]>
Commit | Line | Data |
---|---|---|
c0bbd038 DM |
1 | package PVE::HA::Manager; |
2 | ||
3 | use strict; | |
4 | use warnings; | |
c4a221bc | 5 | use Digest::MD5 qw(md5_base64); |
c0bbd038 | 6 | |
c142ebc9 | 7 | use PVE::Tools; |
a89ff919 | 8 | use PVE::HA::Tools ':exit_codes'; |
c0bbd038 | 9 | use PVE::HA::NodeStatus; |
5d724d4d | 10 | use PVE::HA::Usage::Basic; |
561e7f4b | 11 | use PVE::HA::Usage::Static; |
c0bbd038 | 12 | |
a3ffb0b3 TL |
13 | ## Variable Name & Abbreviations Convention |
14 | # | |
15 | # The HA stack has some variables it uses frequently and thus abbreviates it such that it may be | |
16 | # confusing for new readers. Here's a short list of the most common used. | |
17 | # | |
18 | # NOTE: variables should be assumed to be read only if not otherwise stated, only use the specific | |
19 | # methods to re-compute/read/alter them. | |
20 | # | |
21 | # - $haenv -> HA environment, the main interface to the simulator/test/real world | |
22 | # - $sid -> Service ID, unique identifier for a service, `type:vmid` is common | |
23 | # | |
24 | # - $ms -> Master/Manager Status, contains runtime info from the current active manager | |
25 | # - $ns -> Node Status, hash holding online/offline status about all nodes | |
26 | # | |
27 | # - $ss -> Service Status, hash holding the current state (last LRM cmd result, failed starts | |
28 | # or migrates, maintenance fallback node, for *all* services ... | |
29 | # - $sd -> Service Data, the service status of a *single* service, iow. $ss->{$sid} | |
30 | # | |
31 | # - $sc -> Service Configuration, hash for all services including target state, group, ... | |
0869c306 | 32 | # - $cd -> Configuration Data, the service config of a *single* service, iow. $sc->{$sid} |
a3ffb0b3 TL |
33 | # |
34 | # Try to avoid adding new two letter (or similar over abbreviated) names, but also don't send | |
35 | # patches for changing above, as that set is mostly sensible and should be easy to remember once | |
36 | # spending a bit time in the HA code base. | |
37 | ||
c0bbd038 | 38 | sub new { |
8f0bb968 | 39 | my ($this, $haenv) = @_; |
c0bbd038 DM |
40 | |
41 | my $class = ref($this) || $this; | |
42 | ||
1b813831 | 43 | my $self = bless { haenv => $haenv, crs => {} }, $class; |
8f0bb968 | 44 | |
6ee64cfc | 45 | my $old_ms = $haenv->read_manager_status(); |
8f0bb968 | 46 | |
6ee64cfc TL |
47 | # we only copy the state part of the manager which cannot be auto generated |
48 | ||
49 | $self->{ns} = PVE::HA::NodeStatus->new($haenv, $old_ms->{node_status} || {}); | |
8f0bb968 | 50 | |
59fd7207 | 51 | # fixme: use separate class PVE::HA::ServiceStatus |
6ee64cfc TL |
52 | $self->{ss} = $old_ms->{service_status} || {}; |
53 | ||
54 | $self->{ms} = { master_node => $haenv->nodename() }; | |
c0bbd038 | 55 | |
086f7075 | 56 | $self->update_crs_scheduler_mode(); # initial set, we update it once every loop |
f74f8ffb | 57 | |
c0bbd038 DM |
58 | return $self; |
59 | } | |
60 | ||
cb06cd42 TL |
61 | sub update_crs_scheduler_mode { |
62 | my ($self) = @_; | |
63 | ||
64 | my $haenv = $self->{haenv}; | |
65 | my $dc_cfg = $haenv->get_datacenter_settings(); | |
66 | ||
