]> git.proxmox.com Git - pve-ha-manager.git/blame - src/PVE/HA/Manager.pm
log errors when writing lrm status
[pve-ha-manager.git] / src / PVE / HA / Manager.pm
CommitLineData
c0bbd038
DM
1package PVE::HA::Manager;
2
3use strict;
4use warnings;
c4a221bc 5use Digest::MD5 qw(md5_base64);
c0bbd038
DM
6
7use Data::Dumper;
c142ebc9 8use PVE::Tools;
c0bbd038
DM
9use PVE::HA::NodeStatus;
10
ceac1930 11my $fence_delay = 60;
5385a606 12
c0bbd038 13sub new {
8f0bb968 14 my ($this, $haenv) = @_;
c0bbd038
DM
15
16 my $class = ref($this) || $this;
17
8f0bb968
DM
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
59fd7207
DM
24 # fixme: use separate class PVE::HA::ServiceStatus
25 my $ss = $ms->{service_status} || {};
26
c0bbd038 27 my $self = bless {
8f0bb968
DM
28 haenv => $haenv,
29 ms => $ms, # master status
30 ns => $ns, # PVE::HA::NodeStatus
59fd7207 31 ss => $ss, # service status
c0bbd038
DM
32 }, $class;
33
34 return $self;
35}
36
d84da043
DM
37sub cleanup {
38 my ($self) = @_;
39
40 # todo: ?
41}
42
8f0bb968 43sub flush_master_status {
c0bbd038
DM
44 my ($self) = @_;
45
59fd7207 46 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
c0bbd038 47
8f0bb968 48 $ms->{node_status} = $ns->{status};
59fd7207
DM
49 $ms->{service_status} = $ss;
50
8f0bb968
DM
51 $haenv->write_manager_status($ms);
52}
c0bbd038 53
f7ccd1b3 54sub select_service_node {
270d4406 55 my ($groups, $online_node_usage, $service_conf, $current_node, $try_next) = @_;
f7ccd1b3 56
abc920b4
DM
57 my $group = { 'nodes' => $service_conf->{node} }; # default group
58
270d4406
DM
59 $group = $groups->{ids}->{$service_conf->{group}} if $service_conf->{group} &&
60 $groups->{ids}->{$service_conf->{group}};
abc920b4
DM
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, $2);
68 }
270d4406 69 next if !defined($online_node_usage->{$node}); # offline
abc920b4
DM
70 $pri_groups->{$pri}->{$node} = 1;
71 $group_members->{$node} = $pri;
72 }
f7ccd1b3 73
270d4406 74
abc920b4
DM
75 # add non-group members to unrestricted groups (priority -1)
76 if (!$group->{restricted}) {
77 my $pri = -1;
270d4406 78 foreach my $node (keys %$online_node_usage) {
abc920b4
DM
79 next if defined($group_members->{$node});
80 $pri_groups->{$pri}->{$node} = 1;
81 $group_members->{$node} = -1;
82 }
83 }
84
270d4406 85
abc920b4
DM
86 my @pri_list = sort {$b <=> $a} keys %$pri_groups;
87 return undef if !scalar(@pri_list);
270d4406 88
abc920b4
DM
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
c142ebc9
DM
97 my @nodes = sort {
98 $online_node_usage->{$a} <=> $online_node_usage->{$b} || $a cmp $b
99 } keys %{$pri_groups->{$top_pri}};
abc920b4
DM
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 = $i;
106 last;
107 }
108 }
109
abc920b4
DM
110 if ($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 }
f7ccd1b3
DM
125}
126
c4a221bc
DM
127my $uid_counter = 0;
128
d55aa611
DM
129sub compute_new_uuid {
130 my ($state) = @_;
131
132 $uid_counter++;
133 return md5_base64($state . $$ . time() . $uid_counter);
134}
135
618fbeda
DM
136my $valid_service_states = {
137 stopped => 1,
138 request_stop => 1,
139 started => 1,
140 fence => 1,
141 migrate => 1,
b0fdf86a 142 relocate => 1,
9c7d068b 143 freeze => 1,
618fbeda
DM
144 error => 1,
145};
146
270d4406
DM
147sub recompute_online_node_usage {
148 my ($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') || ($state eq 'request_stop') ||
