         $haenv->log('info', "master seems offline");
 391         return;
 392     }
 393
 394     my $sc = $haenv->read_service_config();
 395
 396     $self->{groups} = $haenv->read_group_config(); # update
 397
 398     # compute new service status
 399
 400     # add new service
 401     foreach my $sid (sort keys %$sc) {
 402         next if $ss->{$sid}; # already there
 403         my $cd = $sc->{$sid};
 404         next if $cd->{state} eq 'ignored';
 405
 406         $haenv->log('info', "adding new service '$sid' on node '$cd->{node}'");
 407         # assume we are running to avoid relocate running service at add
 408         my $state = ($cd->{state} eq 'started') ? 'started' : 'request_stop';
 409         $ss->{$sid} = { state => $state, node => $cd->{node},
 410                         uid => compute_new_uuid('started') };
 411     }
 412
 413     # remove stale or ignored services from manager state
 414     foreach my $sid (keys %$ss) {
 415         next if $sc->{$sid} && $sc->{$sid}->{state} ne 'ignored';
 416
 417         my $reason =  defined($sc->{$sid}) ? 'ignored state requested' : 'no config';
 418         $haenv->log('info', "removing stale service '$sid' ($reason)");
 419
 420         # remove all service related state information
 421         delete $ss->{$sid};
 422     }
 423
 424     $self->update_crm_commands();
 425
 426     for (;;) {
 427         my $repeat = 0;
 428
 429         $self->recompute_online_node_usage();
 430
 431         foreach my $sid (sort keys %$ss) {
 432             my $sd = $ss->{$sid};
 433             my $cd = $sc->{$sid} || { state => 'disabled' };
 434
 435             my $lrm_res = $sd->{uid} ? $lrm_results->{$sd->{uid}} : undef;
 436
 437             my $last_state = $sd->{state};
 438
 439             if ($last_state eq 'stopped') {
 440
 441                 $self->next_state_stopped($sid, $cd, $sd, $lrm_res);
 442
 443             } elsif ($last_state eq 'started') {
 444
 445                 $self->next_state_started($sid, $cd, $sd, $lrm_res);
 446
 447             } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') {
 448
 449                 $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res);
 450
 451             } elsif ($last_state eq 'fence') {
 452
 453                 # do nothing here - wait until fenced
 454
 455             } elsif ($last_state eq 'recovery') {
 456
 457                 $self->next_state_recovery($sid, $cd, $sd, $lrm_res);
 458
 459             } elsif ($last_state eq 'request_stop') {
 460
 461                 $self->next_state_request_stop($sid, $cd, $sd, $lrm_res);
 462
 463             } elsif ($last_state eq 'freeze') {
 464
 465                 my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef;
 466                 # unfreeze
 467                 my $state = ($cd->{state} eq 'started') ? 'started' : 'request_stop';
 468                 &$change_service_state($self, $sid, $state)
 469                     if $lrm_mode && $lrm_mode eq 'active';
 470
 471             } elsif ($last_state eq 'error') {
 472
 473                 $self->next_state_error($sid, $cd, $sd, $lrm_res);
 474
 475             } else {
 476
 477                 die "unknown service state '$last_state'";
 478             }
 479
 480             my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef;
 481             if ($lrm_mode && $lrm_mode eq 'restart') {
 482                 if (($sd->{state} eq 'started' || $sd->{state} eq 'stopped' ||
 483                      $sd->{state} eq 'request_stop')) {
 484                     &$change_service_state($self, $sid, 'freeze');
 485                 }
 486             }
 487
 488             $repeat = 1 if $sd->{state} ne $last_state;
 489         }
 490
 491         # handle fencing
 492         my $fenced_nodes = {};
 493         foreach my $sid (sort keys %$ss) {
 494             my ($service_state, $service_node) = $ss->{$sid}->@{'state', 'node'};
 495             next if $service_state ne 'fence';
 496
 497             if (!defined($fenced_nodes->{$service_node})) {
 498                 $fenced_nodes->{$service_node} = $ns->fence_node($service_node) || 0;
 499             }
 500
 501             next if !$fenced_nodes->{$service_node};
 502
 503             # node fence was successful - recover service
 504             $change_service_state->($self, $sid, 'recovery');
 505             $repeat = 1; # for faster recovery execution
 506         }
 507
 508         # Avoid that a node without services in 'fence' state (e.g., removed
 509         # manually by admin) is stuck with the 'fence' node state.