1b813831 | 67 | my $old_mode = $self->{crs}->{scheduler}; |
cb06cd42 TL |
68 | my $new_mode = $dc_cfg->{crs}->{ha} || 'basic'; |
69 | ||
70 | if (!defined($old_mode)) { | |
71 | $haenv->log('info', "using scheduler mode '$new_mode'") if $new_mode ne 'basic'; | |
086f7075 TL |
72 | } elsif ($new_mode eq $old_mode) { |
73 | return; # nothing to do | |
74 | } else { | |
75 | $haenv->log('info', "switching scheduler mode from '$old_mode' to '$new_mode'"); | |
cb06cd42 TL |
76 | } |
77 | ||
1b813831 | 78 | $self->{crs}->{scheduler} = $new_mode; |
cb06cd42 TL |
79 | |
80 | return; | |
81 | } | |
82 | ||
d84da043 DM |
83 | sub cleanup { |
84 | my ($self) = @_; | |
85 | ||
86 | # todo: ? | |
87 | } | |
88 | ||
8f0bb968 | 89 | sub flush_master_status { |
c0bbd038 DM |
90 | my ($self) = @_; |
91 | ||
59fd7207 | 92 | my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss}); |
c0bbd038 | 93 | |
8f0bb968 | 94 | $ms->{node_status} = $ns->{status}; |
59fd7207 | 95 | $ms->{service_status} = $ss; |
d2f612cf | 96 | $ms->{timestamp} = $haenv->get_time(); |
289e4784 | 97 | |
8f0bb968 | 98 | $haenv->write_manager_status($ms); |
289e4784 | 99 | } |
c0bbd038 | 100 | |
48a6ba2a TL |
101 | sub get_service_group { |
102 | my ($groups, $online_node_usage, $service_conf) = @_; | |
f7ccd1b3 | 103 | |
09c5c4bf TL |
104 | my $group = {}; |
105 | # add all online nodes to default group to allow try_next when no group set | |
5d724d4d | 106 | $group->{nodes}->{$_} = 1 for $online_node_usage->list_nodes(); |
abc920b4 | 107 | |
09c5c4bf | 108 | # overwrite default if service is bound to a specific group |
3458a0e3 TL |
109 | if (my $group_id = $service_conf->{group}) { |
110 | $group = $groups->{ids}->{$group_id} if $groups->{ids}->{$group_id}; | |
111 | } | |
abc920b4 | 112 | |
48a6ba2a TL |
113 | return $group; |
114 | } | |
115 | ||
116 | # groups available nodes with their priority as group index | |
117 | sub get_node_priority_groups { | |
118 | my ($group, $online_node_usage) = @_; | |
119 | ||
abc920b4 DM |
120 | my $pri_groups = {}; |
121 | my $group_members = {}; | |
e0a56314 | 122 | foreach my $entry (keys %{$group->{nodes}}) { |
abc920b4 DM |
123 | my ($node, $pri) = ($entry, 0); |
124 | if ($entry =~ m/^(\S+):(\d+)$/) { | |
125 | ($node, $pri) = ($1, $2); | |
126 | } | |
5d724d4d | 127 | next if !$online_node_usage->contains_node($node); # offline |
abc920b4 DM |
128 | $pri_groups->{$pri}->{$node} = 1; |
129 | $group_members->{$node} = $pri; | |
130 | } | |
f7ccd1b3 | 131 | |
abc920b4 DM |
132 | # add non-group members to unrestricted groups (priority -1) |
133 | if (!$group->{restricted}) { | |
134 | my $pri = -1; | |
5d724d4d | 135 | for my $node ($online_node_usage->list_nodes()) { |
abc920b4 DM |
136 | next if defined($group_members->{$node}); |
137 | $pri_groups->{$pri}->{$node} = 1; | |
138 | $group_members->{$node} = -1; | |
139 | } | |
140 | } | |
141 | ||
48a6ba2a TL |
142 | return ($pri_groups, $group_members); |
143 | } | |
144 | ||
145 | sub select_service_node { | |
2fdf40f2 | 146 | my ($groups, $online_node_usage, $sid, $service_conf, $current_node, $try_next, $tried_nodes, $maintenance_fallback, $best_scored) = @_; |
