]> git.proxmox.com Git - pve-ha-manager.git/blob - src/PVE/HA/Manager.pm
ce257afd195368834fb068419e1fe62370228543
[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::NodeStatus;
10
11 my $fence_delay = 60;
12
13 sub new {
14 my ($this, $haenv) = @_;
15
16 my $class = ref($this) || $this;
17
18 my $ms = $haenv->read_manager_status();
19
20 $ms->{master_node} = $haenv->nodename();
21
22 my $ns = PVE::HA::NodeStatus->new($haenv, $ms->{node_status} || {});
23
24 # fixme: use separate class PVE::HA::ServiceStatus
25 my $ss = $ms->{service_status} || {};
26
27 my $self = bless {
28 haenv => $haenv,
29 ms => $ms, # master status
30 ns => $ns, # PVE::HA::NodeStatus
31 ss => $ss, # service status
32 }, $class;
33
34 return $self;
35 }
36
37 sub cleanup {
38 my ($self) = @_;
39
40 # todo: ?
41 }
42
43 sub flush_master_status {
44 my ($self) = @_;
45
46 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
47
48 $ms->{node_status} = $ns->{status};
49 $ms->{service_status} = $ss;
50
51 $haenv->write_manager_status($ms);
52 }
53
54 sub select_service_node {
55 my ($groups, $online_node_usage, $service_conf, $current_node, $try_next) = @_;
56
57 my $group = { 'nodes' => $service_conf->{node} }; # default group
58
59 $group = $groups->{ids}->{$service_conf->{group}} if $service_conf->{group} &&
60 $groups->{ids}->{$service_conf->{group}};
61
62 my $pri_groups = {};
63 my $group_members = {};
64 foreach my $entry (PVE::Tools::split_list($group->{nodes})) {
65 my ($node, $pri) = ($entry, 0);
66 if ($entry =~ m/^(\S+):(\d+)$/) {
67 ($node, $pri) = ($1, $2);
68 }
69 next if !defined($online_node_usage->{$node}); # offline
70 $pri_groups->{$pri}->{$node} = 1;
71 $group_members->{$node} = $pri;
72 }
73
74
75 # add non-group members to unrestricted groups (priority -1)
76 if (!$group->{restricted}) {
77 my $pri = -1;
78 foreach my $node (keys %$online_node_usage) {
79 next if defined($group_members->{$node});
80 $pri_groups->{$pri}->{$node} = 1;
81 $group_members->{$node} = -1;
82 }
83 }
84
85
86 my @pri_list = sort {$b <=> $a} keys %$pri_groups;
87 return undef if !scalar(@pri_list);
88
89 if (!$try_next && $group->{nofailback} && defined($group_members->{$current_node})) {
90 return $current_node;
91 }
92
93 # select node from top priority node list
94
95 my $top_pri = $pri_list[0];
96
97 my @nodes = sort {
98 $online_node_usage->{$a} <=> $online_node_usage->{$b} || $a cmp $b
99 } keys %{$pri_groups->{$top_pri}};
100
101 my $found;
102 for (my $i = scalar(@nodes) - 1; $i >= 0; $i--) {
103 my $node = $nodes[$i];
104 if ($node eq $current_node) {
105 $found = $i;
106 last;
107 }
108 }
109
110 if ($try_next) {
111
112 if (defined($found) && ($found < (scalar(@nodes) - 1))) {
113 return $nodes[$found + 1];
114 } else {
115 return $nodes[0];
116 }
117
118 } else {
119
120 return $nodes[$found] if defined($found);
121
122 return $nodes[0];
123
124 }
125 }
126
127 my $uid_counter = 0;
128
129 sub compute_new_uuid {
130 my ($state) = @_;
131
132 $uid_counter++;
133 return md5_base64($state . $$ . time() . $uid_counter);
134 }
135
136 my $valid_service_states = {
137 stopped => 1,
138 request_stop => 1,
139 started => 1,
140 fence => 1,
141 migrate => 1,
142 relocate => 1,
143 freeze => 1,
144 error => 1,
145 };
146
147 sub recompute_online_node_usage {
148 my ($self) = @_;
149
150 my $online_node_usage = {};
151
152 my $online_nodes = $self->{ns}->list_online_nodes();
153
154 foreach my $node (@$online_nodes) {
155 $online_node_usage->{$node} = 0;
156 }
157
