]> git.proxmox.com Git - pve-ha-manager.git/blob - src/PVE/HA/Manager.pm
resource agents: fix relocate
[pve-ha-manager.git] / 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 Data::Dumper;
8 use PVE::Tools;
9 use PVE::HA::Tools ':exit_codes';
10 use PVE::HA::NodeStatus;
11
12 my $fence_delay = 60;
13
14 sub new {
15 my ($this, $haenv) = @_;
16
17 my $class = ref($this) || $this;
18
19 my $ms = $haenv->read_manager_status();
20
21 $ms->{master_node} = $haenv->nodename();
22
23 my $ns = PVE::HA::NodeStatus->new($haenv, $ms->{node_status} || {});
24
25 # fixme: use separate class PVE::HA::ServiceStatus
26 my $ss = $ms->{service_status} || {};
27
28 my $self = bless {
29 haenv => $haenv,
30 ms => $ms, # master status
31 ns => $ns, # PVE::HA::NodeStatus
32 ss => $ss, # service status
33 }, $class;
34
35 return $self;
36 }
37
38 sub cleanup {
39 my ($self) = @_;
40
41 # todo: ?
42 }
43
44 sub flush_master_status {
45 my ($self) = @_;
46
47 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
48
49 $ms->{node_status} = $ns->{status};
50 $ms->{service_status} = $ss;
51 $ms->{timestamp} = $haenv->get_time();
52
53 $haenv->write_manager_status($ms);
54 }
55
56 sub select_service_node {
57 my ($groups, $online_node_usage, $service_conf, $current_node, $try_next) = @_;
58
59 my $group = { 'nodes' => { $service_conf->{node} => 1 } }; # default group
60
61 $group = $groups->{ids}->{$service_conf->{group}} if $service_conf->{group} &&
62 $groups->{ids}->{$service_conf->{group}};
63
64 my $pri_groups = {};
65 my $group_members = {};
66 foreach my $entry (keys %{$group->{nodes}}) {
67 my ($node, $pri) = ($entry, 0);
68 if ($entry =~ m/^(\S+):(\d+)$/) {
69 ($node, $pri) = ($1, $2);
70 }
71 next if !defined($online_node_usage->{$node}); # offline
72 $pri_groups->{$pri}->{$node} = 1;
73 $group_members->{$node} = $pri;
74 }
75
76
77 # add non-group members to unrestricted groups (priority -1)
78 if (!$group->{restricted}) {
79 my $pri = -1;
80 foreach my $node (keys %$online_node_usage) {
81 next if defined($group_members->{$node});
82 $pri_groups->{$pri}->{$node} = 1;
83 $group_members->{$node} = -1;
84 }
85 }
86
87
88 my @pri_list = sort {$b <=> $a} keys %$pri_groups;
89 return undef if !scalar(@pri_list);
90
91 if (!$try_next && $group->{nofailback} && defined($group_members->{$current_node})) {
92 return $current_node;
93 }
94
95 # select node from top priority node list
96
97 my $top_pri = $pri_list[0];
98
99 my @nodes = sort {
100 $online_node_usage->{$a} <=> $online_node_usage->{$b} || $a cmp $b
101 } keys %{$pri_groups->{$top_pri}};
102
103 my $found;
104 for (my $i = scalar(@nodes) - 1; $i >= 0; $i--) {
105 my $node = $nodes[$i];
106 if ($node eq $current_node) {
107 $found = $i;
108 last;
109 }
110 }
111
112 if ($try_next) {
113
114 if (defined($found) && ($found < (scalar(@nodes) - 1))) {
115 return $nodes[$found + 1];
116 } else {
117 return $nodes[0];
118 }
119
120 } else {
121
122 return $nodes[$found] if defined($found);
123
124 return $nodes[0];
125
126 }
127 }
128
129 my $uid_counter = 0;
130
131 sub compute_new_uuid {
132 my ($state) = @_;
133
134 $uid_counter++;
135 return md5_base64($state . $$ . time() . $uid_counter);
136 }
137
138 my $valid_service_states = {
139 stopped => 1,
140 request_stop => 1,
141 started => 1,
142 fence => 1,
143 migrate => 1,
144 relocate => 1,
145 freeze => 1,
146 error => 1,
147 };
148
149 sub recompute_online_node_usage {
150 my ($self) = @_;
151
152 my $online_node_usage = {};
153
154 my $online_nodes = $self->{ns}->list_online_nodes();
155
156 foreach my $node (@$online_nodes) {
157 $online_node_usage->{$node} = 0;
158 }
159
160 foreach my $sid (keys %{$self->{ss}}) {
161 my $sd = $self->{ss}->{$sid};
162 my $state = $sd->{state};
163 if (defined($online_node_usage->{$sd->{node}})) {
164 if (($state eq 'started') || ($state eq 'request_stop') ||
165 ($state eq 'fence') || ($state eq 'freeze') || ($state eq 'error')) {
166 $online_node_usage->{$sd->{node}}++;
167 } elsif (($state eq 'migrate') || ($state eq 'relocate')) {
168 $online_node_usage->{$sd->{target}}++;
169 } elsif ($state eq 'stopped') {
170 # do nothing
171 } else {
172 die "should not be reached";
173 }
174 }
175 }