48a6ba2a TL |
147 | |
148 | my $group = get_service_group($groups, $online_node_usage, $service_conf); | |
149 | ||
150 | my ($pri_groups, $group_members) = get_node_priority_groups($group, $online_node_usage); | |
151 | ||
abc920b4 DM |
152 | my @pri_list = sort {$b <=> $a} keys %$pri_groups; |
153 | return undef if !scalar(@pri_list); | |
09c5c4bf TL |
154 | |
155 | # stay on current node if possible (avoids random migrations) | |
2fdf40f2 | 156 | if ((!$try_next && !$best_scored) && $group->{nofailback} && defined($group_members->{$current_node})) { |
abc920b4 DM |
157 | return $current_node; |
158 | } | |
159 | ||
160 | # select node from top priority node list | |
161 | ||
162 | my $top_pri = $pri_list[0]; | |
163 | ||
e6eeb7dc TL |
164 | # try to avoid nodes where the service failed already if we want to relocate |
165 | if ($try_next) { | |
166 | foreach my $node (@$tried_nodes) { | |
167 | delete $pri_groups->{$top_pri}->{$node}; | |
168 | } | |
169 | } | |
170 | ||
631ba60e FE |
171 | return $maintenance_fallback |
172 | if defined($maintenance_fallback) && $pri_groups->{$top_pri}->{$maintenance_fallback}; | |
173 | ||
2fdf40f2 | 174 | return $current_node if (!$try_next && !$best_scored) && $pri_groups->{$top_pri}->{$current_node}; |
c724ce1b | 175 | |
5d724d4d | 176 | my $scores = $online_node_usage->score_nodes_to_start_service($sid, $current_node); |
289e4784 | 177 | my @nodes = sort { |
5d724d4d | 178 | $scores->{$a} <=> $scores->{$b} || $a cmp $b |
c142ebc9 | 179 | } keys %{$pri_groups->{$top_pri}}; |
abc920b4 DM |
180 | |
181 | my $found; | |
182 | for (my $i = scalar(@nodes) - 1; $i >= 0; $i--) { | |
183 | my $node = $nodes[$i]; | |
184 | if ($node eq $current_node) { | |
185 | $found = $i; | |
abc920b4 DM |
186 | } |
187 | } | |
188 | ||
abc920b4 | 189 | if ($try_next) { |
2fdf40f2 | 190 | if (!$best_scored && defined($found) && ($found < (scalar(@nodes) - 1))) { |
abc920b4 DM |
191 | return $nodes[$found + 1]; |
192 | } else { | |
193 | return $nodes[0]; | |
194 | } | |
abc920b4 | 195 | } else { |
abc920b4 | 196 | return $nodes[0]; |
abc920b4 | 197 | } |
f7ccd1b3 DM |
198 | } |
199 | ||
c4a221bc DM |
200 | my $uid_counter = 0; |
201 | ||
d55aa611 DM |
202 | sub compute_new_uuid { |
203 | my ($state) = @_; | |
289e4784 | 204 | |
d55aa611 DM |
205 | $uid_counter++; |
206 | return md5_base64($state . $$ . time() . $uid_counter); | |
207 | } | |
208 | ||
618fbeda DM |
209 | my $valid_service_states = { |
210 | stopped => 1, | |
211 | request_stop => 1, | |
4931b586 | 212 | request_start => 1, |
618fbeda DM |
213 | started => 1, |
214 | fence => 1, | |
c259b1a8 | 215 | recovery => 1, |
618fbeda | 216 | migrate => 1, |
b0fdf86a | 217 | relocate => 1, |
9c7d068b | 218 | freeze => 1, |
618fbeda DM |
219 | error => 1, |
220 | }; | |
221 | ||
561e7f4b FE |
222 | # FIXME with 'static' mode and thousands of services, the overhead can be noticable and the fact |
223 | # that this function is called for each state change and upon recovery doesn't help. | |
270d4406 DM |
224 | sub recompute_online_node_usage { |
225 | my ($self) = @_; | |
226 | ||
561e7f4b | 227 | my $haenv = $self->{haenv}; |