158 foreach my $sid (keys %{$self->{ss}}) {
159 my $sd = $self->{ss}->{$sid};
160 my $state = $sd->{state};
161 if (defined($online_node_usage->{$sd->{node}})) {
162 if (($state eq 'started') || ($state eq 'request_stop') ||
163 ($state eq 'fence') || ($state eq 'freeze') || ($state eq 'error')) {
164 $online_node_usage->{$sd->{node}}++;
165 } elsif (($state eq 'migrate') || ($state eq 'relocate')) {
166 $online_node_usage->{$sd->{target}}++;
167 } elsif ($state eq 'stopped') {
168 # do nothing
169 } else {
170 die "should not be reached";
171 }
172 }
173 }
174
175 $self->{online_node_usage} = $online_node_usage;
176 }
177
178 my $change_service_state = sub {
179 my ($self, $sid, $new_state, %params) = @_;
180
181 my ($haenv, $ss) = ($self->{haenv}, $self->{ss});
182
183 my $sd = $ss->{$sid} || die "no such service '$sid";
184
185 my $old_state = $sd->{state};
186 my $old_node = $sd->{node};
187
188 die "no state change" if $old_state eq $new_state; # just to be sure
189
190 die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state};
191
192 foreach my $k (keys %$sd) { delete $sd->{$k}; };
193
194 $sd->{state} = $new_state;
195 $sd->{node} = $old_node;
196
197 my $text_state = '';
198 foreach my $k (keys %params) {
199 my $v = $params{$k};
200 $text_state .= ", " if $text_state;
201 $text_state .= "$k = $v";
202 $sd->{$k} = $v;
203 }
204
205 $self->recompute_online_node_usage();
206
207 $sd->{uid} = compute_new_uuid($new_state);
208
209
210 $text_state = " ($text_state)" if $text_state;
211 $haenv->log('info', "service '$sid': state changed from '${old_state}' to '${new_state}' $text_state");
212 };
213
214 # read LRM status for all nodes
215 sub read_lrm_status {
216 my ($self) = @_;
217
218 my $nodes = $self->{ns}->list_nodes();
219 my $haenv = $self->{haenv};
220
221 my $results = {};
222 my $modes = {};
223 foreach my $node (@$nodes) {
224 my $lrm_status = $haenv->read_lrm_status($node);
225 $modes->{$node} = $lrm_status->{mode} || 'active';
226 foreach my $uid (keys %{$lrm_status->{results}}) {
227 next if $results->{$uid}; # should not happen
228 $results->{$uid} = $lrm_status->{results}->{$uid};
229 }
230 }
231
232
233 return ($results, $modes);
234 }
235
236 # read new crm commands and save them into crm master status
237 sub update_crm_commands {
238 my ($self) = @_;
239
240 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
241
242 my $cmdlist = $haenv->read_crm_commands();
243
244 foreach my $cmd (split(/\n/, $cmdlist)) {
245 chomp $cmd;
246
247 if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) {
248 my ($task, $sid, $node) = ($1, $2, $3);
249 if (my $sd = $ss->{$sid}) {
250 if (!$ns->node_is_online($node)) {
251 $haenv->log('err', "crm command error - node not online: $cmd");
252 } else {
253 if ($node eq $sd->{node}) {
254 $haenv->log('info', "ignore crm command - service already on target node: $cmd");
255 } else {
256 $haenv->log('info', "got crm command: $cmd");
257 $ss->{$sid}->{cmd} = [ $task, $node];
258 }
259 }
260 } else {
261 $haenv->log('err', "crm command error - no such service: $cmd");
262 }
263
264 } else {
265 $haenv->log('err', "unable to parse crm command: $cmd");
266 }
267 }
268
269 }
270
271 sub manage {
272 my ($self) = @_;
273
274 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
275
276 $ns->update($haenv->get_node_info());
277
278 if (!$ns->node_is_online($haenv->nodename())) {
279 $haenv->log('info', "master seems offline");
280 return;
281 }
282
283 my ($lrm_results, $lrm_modes) = $self->read_lrm_status();