 510         for my $node (sort grep { !defined($fenced_nodes->{$_}) } keys $ns->{status}->%*) {
 511             next if $ns->get_node_state($node) ne 'fence';
 512
 513             $haenv->log('notice', "node '$node' in fence state but no services to-fence! admin interference?!");
 514             $repeat = 1 if $ns->fence_node($node);
 515         }
 516
 517         last if !$repeat;
 518     }
 519
 520     $self->flush_master_status();
 521 }
 522
 523 # functions to compute next service states
 524 # $cd: service configuration data (read only)
 525 # $sd: service status data (read only)
 526 #
 527 # Note: use change_service_state() to alter state
 528 #
 529
 530 sub next_state_request_stop {
 531     my ($self, $sid, $cd, $sd, $lrm_res) = @_;
 532
 533     my $haenv = $self->{haenv};
 534     my $ns = $self->{ns};
 535
 536     # check result from LRM daemon
 537     if ($lrm_res) {
 538         my $exit_code = $lrm_res->{exit_code};
 539         if ($exit_code == SUCCESS) {
 540             &$change_service_state($self, $sid, 'stopped');
 541             return;
 542         } else {
 543             $haenv->log('err', "service '$sid' stop failed (exit code $exit_code)");
 544             &$change_service_state($self, $sid, 'error'); # fixme: what state?
 545             return;
 546         }
 547     }
 548
 549     if ($ns->node_is_offline_delayed($sd->{node})) {
 550         &$change_service_state($self, $sid, 'fence');
 551         return;
 552     }
 553 }
 554
 555 sub next_state_migrate_relocate {
 556     my ($self, $sid, $cd, $sd, $lrm_res) = @_;
 557
 558     my $haenv = $self->{haenv};
 559     my $ns = $self->{ns};
 560
 561     # check result from LRM daemon
 562     if ($lrm_res) {
 563         my $exit_code = $lrm_res->{exit_code};
 564         my $req_state = $cd->{state} eq 'started' ? 'started' : 'request_stop';
 565         if ($exit_code == SUCCESS) {
 566             &$change_service_state($self, $sid, $req_state, node => $sd->{target});
 567             return;
 568         } elsif ($exit_code == EWRONG_NODE) {
 569             $haenv->log('err', "service '$sid' - migration failed: service" .
 570                         " registered on wrong node!");
 571             &$change_service_state($self, $sid, 'error');
 572         } else {
 573             $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)");
 574             &$change_service_state($self, $sid, $req_state, node => $sd->{node});
 575             return;
 576         }
 577     }
 578
 579     if ($ns->node_is_offline_delayed($sd->{node})) {
 580         &$change_service_state($self, $sid, 'fence');
 581         return;
 582     }
 583 }
 584
 585 sub next_state_stopped {
 586     my ($self, $sid, $cd, $sd, $lrm_res) = @_;
 587
 588     my $haenv = $self->{haenv};
 589     my $ns = $self->{ns};
 590
 591     if ($sd->{node} ne $cd->{node}) {
 592         # this can happen if we fence a node with active migrations
 593         # hack: modify $sd (normally this should be considered read-only)
 594         $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node})");
 595         $sd->{node} = $cd->{node};
 596     }
 597
 598     if ($sd->{cmd}) {
 599         my $cmd = shift @{$sd->{cmd}};
 600
 601         if ($cmd eq 'migrate' || $cmd eq 'relocate') {
 602             my $target = shift @{$sd->{cmd}};
 603             if (!$ns->node_is_online($target)) {
 604                 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
 605             } elsif ($sd->{node} eq $target) {
 606                 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
 607             } else {
 608                 &$change_service_state($self, $sid, $cmd, node => $sd->{node},
 609                                        target => $target);
 610                 return;
 611             }
 612         } elsif ($cmd eq 'stop') {
 613                 $haenv->log('info', "ignore service '$sid' $cmd request - service already stopped");
 614         } else {
 615             $haenv->log('err', "unknown command '$cmd' for service '$sid'");
 616         }
 617         delete $sd->{cmd};
 618     }
 619
 620     if ($cd->{state} eq 'disabled') {
 621         # NOTE: do nothing here, the stop state is an exception as we do not
 622         # process the LRM result here, thus the LRM always tries to stop the
 623         # service (protection for the case no CRM is active)
 624         return;
 625     }
 626
 627     if ($ns->node_is_offline_delayed($sd->{node}) && $ns->get_node_state($sd->{node}) ne 'maintenance') {
 628         &$change_service_state($self, $sid, 'fence');
 629         return;
 630     }
 631
 632     if ($cd->{state} eq 'stopped') {
 633         # almost the same as 'disabled' state but the service will also get recovered
 634         return;
 635     }
 636
 637     if ($cd->{state} eq 'started') {
 638         # simply mark it started, if it's on the wrong node