270d4406 DM |
228 | |
229 | my $online_nodes = $self->{ns}->list_online_nodes(); | |
230 | ||
561e7f4b FE |
231 | my $online_node_usage; |
232 | ||
1b813831 | 233 | if (my $mode = $self->{crs}->{scheduler}) { |
561e7f4b FE |
234 | if ($mode eq 'static') { |
235 | $online_node_usage = eval { | |
236 | my $scheduler = PVE::HA::Usage::Static->new($haenv); | |
237 | $scheduler->add_node($_) for $online_nodes->@*; | |
238 | return $scheduler; | |
239 | }; | |
f2c72982 | 240 | $haenv->log('warning', "fallback to 'basic' scheduler mode, init for 'static' failed - $@") |
561e7f4b | 241 | if $@; |
c2d8b56a TL |
242 | } elsif ($mode eq 'basic') { |
243 | # handled below in the general fall-back case | |
244 | } else { | |
561e7f4b FE |
245 | $haenv->log('warning', "got unknown scheduler mode '$mode', using 'basic'"); |
246 | } | |
247 | } | |
248 | ||
c2d8b56a | 249 | # fallback to the basic algorithm in any case |
561e7f4b FE |
250 | if (!$online_node_usage) { |
251 | $online_node_usage = PVE::HA::Usage::Basic->new($haenv); | |
252 | $online_node_usage->add_node($_) for $online_nodes->@*; | |
253 | } | |
270d4406 | 254 | |
7fd7af67 | 255 | foreach my $sid (sort keys %{$self->{ss}}) { |
270d4406 DM |
256 | my $sd = $self->{ss}->{$sid}; |
257 | my $state = $sd->{state}; | |
6f818da1 | 258 | my $target = $sd->{target}; # optional |
5d724d4d | 259 | if ($online_node_usage->contains_node($sd->{node})) { |
c259b1a8 | 260 | if ( |
4931b586 TL |
261 | $state eq 'started' || $state eq 'request_stop' || $state eq 'fence' |
262 | || $state eq 'freeze' || $state eq 'error' || $state eq 'recovery' | |
c259b1a8 | 263 | ) { |
5d724d4d | 264 | $online_node_usage->add_service_usage_to_node($sd->{node}, $sid, $sd->{node}); |
270d4406 | 265 | } elsif (($state eq 'migrate') || ($state eq 'relocate')) { |
5d724d4d | 266 | my $source = $sd->{node}; |
5c2eef4b | 267 | # count it for both, source and target as load is put on both |
5d724d4d FE |
268 | $online_node_usage->add_service_usage_to_node($source, $sid, $source, $target); |
269 | $online_node_usage->add_service_usage_to_node($target, $sid, $source, $target); | |
4931b586 | 270 | } elsif ($state eq 'stopped' || $state eq 'request_start') { |
270d4406 DM |
271 | # do nothing |
272 | } else { | |
feea3913 | 273 | die "should not be reached (sid = '$sid', state = '$state')"; |
270d4406 | 274 | } |
5d724d4d | 275 | } elsif (defined($target) && $online_node_usage->contains_node($target)) { |
066fd016 TL |
276 | if ($state eq 'migrate' || $state eq 'relocate') { |
277 | # to correctly track maintenance modi and also consider the target as used for the | |
278 | # case a node dies, as we cannot really know if the to-be-aborted incoming migration | |
279 | # has already cleaned up all used resources | |
5d724d4d | 280 | $online_node_usage->add_service_usage_to_node($target, $sid, $sd->{node}, $target); |
066fd016 | 281 | } |
270d4406 DM |
282 | } |
283 | } | |
284 | ||
285 | $self->{online_node_usage} = $online_node_usage; | |
286 | } | |
287 | ||
4e01bc86 DM |
288 | my $change_service_state = sub { |
289 | my ($self, $sid, $new_state, %params) = @_; | |
290 | ||