284
285 my $sc = $haenv->read_service_config();
286
287 $self->{groups} = $haenv->read_group_config(); # update
288
289 # compute new service status
290
291 # add new service
292 foreach my $sid (sort keys %$sc) {
293 next if $ss->{$sid}; # already there
294 $haenv->log('info', "adding new service '$sid' on node '$sc->{$sid}->{node}'");
295 # assume we are running to avoid relocate running service at add
296 $ss->{$sid} = { state => 'started', node => $sc->{$sid}->{node},
297 uid => compute_new_uuid('started') };
298 }
299
300 # remove stale service from manager state
301 foreach my $sid (keys %$ss) {
302 next if $sc->{$sid};
303 $haenv->log('info', "removing stale service '$sid' (no config)");
304 delete $ss->{$sid};
305 }
306
307 $self->update_crm_commands();
308
309 for (;;) {
310 my $repeat = 0;
311
312 $self->recompute_online_node_usage();
313
314 foreach my $sid (keys %$ss) {
315 my $sd = $ss->{$sid};
316 my $cd = $sc->{$sid} || { state => 'disabled' };
317
318 my $lrm_res = $sd->{uid} ? $lrm_results->{$sd->{uid}} : undef;
319
320 my $last_state = $sd->{state};
321
322 if ($last_state eq 'stopped') {
323
324 $self->next_state_stopped($sid, $cd, $sd, $lrm_res);
325
326 } elsif ($last_state eq 'started') {
327
328 $self->next_state_started($sid, $cd, $sd, $lrm_res);
329
330 } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') {
331
332 $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res);
333
334 } elsif ($last_state eq 'fence') {
335
336 # do nothing here - wait until fenced
337
338 } elsif ($last_state eq 'request_stop') {
339
340 $self->next_state_request_stop($sid, $cd, $sd, $lrm_res);
341
342 } elsif ($last_state eq 'freeze') {
343
344 my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef;
345 # unfreeze
346 &$change_service_state($self, $sid, 'started')
347 if $lrm_mode && $lrm_mode eq 'active';
348
349 } elsif ($last_state eq 'error') {
350
351 # fixme:
352
353 } else {
354
355 die "unknown service state '$last_state'";
356 }
357
358
359 my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef;
360 $lrm_mode = 'unknown'if !$lrm_mode;
361 if (($sd->{state} eq 'started' || $sd->{state} eq 'stopped' ||
362 $sd->{state} eq 'request_stop') && ($lrm_mode ne 'active')) {
363 &$change_service_state($self, $sid, 'freeze');
364 }
365
366 $repeat = 1 if $sd->{state} ne $last_state;
367 }
368
369 # handle fencing
370 my $fenced_nodes = {};
371 foreach my $sid (keys %$ss) {
372 my $sd = $ss->{$sid};
373 next if $sd->{state} ne 'fence';
374
375 if (!defined($fenced_nodes->{$sd->{node}})) {
376 $fenced_nodes->{$sd->{node}} = $ns->fence_node($sd->{node}) || 0;
377 }
378
379 next if !$fenced_nodes->{$sd->{node}};
380
381 # node fence was sucessful - mark service as stopped
382 &$change_service_state($self, $sid, 'stopped');
383 }
384
385 last if !$repeat;
386 }
387
388 $self->flush_master_status();
389 }
390
391 # functions to compute next service states
392 # $cd: service configuration data (read only)
393 # $sd: service status data (read only)
394 #
395 # Note: use change_service_state() to alter state
396 #
397
398 sub next_state_request_stop {
399 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
400
401 my $haenv = $self->{haenv};
402 my $ns = $self->{ns};
403
404 # check result from LRM daemon
405 if ($lrm_res) {
406 my $exit_code = $lrm_res->{exit_code};
407 if ($exit_code == 0) {
408 &$change_service_state($self, $sid, 'stopped');
409 return;
410 } else {
411 &$change_service_state($self, $sid, 'error'); # fixme: what state?