9c7d068b 163 ($state eq 'fence') || ($state eq 'freeze') || ($state eq 'error')) {
270d4406
DM
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
4e01bc86
DM
178my $change_service_state = sub {
179 my ($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};
e4ffb299 186 my $old_node = $sd->{node};
4e01bc86
DM
187
188 die "no state change" if $old_state eq $new_state; # just to be sure
189
618fbeda
DM
190 die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state};
191
e4ffb299
DM
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 = '';
4e01bc86
DM
198 foreach my $k (keys %params) {
199 my $v = $params{$k};
e4ffb299
DM
200 $text_state .= ", " if $text_state;
201 $text_state .= "$k = $v";
4e01bc86
DM
202 $sd->{$k} = $v;
203 }
270d4406
DM
204
205 $self->recompute_online_node_usage();
206
d55aa611
DM
207 $sd->{uid} = compute_new_uuid($new_state);
208
4e01bc86 209
e4ffb299 210 $text_state = " ($text_state)" if $text_state;
e5986717 211 $haenv->log('info', "service '$sid': state changed from '${old_state}' to '${new_state}' $text_state");
4e01bc86
DM
212};
213
9c7d068b 214# read LRM status for all nodes
c4a221bc 215sub read_lrm_status {
332170bd 216 my ($self) = @_;
c4a221bc 217
9c7d068b 218 my $nodes = $self->{ns}->list_nodes();
c4a221bc
DM
219 my $haenv = $self->{haenv};
220
9c7d068b
DM
221 my $results = {};
222 my $modes = {};
332170bd 223 foreach my $node (@$nodes) {
9c7d068b
DM
224 my $lrm_status = $haenv->read_lrm_status($node);
225 $modes->{$node} = $lrm_status->{mode} || 'unknown';
226 foreach my $uid (keys %{$lrm_status->{results}}) {
227 next if $results->{$uid}; # should not happen
228 $results->{$uid} = $lrm_status->{results}->{$uid};
c4a221bc
DM
229 }
230 }
231
9c7d068b
DM
232
233 return ($results, $modes);
c4a221bc
DM
234}
235
aa98a844
DM
236# read new crm commands and save them into crm master status
237sub update_crm_commands {
238 my ($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 $cmd;
246
b0fdf86a
DM
247 if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) {
248 my ($task, $sid, $node) = ($1, $2, $3);
aa98a844
DM
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");
b0fdf86a 257 $ss->{$sid}->{cmd} = [ $task, $node];
aa98a844
DM
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: $cmd");
266 }
267 }
268
269}
270
8f0bb968
DM
271sub manage {
272 my ($self) = @_;
c0bbd038 273
59fd7207 274 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
c0bbd038 275
332170bd 276 $ns->update($haenv->get_node_info());
c79442f2
DM
277
278 if (!$ns->node_is_online($haenv->nodename())) {
e5986717 279 $haenv->log('info', "master seems offline");
c79442f2
DM
280 return;
281 }
282
9c7d068b 283 my ($lrm_results, $lrm_modes) = $self->read_lrm_status();
c4a221bc 284
f7ccd1b3
DM
285 my $sc = $haenv->read_service_config();
286
abc920b4
DM
287 $self->{groups} = $haenv->read_group_config(); # update
288
f7ccd1b3
DM
289 # compute new service status
290
291 # add new service
cc32a8f3 292 foreach my $sid (sort keys %$sc) {
f7ccd1b3 293 next if $ss->{$sid}; # already there
cc32a8f3 294 $haenv->log('info', "adding new service '$sid' on node '$sc->{$sid}->{node}'");
f7ccd1b3 295 # assume we are running to avoid relocate running service at add
d55aa611
DM
296 $ss->{$sid} = { state => 'started', node => $sc->{$sid}->{node},
