]> git.proxmox.com Git - pve-ha-manager.git/blob - src/PVE/HA/Manager.pm
remove stale service from manager state
[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 $fence_delay = 30;
12
13 sub new {
14 my ($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 $valid_service_states = {
137 stopped => 1,
138 request_stop => 1,
139 started => 1,
140 fence => 1,
141 migrate => 1,
142 relocate => 1,
143 error => 1,
144 };
145
146 sub recompute_online_node_usage {
147 my ($self) = @_;
148
149 my $online_node_usage = {};
150
151 my $online_nodes = $self->{ns}->list_online_nodes();
152
153 foreach my $node (@$online_nodes) {
154 $online_node_usage->{$node} = 0;
155 }
156
157 foreach my $sid (keys %{$self->{ss}}) {
158 my $sd = $self->{ss}->{$sid};
159 my $state = $sd->{state};
160 if (defined($online_node_usage->{$sd->{node}})) {
161 if (($state eq 'started') || ($state eq 'request_stop') ||
162 ($state eq 'fence') || ($state eq 'error')) {
163 $online_node_usage->{$sd->{node}}++;
164 } elsif (($state eq 'migrate') || ($state eq 'relocate')) {
165 $online_node_usage->{$sd->{target}}++;
166 } elsif ($state eq 'stopped') {
167 # do nothing
168 } else {
169 die "should not be reached";
170 }
171 }
172 }
173
174 $self->{online_node_usage} = $online_node_usage;
175 }
176
177 my $change_service_state = sub {
178 my ($self, $sid, $new_state, %params) = @_;
179
180 my ($haenv, $ss) = ($self->{haenv}, $self->{ss});
181
182 my $sd = $ss->{$sid} || die "no such service '$sid";
183
184 my $old_state = $sd->{state};
185 my $old_node = $sd->{node};
186
187 die "no state change" if $old_state eq $new_state; # just to be sure
188
189 die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state};
190
191 foreach my $k (keys %$sd) { delete $sd->{$k}; };
192
193 $sd->{state} = $new_state;
194 $sd->{node} = $old_node;
195
196 my $text_state = '';
197 foreach my $k (keys %params) {
198 my $v = $params{$k};
199 $text_state .= ", " if $text_state;
200 $text_state .= "$k = $v";
201 $sd->{$k} = $v;
202 }
203
204 $self->recompute_online_node_usage();
205
206 $sd->{uid} = compute_new_uuid($new_state);
207
208
209 $text_state = " ($text_state)" if $text_state;
210 $haenv->log('info', "service '$sid': state changed from '${old_state}' to '${new_state}' $text_state\n");
211 };
212
213 # read LRM status for all active nodes
214 sub read_lrm_status {
215 my ($self) = @_;
216
217 my $nodes = $self->{ns}->list_online_nodes();
218 my $haenv = $self->{haenv};
219
220 my $res = {};
221
222 foreach my $node (@$nodes) {
223 my $ls = $haenv->read_lrm_status($node);
224 foreach my $uid (keys %$ls) {
225 next if $res->{$uid}; # should not happen
226 $res->{$uid} = $ls->{$uid};
227 }
228 }
229
230 return $res;
231 }
232
233 # read new crm commands and save them into crm master status
234 sub update_crm_commands {
235 my ($self) = @_;
236
237 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
238
239 my $cmdlist = $haenv->read_crm_commands();
240
241 foreach my $cmd (split(/\n/, $cmdlist)) {
242 chomp $cmd;
243
244 if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) {
245 my ($task, $sid, $node) = ($1, $2, $3);
246 if (my $sd = $ss->{$sid}) {
247 if (!$ns->node_is_online($node)) {
248 $haenv->log('err', "crm command error - node not online: $cmd");
249 } else {
250 if ($node eq $sd->{node}) {
251 $haenv->log('info', "ignore crm command - service already on target node: $cmd");
252 } else {
253 $haenv->log('info', "got crm command: $cmd");
254 $ss->{$sid}->{cmd} = [ $task, $node];