 639         # next_state_started will fix that for us
 640         &$change_service_state($self, $sid, 'started', node => $sd->{node});
 641         return;
 642     }
 643
 644     $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
 645 }
 646
 647 sub record_service_failed_on_node {
 648     my ($self, $sid, $node) = @_;
 649
 650     if (!defined($self->{ss}->{$sid}->{failed_nodes})) {
 651         $self->{ss}->{$sid}->{failed_nodes} = [];
 652     }
 653
 654     push @{$self->{ss}->{$sid}->{failed_nodes}}, $node;
 655 }
 656
 657 sub next_state_started {
 658     my ($self, $sid, $cd, $sd, $lrm_res) = @_;
 659
 660     my $haenv = $self->{haenv};
 661     my $master_status = $self->{ms};
 662     my $ns = $self->{ns};
 663
 664     if (!$ns->node_is_online($sd->{node})) {
 665         if ($ns->node_is_offline_delayed($sd->{node})) {
 666             &$change_service_state($self, $sid, 'fence');
 667         }
 668         if ($ns->get_node_state($sd->{node}) ne 'maintenance') {
 669             return;
 670         } else {
 671             # save current node as fallback for when it comes out of
 672             # maintenance
 673             $sd->{maintenance_node} = $sd->{node};
 674         }
 675     }
 676
 677     if ($cd->{state} eq 'disabled' || $cd->{state} eq 'stopped') {
 678         &$change_service_state($self, $sid, 'request_stop');
 679         return;
 680     }
 681
 682     if ($cd->{state} eq 'started') {
 683
 684         if ($sd->{cmd}) {
 685             my $cmd = shift @{$sd->{cmd}};
 686
 687             if ($cmd eq 'migrate' || $cmd eq 'relocate') {
 688                 my $target = shift @{$sd->{cmd}};
 689                 if (!$ns->node_is_online($target)) {
 690                     $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
 691                 } elsif ($sd->{node} eq $target) {
 692                     $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
 693                 } else {
 694                     $haenv->log('info', "$cmd service '$sid' to node '$target'");
 695                     &$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target);
 696                 }
 697             } elsif ($cmd eq 'stop') {
 698                 my $timeout = shift @{$sd->{cmd}};
 699                 if ($timeout == 0) {
 700                     $haenv->log('info', "request immediate service hard-stop for service '$sid'");
 701                 } else {
 702                     $haenv->log('info', "request graceful stop with timeout '$timeout' for service '$sid'");
 703                 }
 704                 &$change_service_state($self, $sid, 'request_stop', timeout => $timeout);
 705                 $haenv->update_service_config($sid, {'state' => 'stopped'});
 706             } else {
 707                 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
 708             }
 709
 710             delete $sd->{cmd};
 711
 712         } else {
 713
 714             my $try_next = 0;
 715
 716             if ($lrm_res) {
 717
 718                 my $ec = $lrm_res->{exit_code};
 719                 if ($ec == SUCCESS) {
 720
 721                     if (defined($sd->{failed_nodes})) {
 722                         $haenv->log('info', "relocation policy successful for '$sid' on node '$sd->{node}'," .
 723                                     " failed nodes: " . join(', ', @{$sd->{failed_nodes}}) );
 724                     }
 725
 726                     delete $sd->{failed_nodes};
 727
 728                     # store flag to indicate successful start - only valid while state == 'started'
 729                     $sd->{running} = 1;
 730
 731                 } elsif ($ec == ERROR) {
 732
 733                     delete $sd->{running};
 734
 735                     # apply our relocate policy if we got ERROR from the LRM
 736                     $self->record_service_failed_on_node($sid, $sd->{node});
 737
 738                     if (scalar(@{$sd->{failed_nodes}}) <= $cd->{max_relocate}) {
 739
 740                         # tell select_service_node to relocate if possible
 741                         $try_next = 1;
 742
 743                         $haenv->log('warning', "starting service $sid on node".
 744                                    " '$sd->{node}' failed, relocating service.");
 745
 746                     } else {
 747
 748                         $haenv->log('err', "recovery policy for service $sid " .
 749                                     "failed, entering error state. Failed nodes: ".
 750                                     join(', ', @{$sd->{failed_nodes}}));
 751                         &$change_service_state($self, $sid, 'error');
 752                         return;
 753
 754                     }
 755                 } else {
 756                     $self->record_service_failed_on_node($sid, $sd->{node});
 757
 758                     $haenv->log('err', "service '$sid' got unrecoverable error" .