412 return;
413 }
414 }
415
416 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
417 &$change_service_state($self, $sid, 'fence');
418 return;
419 }
420 }
421
422 sub next_state_migrate_relocate {
423 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
424
425 my $haenv = $self->{haenv};
426 my $ns = $self->{ns};
427
428 # check result from LRM daemon
429 if ($lrm_res) {
430 my $exit_code = $lrm_res->{exit_code};
431 if ($exit_code == 0) {
432 &$change_service_state($self, $sid, 'started', node => $sd->{target});
433 return;
434 } else {
435 $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)");
436 &$change_service_state($self, $sid, 'started', node => $sd->{node});
437 return;
438 }
439 }
440
441 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
442 &$change_service_state($self, $sid, 'fence');
443 return;
444 }
445 }
446
447
448 sub next_state_stopped {
449 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
450
451 my $haenv = $self->{haenv};
452 my $ns = $self->{ns};
453
454 if ($sd->{node} ne $cd->{node}) {
455 # this can happen if we fence a node with active migrations
456 # hack: modify $sd (normally this should be considered read-only)
457 $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node}");
458 $sd->{node} = $cd->{node};
459 }
460
461 if ($sd->{cmd}) {
462 my ($cmd, $target) = @{$sd->{cmd}};
463 delete $sd->{cmd};
464
465 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
466 if (!$ns->node_is_online($target)) {
467 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
468 } elsif ($sd->{node} eq $target) {
469 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
470 } else {
471 eval {
472 $haenv->change_service_location($sid, $sd->{node}, $target);
473 $cd->{node} = $sd->{node} = $target; # fixme: $sd is read-only??!!
474 $haenv->log('info', "$cmd service '$sid' to node '$target' (stopped)");
475 };
476 if (my $err = $@) {
477 $haenv->log('err', "$cmd service '$sid' to node '$target' failed - $err");
478 }
479 }
480 } else {
481 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
482 }
483 }
484
485 if ($cd->{state} eq 'disabled') {
486 # do nothing
487 return;
488 }
489
490 if ($cd->{state} eq 'enabled') {
491 if (my $node = select_service_node($self->{groups}, $self->{online_node_usage}, $cd, $sd->{node})) {
492 if ($node && ($sd->{node} ne $node)) {
493 eval {
494 $haenv->change_service_location($sid, $sd->{node}, $node);
495 $cd->{node} = $sd->{node} = $node; # fixme: $sd is read-only??!!
496 };
497 if (my $err = $@) {
498 $haenv->log('err', "move service '$sid' to node '$node' failed - $err");
499 } else {
500 &$change_service_state($self, $sid, 'started', node => $node);
501 }
502 } else {
503 &$change_service_state($self, $sid, 'started', node => $node);
504 }
505 } else {
506 # fixme: warn
507 }
508
509 return;
510 }
511
512 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
513 }
514
515 sub next_state_started {
516 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
517
518 my $haenv = $self->{haenv};
519 my $ns = $self->{ns};
520
521 if (!$ns->node_is_online($sd->{node})) {
522 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
523 &$change_service_state($self, $sid, 'fence');
524 }
525 return;
526 }
527
528 if ($cd->{state} eq 'disabled') {
529 &$change_service_state($self, $sid, 'request_stop');
530 return;
531 }
532
533 if ($cd->{state} eq 'enabled') {
534
535 if ($sd->{cmd}) {
536 my ($cmd, $target) = @{$sd->{cmd}};
537 delete $sd->{cmd};
538
539 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
540 if (!$ns->node_is_online($target)) {
541 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
542 } elsif ($sd->{node} eq $target) {
543 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
544 } else {
545 $haenv->log('info', "$cmd service '$sid' to node '$target' (running)");
546 &$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target);
547 }
548 } else {
549 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
550 }
551 } else {
552
553 my $try_next = 0;
554 if ($lrm_res && ($lrm_res->{exit_code} != 0)) { # fixme: other exit codes?
555 $try_next = 1;
556 }
557
558 my $node = select_service_node($self->{groups}, $self->{online_node_usage},
559 $cd, $sd->{node}, $try_next);
560
561 if ($node && ($sd->{node} ne $node)) {
562 $haenv->log('info', "migrate service '$sid' to node '$node' (running)");
563 &$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node);
564 } else {
565 # do nothing
566 }
567 }
568
569 return;
570 }
571
572 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
573 }
574
575 1;