255 }
256 }
257 } else {
258 $haenv->log('err', "crm command error - no such service: $cmd");
259 }
260
261 } else {
262 $haenv->log('err', "unable to parse crm command: $cmd");
263 }
264 }
265
266 }
267
268 sub manage {
269 my ($self) = @_;
270
271 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
272
273 $ns->update($haenv->get_node_info());
274
275 if (!$ns->node_is_online($haenv->nodename())) {
276 $haenv->log('info', "master seems offline\n");
277 return;
278 }
279
280 my $lrm_status = $self->read_lrm_status();
281
282 my $sc = $haenv->read_service_config();
283
284 $self->{groups} = $haenv->read_group_config(); # update
285
286 # compute new service status
287
288 # add new service
289 foreach my $sid (keys %$sc) {
290 next if $ss->{$sid}; # already there
291 $haenv->log('info', "Adding new service '$sid'\n");
292 # assume we are running to avoid relocate running service at add
293 $ss->{$sid} = { state => 'started', node => $sc->{$sid}->{node},
294 uid => compute_new_uuid('started') };
295 }
296
297 # remove stale service from manager state
298 foreach my $sid (keys %$ss) {
299 next if $sc->{$sid};
300 $haenv->log('info', "removing stale service '$sid' (no config)");
301 delete $ss->{$sid};
302 }
303
304 $self->update_crm_commands();
305
306 for (;;) {
307 my $repeat = 0;
308
309 $self->recompute_online_node_usage();
310
311 foreach my $sid (keys %$ss) {
312 my $sd = $ss->{$sid};
313 my $cd = $sc->{$sid} || { state => 'disabled' };
314
315 my $lrm_res = $sd->{uid} ? $lrm_status->{$sd->{uid}} : undef;
316
317 my $last_state = $sd->{state};
318
319 if ($last_state eq 'stopped') {
320
321 $self->next_state_stopped($sid, $cd, $sd, $lrm_res);
322
323 } elsif ($last_state eq 'started') {
324
325 $self->next_state_started($sid, $cd, $sd, $lrm_res);
326
327 } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') {
328
329 $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res);
330
331 } elsif ($last_state eq 'fence') {
332
333 # do nothing here - wait until fenced
334
335 } elsif ($last_state eq 'request_stop') {
336
337 $self->next_state_request_stop($sid, $cd, $sd, $lrm_res);
338
339 } elsif ($last_state eq 'error') {
340
341 # fixme:
342
343 } else {
344
345 die "unknown service state '$last_state'";
346 }
347
348 $repeat = 1 if $sd->{state} ne $last_state;
349 }
350
351 # handle fencing
352 my $fenced_nodes = {};
353 foreach my $sid (keys %$ss) {
354 my $sd = $ss->{$sid};
355 next if $sd->{state} ne 'fence';
356
357 if (!defined($fenced_nodes->{$sd->{node}})) {
358 $fenced_nodes->{$sd->{node}} = $ns->fence_node($sd->{node}) || 0;
359 }
360
361 next if !$fenced_nodes->{$sd->{node}};
362
363 # node fence was sucessful - mark service as stopped
364 &$change_service_state($self, $sid, 'stopped');
365 }
366
367 last if !$repeat;
368 }
369
370 $self->flush_master_status();
371 }
372
373 # functions to compute next service states
374 # $cd: service configuration data (read only)
375 # $sd: service status data (read only)
376 #
377 # Note: use change_service_state() to alter state
378 #
379
380 sub next_state_request_stop {
381 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
382
383 my $haenv = $self->{haenv};
384 my $ns = $self->{ns};
385
386 # check result from LRM daemon
387 if ($lrm_res) {
388 my $exit_code = $lrm_res->{exit_code};
389 if ($exit_code == 0) {
390 &$change_service_state($self, $sid, 'stopped');
391 return;
392 } else {
393 &$change_service_state($self, $sid, 'error'); # fixme: what state?