 759                                 " (exit code $ec))");
 760                     # we have no save way out (yet) for other errors
 761                     &$change_service_state($self, $sid, 'error');
 762                     return;
 763                 }
 764             }
 765
 766             my $node = select_service_node(
 767                 $self->{groups},
 768                 $self->{online_node_usage},
 769                 $cd,
 770                 $sd->{node},
 771                 $try_next,
 772                 $sd->{failed_nodes},
 773                 $sd->{maintenance_node},
 774             );
 775
 776             if ($node && ($sd->{node} ne $node)) {
 777                 $self->{online_node_usage}->{$node}++;
 778
 779                 if (defined(my $fallback = $sd->{maintenance_node})) {
 780                     if ($node eq $fallback) {
 781                         $haenv->log('info', "moving service '$sid' back to '$fallback', node came back from maintenance.");
 782                         delete $sd->{maintenance_node};
 783                     } elsif ($sd->{node} ne $fallback) {
 784                         $haenv->log('info', "dropping maintenance fallback node '$fallback' for '$sid'");
 785                         delete $sd->{maintenance_node};
 786                     }
 787                 }
 788
 789                 if ($cd->{type} eq 'vm') {
 790                     $haenv->log('info', "migrate service '$sid' to node '$node' (running)");
 791                     &$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node);
 792                 } else {
 793                     $haenv->log('info', "relocate service '$sid' to node '$node'");
 794                     &$change_service_state($self, $sid, 'relocate', node => $sd->{node}, target => $node);
 795                 }
 796             } else {
 797                 if ($try_next && !defined($node)) {
 798                     $haenv->log('warning', "Start Error Recovery: Tried all available " .
 799                                 " nodes for service '$sid', retry start on current node. " .
 800                                 "Tried nodes: " . join(', ', @{$sd->{failed_nodes}}));
 801                 }
 802                 # ensure service get started again if it went unexpected down
 803                 # but ensure also no LRM result gets lost
 804                 $sd->{uid} = compute_new_uuid($sd->{state}) if defined($lrm_res);
 805             }
 806         }
 807
 808         return;
 809     }
 810
 811     $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
 812 }
 813
 814 sub next_state_error {
 815     my ($self, $sid, $cd, $sd, $lrm_res) = @_;
 816
 817     my $ns = $self->{ns};
 818     my $ms = $self->{ms};
 819
 820     if ($cd->{state} eq 'disabled') {
 821         # clean up on error recovery
 822         delete $sd->{failed_nodes};
 823
 824         &$change_service_state($self, $sid, 'stopped');
 825         return;
 826     }
 827
 828 }
 829
 830 # after a node was fenced this recovers the service to a new node
 831 sub next_state_recovery {
 832     my ($self, $sid, $cd, $sd, $lrm_res) = @_;
 833
 834     my ($haenv, $ss) = ($self->{haenv}, $self->{ss});
 835     my $ns = $self->{ns};
 836     my $ms = $self->{ms};
 837
 838     if ($sd->{state} ne 'recovery') { # should not happen
 839         $haenv->log('err', "cannot recover service '$sid' from fencing, wrong state '$sd->{state}'");
 840         return;
 841     }
 842
 843     my $fenced_node = $sd->{node}; # for logging purpose
 844
 845     $self->recompute_online_node_usage(); # we want the most current node state
 846
 847     my $recovery_node = select_service_node(
 848         $self->{groups},
 849         $self->{online_node_usage},
 850         $cd,
 851         $sd->{node},
 852     );
 853
 854     if ($recovery_node) {
 855         my $msg = "recover service '$sid' from fenced node '$fenced_node' to node '$recovery_node'";
 856         if ($recovery_node eq $fenced_node) {
 857             # can happen if restriced groups and the node came up again OK
 858             $msg = "recover service '$sid' to previous failed and fenced node '$fenced_node' again";
 859         }
 860         $haenv->log('info', "$msg");
 861
 862         $fence_recovery_cleanup->($self, $sid, $fenced_node);
 863
 864         $haenv->steal_service($sid, $sd->{node}, $recovery_node);
 865         $self->{online_node_usage}->{$recovery_node}++;
 866
 867         # NOTE: $sd *is normally read-only*, fencing is the exception
 868         $cd->{node} = $sd->{node} = $recovery_node;
 869         my $new_state = ($cd->{state} eq 'started') ? 'started' : 'request_stop';
 870         $change_service_state->($self, $sid, $new_state, node => $recovery_node);
 871     } else {
 872         # no possible node found, cannot recover - but retry later, as we always try to make it available
 873         $haenv->log('err', "recovering service '$sid' from fenced node '$fenced_node' failed, no recovery node found");
 874
 875         if ($cd->{state} eq 'disabled') {
 876             # allow getting a service out of recovery manually if an admin disables it.
 877             delete $sd->{failed_nodes}; # clean up on recovery to stopped
 878             $change_service_state->($self, $sid, 'stopped'); # must NOT go through request_stop
 879             return;
 880         }
 881     }
 882 }
 883
 884 1;