297 uid => compute_new_uuid('started') };
f7ccd1b3
DM
298 }
299
4e5764af
DM
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
aa98a844
DM
307 $self->update_crm_commands();
308
c79442f2
DM
309 for (;;) {
310 my $repeat = 0;
270d4406
DM
311
312 $self->recompute_online_node_usage();
f7ccd1b3 313
c79442f2
DM
314 foreach my $sid (keys %$ss) {
315 my $sd = $ss->{$sid};
316 my $cd = $sc->{$sid} || { state => 'disabled' };
f7ccd1b3 317
9c7d068b 318 my $lrm_res = $sd->{uid} ? $lrm_results->{$sd->{uid}} : undef;
a875fbe8 319
c79442f2
DM
320 my $last_state = $sd->{state};
321
322 if ($last_state eq 'stopped') {
323
abc920b4 324 $self->next_state_stopped($sid, $cd, $sd, $lrm_res);
f7ccd1b3 325
c79442f2 326 } elsif ($last_state eq 'started') {
f7ccd1b3 327
abc920b4 328 $self->next_state_started($sid, $cd, $sd, $lrm_res);
f7ccd1b3 329
b0fdf86a 330 } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') {
f7ccd1b3 331
8aaa0e36 332 $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res);
f7ccd1b3 333
c79442f2 334 } elsif ($last_state eq 'fence') {
f7ccd1b3 335
21e37ed4 336 # do nothing here - wait until fenced
f7ccd1b3 337
c79442f2 338 } elsif ($last_state eq 'request_stop') {
f7ccd1b3 339
0df5b3dd 340 $self->next_state_request_stop($sid, $cd, $sd, $lrm_res);
618fbeda 341
9c7d068b
DM
342 } elsif ($last_state eq 'freeze') {
343
344 my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef;
345 $lrm_mode = 'unknown'if !$lrm_mode;
346 # unfreeze
347 &$change_service_state($self, $sid, 'started')
348 if $lrm_mode eq 'active';
349
e88469ba
DM
350 } elsif ($last_state eq 'error') {
351
352 # fixme:
353
a875fbe8
DM
354 } else {
355
356 die "unknown service state '$last_state'";
618fbeda 357 }
21e37ed4 358
9c7d068b
DM
359
360 my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef;
361 $lrm_mode = 'unknown'if !$lrm_mode;
362 if (($sd->{state} eq 'started' || $sd->{state} eq 'stopped' ||
363 $sd->{state} eq 'request_stop') && ($lrm_mode ne 'active')) {
364 &$change_service_state($self, $sid, 'freeze');
365 }
366
c79442f2 367 $repeat = 1 if $sd->{state} ne $last_state;
f7ccd1b3
DM
368 }
369
21e37ed4
DM
370 # handle fencing
371 my $fenced_nodes = {};
372 foreach my $sid (keys %$ss) {
373 my $sd = $ss->{$sid};
374 next if $sd->{state} ne 'fence';
375
376 if (!defined($fenced_nodes->{$sd->{node}})) {
377 $fenced_nodes->{$sd->{node}} = $ns->fence_node($sd->{node}) || 0;
378 }
379
380 next if !$fenced_nodes->{$sd->{node}};
381
382 # node fence was sucessful - mark service as stopped
383 &$change_service_state($self, $sid, 'stopped');
384 }
385
c79442f2 386 last if !$repeat;
f7ccd1b3 387 }
f7ccd1b3 388
8f0bb968 389 $self->flush_master_status();
c0bbd038
DM
390}
391
a875fbe8
DM
392# functions to compute next service states
393# $cd: service configuration data (read only)
394# $sd: service status data (read only)
395#
396# Note: use change_service_state() to alter state
397#
398
0df5b3dd
DM
399sub next_state_request_stop {
400 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
401
402 my $haenv = $self->{haenv};
403 my $ns = $self->{ns};
404
405 # check result from LRM daemon
406 if ($lrm_res) {
407 my $exit_code = $lrm_res->{exit_code};
408 if ($exit_code == 0) {
409 &$change_service_state($self, $sid, 'stopped');
410 return;
411 } else {
412 &$change_service_state($self, $sid, 'error'); # fixme: what state?