394 return;
395 }
396 }
397
398 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
399 &$change_service_state($self, $sid, 'fence');
400 return;
401 }
402 }
403
404 sub next_state_migrate_relocate {
405 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
406
407 my $haenv = $self->{haenv};
408 my $ns = $self->{ns};
409
410 # check result from LRM daemon
411 if ($lrm_res) {
412 my $exit_code = $lrm_res->{exit_code};
413 if ($exit_code == 0) {
414 &$change_service_state($self, $sid, 'started', node => $sd->{target});
415 return;
416 } else {
417 $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)");
418 &$change_service_state($self, $sid, 'started', node => $sd->{node});
419 return;
420 }
421 }
422
423 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
424 &$change_service_state($self, $sid, 'fence');
425 return;
426 }
427 }
428
429
430 sub next_state_stopped {
431 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
432
433 my $haenv = $self->{haenv};
434 my $ns = $self->{ns};
435
436 if ($sd->{node} ne $cd->{node}) {
437 # this can happen if we fence a node with active migrations
438 # hack: modify $sd (normally this should be considered read-only)
439 $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node}");
440 $sd->{node} = $cd->{node};
441 }
442
443 if ($sd->{cmd}) {
444 my ($cmd, $target) = @{$sd->{cmd}};
445 delete $sd->{cmd};
446
447 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
448 if (!$ns->node_is_online($target)) {
449 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
450 } elsif ($sd->{node} eq $target) {
451 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
452 } else {
453 $haenv->change_service_location($sid, $target);
454 $cd->{node} = $sd->{node} = $target; # fixme: $sd is read-only??!!
455 $haenv->log('info', "$cmd service '$sid' to node '$target' (stopped)");
456 }
457 } else {
458 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
459 }
460 }
461
462 if ($cd->{state} eq 'disabled') {
463 # do nothing
464 return;
465 }
466
467 if ($cd->{state} eq 'enabled') {
468 if (my $node = select_service_node($self->{groups}, $self->{online_node_usage}, $cd, $sd->{node})) {
469 if ($node && ($sd->{node} ne $node)) {
470 $haenv->change_service_location($sid, $node);
471 }
472 &$change_service_state($self, $sid, 'started', node => $node);
473 } else {
474 # fixme: warn
475 }
476
477 return;
478 }
479
480 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
481 }
482
483 sub next_state_started {
484 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
485
486 my $haenv = $self->{haenv};
487 my $ns = $self->{ns};
488
489 if (!$ns->node_is_online($sd->{node})) {
490 if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) {
491 &$change_service_state($self, $sid, 'fence');
492 }
493 return;
494 }
495
496 if ($cd->{state} eq 'disabled') {
497 &$change_service_state($self, $sid, 'request_stop');
498 return;
499 }
500
501 if ($cd->{state} eq 'enabled') {
502
503 if ($sd->{cmd}) {
504 my ($cmd, $target) = @{$sd->{cmd}};
505 delete $sd->{cmd};
506
507 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
508 if (!$ns->node_is_online($target)) {
509 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
510 } elsif ($sd->{node} eq $target) {
511 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
512 } else {
513 $haenv->log('info', "$cmd service '$sid' to node '$target' (running)");
514 &$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target);
515 }
516 } else {
517 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
518 }
519 } else {
520
521 my $try_next = 0;
522 if ($lrm_res && ($lrm_res->{exit_code} != 0)) { # fixme: other exit codes?
523 $try_next = 1;
524 }
525
526 my $node = select_service_node($self->{groups}, $self->{online_node_usage},
527 $cd, $sd->{node}, $try_next);
528
529 if ($node && ($sd->{node} ne $node)) {
530 $haenv->log('info', "migrate service '$sid' to node '$node' (running)");
531 &$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node);
532 } else {
533 # do nothing
534 }
535 }
536
537 return;
538 }
539
540 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
541 }
542
543 1;