176
177 $self->{online_node_usage} = $online_node_usage;
178 }
179
180 my $change_service_state = sub {
181 my ($self, $sid, $new_state, %params) = @_;
182
183 my ($haenv, $ss) = ($self->{haenv}, $self->{ss});
184
185 my $sd = $ss->{$sid} || die "no such service '$sid";
186
187 my $old_state = $sd->{state};
188 my $old_node = $sd->{node};
189
190 die "no state change" if $old_state eq $new_state; # just to be sure
191
192 die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state};
193
194 foreach my $k (keys %$sd) { delete $sd->{$k}; };
195
196 $sd->{state} = $new_state;
197 $sd->{node} = $old_node;
198
199 my $text_state = '';
200 foreach my $k (sort keys %params) {
201 my $v = $params{$k};
202 $text_state .= ", " if $text_state;
203 $text_state .= "$k = $v";
204 $sd->{$k} = $v;
205 }
206
207 $self->recompute_online_node_usage();
208
209 $sd->{uid} = compute_new_uuid($new_state);
210
211
212 $text_state = " ($text_state)" if $text_state;
213 $haenv->log('info', "service '$sid': state changed from '${old_state}' to '${new_state}' $text_state");
214 };
215
216 # read LRM status for all nodes
217 sub read_lrm_status {
218 my ($self) = @_;
219
220 my $nodes = $self->{ns}->list_nodes();
221 my $haenv = $self->{haenv};
222
223 my $results = {};
224 my $modes = {};
225 foreach my $node (@$nodes) {
226 my $lrm_status = $haenv->read_lrm_status($node);
227 $modes->{$node} = $lrm_status->{mode} || 'active';
228 foreach my $uid (keys %{$lrm_status->{results}}) {
229 next if $results->{$uid}; # should not happen
230 $results->{$uid} = $lrm_status->{results}->{$uid};
231 }
232 }
233
234
235 return ($results, $modes);
236 }
237
238 # read new crm commands and save them into crm master status
239 sub update_crm_commands {
240 my ($self) = @_;
241
242 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
243
244 my $cmdlist = $haenv->read_crm_commands();
245
246 foreach my $cmd (split(/\n/, $cmdlist)) {
247 chomp $cmd;
248
249 if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) {
250 my ($task, $sid, $node) = ($1, $2, $3);
251 if (my $sd = $ss->{$sid}) {
252 if (!$ns->node_is_online($node)) {
253 $haenv->log('err', "crm command error - node not online: $cmd");
254 } else {
255 if ($node eq $sd->{node}) {
256 $haenv->log('info', "ignore crm command - service already on target node: $cmd");
257 } else {
258 $haenv->log('info', "got crm command: $cmd");
259 $ss->{$sid}->{cmd} = [ $task, $node];
260 }
261 }
262 } else {
263 $haenv->log('err', "crm command error - no such service: $cmd");
264 }
265
266 } else {
267 $haenv->log('err', "unable to parse crm command: $cmd");
268 }
269 }
270
271 }
272
273 sub manage {
274 my ($self) = @_;
275
276 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
277
278 $ns->update($haenv->get_node_info());
279
280 if (!$ns->node_is_online($haenv->nodename())) {
281 $haenv->log('info', "master seems offline");
282 return;
283 }
284
285 my ($lrm_results, $lrm_modes) = $self->read_lrm_status();
286
287 my $sc = $haenv->read_service_config();
288
289 $self->{groups} = $haenv->read_group_config(); # update
290
291 # compute new service status
292
293 # add new service
294 foreach my $sid (sort keys %$sc) {
295 next if $ss->{$sid}; # already there
296 $haenv->log('info', "adding new service '$sid' on node '$sc->{$sid}->{node}'");
297 # assume we are running to avoid relocate running service at add
298 $ss->{$sid} = { state => 'started', node => $sc->{$sid}->{node},
299 uid => compute_new_uuid('started') };
300 }
301
302 # remove stale service from manager state
303 foreach my $sid (keys %$ss) {
304 next if $sc->{$sid};
305 $haenv->log('info', "removing stale service '$sid' (no config)");
306 delete $ss->{$sid};
307 }
308
309 $self->update_crm_commands();
310
311 for (;;) {
312 my $repeat = 0;
313
314 $self->recompute_online_node_usage();
315
316 foreach my $sid (keys %$ss) {
317 my $sd = $ss->{$sid};
318 my $cd = $sc->{$sid} || { state => 'disabled' };
319
320 my $lrm_res = $sd->{uid} ? $lrm_results->{$sd->{uid}} : undef;
321
322 my $last_state = $sd->{state};
323
324 if ($last_state eq 'stopped') {
325
326 $self->next_state_stopped($sid, $cd, $sd, $lrm_res);
327
328 } elsif ($last_state eq 'started') {