291 | my ($haenv, $ss) = ($self->{haenv}, $self->{ss}); | |
292 | ||
293 | my $sd = $ss->{$sid} || die "no such service '$sid"; | |
294 | ||
295 | my $old_state = $sd->{state}; | |
e4ffb299 | 296 | my $old_node = $sd->{node}; |
46139211 | 297 | my $old_failed_nodes = $sd->{failed_nodes}; |
2167dd1e | 298 | my $old_maintenance_node = $sd->{maintenance_node}; |
4e01bc86 DM |
299 | |
300 | die "no state change" if $old_state eq $new_state; # just to be sure | |
301 | ||
618fbeda DM |
302 | die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state}; |
303 | ||
e4ffb299 DM |
304 | foreach my $k (keys %$sd) { delete $sd->{$k}; }; |
305 | ||
306 | $sd->{state} = $new_state; | |
307 | $sd->{node} = $old_node; | |
ea998b07 | 308 | $sd->{failed_nodes} = $old_failed_nodes if defined($old_failed_nodes); |
2167dd1e | 309 | $sd->{maintenance_node} = $old_maintenance_node if defined($old_maintenance_node); |
e4ffb299 DM |
310 | |
311 | my $text_state = ''; | |
ba623362 | 312 | foreach my $k (sort keys %params) { |
4e01bc86 | 313 | my $v = $params{$k}; |
e4ffb299 DM |
314 | $text_state .= ", " if $text_state; |
315 | $text_state .= "$k = $v"; | |
4e01bc86 DM |
316 | $sd->{$k} = $v; |
317 | } | |
270d4406 DM |
318 | |
319 | $self->recompute_online_node_usage(); | |
320 | ||
d55aa611 | 321 | $sd->{uid} = compute_new_uuid($new_state); |
4e01bc86 | 322 | |
24678a59 TL |
323 | $text_state = " ($text_state)" if $text_state; |
324 | $haenv->log('info', "service '$sid': state changed from '${old_state}'" . | |
325 | " to '${new_state}'$text_state"); | |
4e01bc86 DM |
326 | }; |
327 | ||
5dd3ed86 TL |
328 | # clean up a possible bad state from a recovered service to allow its start |
329 | my $fence_recovery_cleanup = sub { | |
330 | my ($self, $sid, $fenced_node) = @_; | |
331 | ||
332 | my $haenv = $self->{haenv}; | |
333 | ||
0087839a | 334 | my (undef, $type, $id) = $haenv->parse_sid($sid); |
5dd3ed86 TL |
335 | my $plugin = PVE::HA::Resources->lookup($type); |
336 | ||
337 | # should not happen | |
338 | die "unknown resource type '$type'" if !$plugin; | |
339 | ||
32ea51dd TL |
340 | # locks may block recovery, cleanup those which are safe to remove after fencing, |
341 | # i.e., after the original node was reset and thus all it's state | |
3458a0e3 TL |
342 | my $removable_locks = [ |
343 | 'backup', | |
344 | 'mounted', | |
345 | 'migrate', | |
346 | 'clone', | |
347 | 'rollback', | |
348 | 'snapshot', | |
349 | 'snapshot-delete', | |
350 | 'suspending', | |
351 | 'suspended', | |
352 | ]; | |
5dd3ed86 TL |
353 | if (my $removed_lock = $plugin->remove_locks($haenv, $id, $removable_locks, $fenced_node)) { |
354 | $haenv->log('warning', "removed leftover lock '$removed_lock' from recovered " . | |
355 | "service '$sid' to allow its start."); | |
356 | } | |
357 | }; | |
358 | ||
289e4784 | 359 | # read LRM status for all nodes |
c4a221bc | 360 | sub read_lrm_status { |
332170bd | 361 | my ($self) = @_; |
c4a221bc | 362 | |
9c7d068b | 363 | my $nodes = $self->{ns}->list_nodes(); |
c4a221bc DM |
364 | my $haenv = $self->{haenv}; |
365 | ||
9c7d068b DM |
366 | my $results = {}; |
367 | my $modes = {}; | |