413 return;
414 }
415 }
416
5385a606 417 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
0df5b3dd
DM
418 &$change_service_state($self, $sid, 'fence');
419 return;
420 }
421}
422
8aaa0e36
DM
423sub next_state_migrate_relocate {
424 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
425
426 my $haenv = $self->{haenv};
427 my $ns = $self->{ns};
428
429 # check result from LRM daemon
430 if ($lrm_res) {
431 my $exit_code = $lrm_res->{exit_code};
432 if ($exit_code == 0) {
433 &$change_service_state($self, $sid, 'started', node => $sd->{target});
434 return;
435 } else {
436 $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)");
437 &$change_service_state($self, $sid, 'started', node => $sd->{node});
438 return;
439 }
440 }
441
5385a606 442 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
8aaa0e36
DM
443 &$change_service_state($self, $sid, 'fence');
444 return;
445 }
446}
447
448
a875fbe8 449sub next_state_stopped {
abc920b4 450 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
a875fbe8
DM
451
452 my $haenv = $self->{haenv};
e88469ba 453 my $ns = $self->{ns};
a875fbe8 454
ff6f1c5c
DM
455 if ($sd->{node} ne $cd->{node}) {
456 # this can happen if we fence a node with active migrations
457 # hack: modify $sd (normally this should be considered read-only)
458 $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node}");
459 $sd->{node} = $cd->{node};
460 }
461
94b7ebe2
DM
462 if ($sd->{cmd}) {
463 my ($cmd, $target) = @{$sd->{cmd}};
464 delete $sd->{cmd};
465
b0fdf86a 466 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
94b7ebe2 467 if (!$ns->node_is_online($target)) {
b0fdf86a 468 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
e88469ba 469 } elsif ($sd->{node} eq $target) {
b0fdf86a 470 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
94b7ebe2
DM
471 } else {
472 $haenv->change_service_location($sid, $target);
473 $cd->{node} = $sd->{node} = $target; # fixme: $sd is read-only??!!
b0fdf86a 474 $haenv->log('info', "$cmd service '$sid' to node '$target' (stopped)");
94b7ebe2
DM
475 }
476 } else {
477 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
478 }
479 }
480
a875fbe8
DM
481 if ($cd->{state} eq 'disabled') {
482 # do nothing
e88469ba
DM
483 return;
484 }
485
486 if ($cd->{state} eq 'enabled') {
270d4406 487 if (my $node = select_service_node($self->{groups}, $self->{online_node_usage}, $cd, $sd->{node})) {
a875fbe8
DM
488 if ($node && ($sd->{node} ne $node)) {
489 $haenv->change_service_location($sid, $node);
490 }
491 &$change_service_state($self, $sid, 'started', node => $node);
492 } else {
493 # fixme: warn
494 }
e88469ba
DM
495
496 return;
a875fbe8 497 }
e88469ba
DM
498
499 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
a875fbe8
DM
500}
501
502sub next_state_started {
abc920b4 503 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
a875fbe8
DM
504
505 my $haenv = $self->{haenv};
506 my $ns = $self->{ns};
507
508 if (!$ns->node_is_online($sd->{node})) {
5385a606
DM
509 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
510 &$change_service_state($self, $sid, 'fence');
511 }
e88469ba
DM
512 return;
513 }
a875fbe8 514
e88469ba
DM
515 if ($cd->{state} eq 'disabled') {
516 &$change_service_state($self, $sid, 'request_stop');
517 return;
518 }
519
520 if ($cd->{state} eq 'enabled') {
e88469ba
DM
521
522 if ($sd->{cmd}) {
523 my ($cmd, $target) = @{$sd->{cmd}};
524 delete $sd->{cmd};
525
b0fdf86a 526 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
e88469ba 527 if (!$ns->node_is_online($target)) {
b0fdf86a 528 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
e88469ba 529 } elsif ($sd->{node} eq $target) {
b0fdf86a 530 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
e88469ba 531 } else {
b0fdf86a
DM
532 $haenv->log('info', "$cmd service '$sid' to node '$target' (running)");
533 &$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target);
e88469ba 534 }
a875fbe8 535 } else {
e88469ba 536 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
a875fbe8
DM
537 }
538 } else {
b0fdf86a 539
abc920b4
DM
540 my $try_next = 0;
541 if ($lrm_res && ($lrm_res->{exit_code} != 0)) { # fixme: other exit codes?
542 $try_next = 1;
543 }
544
270d4406
DM
545 my $node = select_service_node($self->{groups}, $self->{online_node_usage},
546 $cd, $sd->{node}, $try_next);
abc920b4 547
b0fdf86a
DM
548 if ($node && ($sd->{node} ne $node)) {
549 $haenv->log('info', "migrate service '$sid' to node '$node' (running)");
550 &$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node);
551 } else {
552 # do nothing
553 }
a875fbe8 554 }
e88469ba
DM
555
556 return;
557 }
558
559 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
a875fbe8 560}
c0bbd038
DM
561
5621;