329
330 $self->next_state_started($sid, $cd, $sd, $lrm_res);
331
332 } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') {
333
334 $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res);
335
336 } elsif ($last_state eq 'fence') {
337
338 # do nothing here - wait until fenced
339
340 } elsif ($last_state eq 'request_stop') {
341
342 $self->next_state_request_stop($sid, $cd, $sd, $lrm_res);
343
344 } elsif ($last_state eq 'freeze') {
345
346 my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef;
347 # unfreeze
348 &$change_service_state($self, $sid, 'started')
349 if $lrm_mode && $lrm_mode eq 'active';
350
351 } elsif ($last_state eq 'error') {
352
353 $self->next_state_error($sid, $cd, $sd, $lrm_res);
354
355 } else {
356
357 die "unknown service state '$last_state'";
358 }
359
360 my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef;
361 if ($lrm_mode && $lrm_mode eq 'restart') {
362 if (($sd->{state} eq 'started' || $sd->{state} eq 'stopped' ||
363 $sd->{state} eq 'request_stop')) {
364 &$change_service_state($self, $sid, 'freeze');
365 }
366 }
367
368 $repeat = 1 if $sd->{state} ne $last_state;
369 }
370
371 # handle fencing
372 my $fenced_nodes = {};
373 foreach my $sid (keys %$ss) {
374 my $sd = $ss->{$sid};
375 next if $sd->{state} ne 'fence';
376
377 if (!defined($fenced_nodes->{$sd->{node}})) {
378 $fenced_nodes->{$sd->{node}} = $ns->fence_node($sd->{node}) || 0;
379 }
380
381 next if !$fenced_nodes->{$sd->{node}};
382
383 # node fence was successful - mark service as stopped
384 &$change_service_state($self, $sid, 'stopped');
385 }
386
387 last if !$repeat;
388 }
389
390 $self->flush_master_status();
391 }
392
393 # functions to compute next service states
394 # $cd: service configuration data (read only)
395 # $sd: service status data (read only)
396 #
397 # Note: use change_service_state() to alter state
398 #
399
400 sub next_state_request_stop {
401 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
402
403 my $haenv = $self->{haenv};
404 my $ns = $self->{ns};
405
406 # check result from LRM daemon
407 if ($lrm_res) {
408 my $exit_code = $lrm_res->{exit_code};
409 if ($exit_code == SUCCESS) {
410 &$change_service_state($self, $sid, 'stopped');
411 return;
412 } else {
413 $haenv->log('err', "service '$sid' stop failed (exit code $exit_code)");
414 &$change_service_state($self, $sid, 'error'); # fixme: what state?
415 return;
416 }
417 }
418
419 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
420 &$change_service_state($self, $sid, 'fence');
421 return;
422 }
423 }
424
425 sub next_state_migrate_relocate {
426 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
427
428 my $haenv = $self->{haenv};
429 my $ns = $self->{ns};
430
431 # check result from LRM daemon
432 if ($lrm_res) {
433 my $exit_code = $lrm_res->{exit_code};
434 if ($exit_code == SUCCESS) {
435 &$change_service_state($self, $sid, 'started', node => $sd->{target});
436 return;
437 } else {
438 $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)");
439 &$change_service_state($self, $sid, 'started', node => $sd->{node});
440 return;
441 }
442 }
443
444 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
445 &$change_service_state($self, $sid, 'fence');
446 return;
447 }
448 }
449
450
451 sub next_state_stopped {
452 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
453
454 my $haenv = $self->{haenv};
455 my $ns = $self->{ns};
456
457 if ($sd->{node} ne $cd->{node}) {
458 # this can happen if we fence a node with active migrations
459 # hack: modify $sd (normally this should be considered read-only)
460 $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node}");
461 $sd->{node} = $cd->{node};
462 }
463
464 if ($sd->{cmd}) {
465 my ($cmd, $target) = @{$sd->{cmd}};
466 delete $sd->{cmd};
467
468 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
469 if (!$ns->node_is_online($target)) {
470 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
471 } elsif ($sd->{node} eq $target) {
472 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
473 } else {
474 eval {
475 $haenv->change_service_location($sid, $sd->{node}, $target);
476 $cd->{node} = $sd->{node} = $target; # fixme: $sd is read-only??!!
477 $haenv->log('info', "$cmd service '$sid' to node '$target' (stopped)");
478 };
479 if (my $err = $@) {
480 $haenv->log('err', "$cmd service '$sid' to node '$target' failed - $err");
481 }
482 }
483 } else {
484 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
485 }
486 }
487
488 if ($cd->{state} eq 'disabled') {
489 # do nothing
490 return;
491 }
492
493 if ($cd->{state} eq 'enabled') {
494 if (my $node = select_service_node($self->{groups}, $self->{online_node_usage}, $cd, $sd->{node})) {
495 if ($node && ($sd->{node} ne $node)) {
496 eval {
497 $haenv->change_service_location($sid, $sd->{node}, $node);
498 $cd->{node} = $sd->{node} = $node; # fixme: $sd is read-only??!!
499 };
500 if (my $err = $@) {
501 $haenv->log('err', "move service '$sid' to node '$node' failed - $err");
502 } else {
503 &$change_service_state($self, $sid, 'started', node => $node);
504 }
505 } else {
506 &$change_service_state($self, $sid, 'started', node => $node);
507 }
508 } else {
509 # fixme: warn
510 }
511
512 return;
513 }
514
515 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
516 }
517
518 sub next_state_started {
519 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
520
521 my $haenv = $self->{haenv};
522 my $master_status = $self->{ms};
523 my $ns = $self->{ns};
524
525 if (!$ns->node_is_online($sd->{node})) {
526 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
527 &$change_service_state($self, $sid, 'fence');
528 }
529 return;
530 }
531
532 if ($cd->{state} eq 'disabled') {
533 &$change_service_state($self, $sid, 'request_stop');
534 return;
535 }
536
537 if ($cd->{state} eq 'enabled') {
538
539 if ($sd->{cmd}) {
540 my ($cmd, $target) = @{$sd->{cmd}};
541 delete $sd->{cmd};
542
543 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
544 if (!$ns->node_is_online($target)) {
545 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
546 } elsif ($sd->{node} eq $target) {
547 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
548 } else {
549 $haenv->log('info', "$cmd service '$sid' to node '$target' (running)");
550 &$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target);
551 }
552 } else {
553 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
554 }
555 } else {
556
557 my $try_next = 0;
558 if ($lrm_res) {
559 if ($lrm_res->{exit_code} == ERROR) {
560
561 my $try = $master_status->{relocate_trial}->{$sid} || 0;
562
563 if ($try < $cd->{max_relocate}) {
564
565 $try++;
566 $try_next = 1; # tell select_service_node to relocate
567
568 $haenv->log('warning', "starting service $sid on node".
569 " '$sd->{node}' failed, relocating service.");
570 $master_status->{relocate_trial}->{$sid} = $try;
571
572 } else {
573
574 $haenv->log('err', "recovery policy for service".
575 " $sid failed, entering error state!");
576 &$change_service_state($self, $sid, 'error');
577 return;
578
579 }
580 } elsif ($lrm_res->{exit_code} == SUCCESS) {
581 $master_status->{relocate_trial}->{$sid} = 0;
582 }
583 }
584
585 my $node = select_service_node($self->{groups}, $self->{online_node_usage},
586 $cd, $sd->{node}, $try_next);
587
588 if ($node && ($sd->{node} ne $node)) {
589 if ($cd->{type} eq 'vm') {
590 $haenv->log('info', "migrate service '$sid' to node '$node' (running)");
591 &$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node);
592 } else {
593 $haenv->log('info', "relocate service '$sid' to node '$node'");
594 &$change_service_state($self, $sid, 'relocate', node => $sd->{node}, target => $node);
595 }
596 } else {
597 # do nothing
598 }
599 }
600
601 return;
602 }
603
604 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
605 }
606
607 sub next_state_error {
608 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
609
610 my $ns = $self->{ns};
611
612 if ($cd->{state} eq 'disabled') {
613 &$change_service_state($self, $sid, 'stopped');
614 return;
615 }
616
617 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
618 &$change_service_state($self, $sid, 'fence');
619 return;
620 }
621
622 }
623
624 1;