332170bd | 368 | foreach my $node (@$nodes) { |
9c7d068b | 369 | my $lrm_status = $haenv->read_lrm_status($node); |
02ffd753 | 370 | $modes->{$node} = $lrm_status->{mode} || 'active'; |
9c7d068b DM |
371 | foreach my $uid (keys %{$lrm_status->{results}}) { |
372 | next if $results->{$uid}; # should not happen | |
373 | $results->{$uid} = $lrm_status->{results}->{$uid}; | |
c4a221bc DM |
374 | } |
375 | } | |
376 | ||
9c7d068b | 377 | return ($results, $modes); |
c4a221bc DM |
378 | } |
379 | ||
aa98a844 DM |
380 | # read new crm commands and save them into crm master status |
381 | sub update_crm_commands { | |
382 | my ($self) = @_; | |
383 | ||
384 | my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss}); | |
385 | ||
386 | my $cmdlist = $haenv->read_crm_commands(); | |
bf7febe3 | 387 | |
aa98a844 DM |
388 | foreach my $cmd (split(/\n/, $cmdlist)) { |
389 | chomp $cmd; | |
390 | ||
b0fdf86a | 391 | if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) { |
289e4784 | 392 | my ($task, $sid, $node) = ($1, $2, $3); |
aa98a844 DM |
393 | if (my $sd = $ss->{$sid}) { |
394 | if (!$ns->node_is_online($node)) { | |
395 | $haenv->log('err', "crm command error - node not online: $cmd"); | |
396 | } else { | |
397 | if ($node eq $sd->{node}) { | |
398 | $haenv->log('info', "ignore crm command - service already on target node: $cmd"); | |
289e4784 | 399 | } else { |
aa98a844 | 400 | $haenv->log('info', "got crm command: $cmd"); |
3d42b01b | 401 | $ss->{$sid}->{cmd} = [ $task, $node ]; |
aa98a844 DM |
402 | } |
403 | } | |
404 | } else { | |
405 | $haenv->log('err', "crm command error - no such service: $cmd"); | |
406 | } | |
407 | ||
21caf0db FE |
408 | } elsif ($cmd =~ m/^stop\s+(\S+)\s+(\S+)$/) { |
409 | my ($sid, $timeout) = ($1, $2); | |
410 | if (my $sd = $ss->{$sid}) { | |
411 | $haenv->log('info', "got crm command: $cmd"); | |
412 | $ss->{$sid}->{cmd} = [ 'stop', $timeout ]; | |
413 | } else { | |
414 | $haenv->log('err', "crm command error - no such service: $cmd"); | |
415 | } | |
aa98a844 DM |
416 | } else { |
417 | $haenv->log('err', "unable to parse crm command: $cmd"); | |
418 | } | |
419 | } | |
420 | ||
421 | } | |
422 | ||
8f0bb968 DM |
423 | sub manage { |
424 | my ($self) = @_; | |
c0bbd038 | 425 | |
59fd7207 | 426 | my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss}); |
c0bbd038 | 427 | |
99278e06 TL |
428 | my ($node_info) = $haenv->get_node_info(); |
429 | my ($lrm_results, $lrm_modes) = $self->read_lrm_status(); | |
430 | ||
431 | $ns->update($node_info, $lrm_modes); | |
c79442f2 | 432 | |
99278e06 | 433 | if (!$ns->node_is_operational($haenv->nodename())) { |
e5986717 | 434 | $haenv->log('info', "master seems offline"); |
c79442f2 DM |
435 | return; |
436 | } | |
437 | ||
086f7075 TL |
438 | $self->update_crs_scheduler_mode(); |
439 | ||
f7ccd1b3 DM |
440 | my $sc = $haenv->read_service_config(); |
441 | ||
abc920b4 DM |
442 | $self->{groups} = $haenv->read_group_config(); # update |
443 | ||
f7ccd1b3 DM |
444 | # compute new service status |
445 | ||
446 | # add new service | |
cc32a8f3 | 447 | foreach my $sid (sort keys %$sc) { |
f7ccd1b3 | 448 | next if $ss->{$sid}; # already there |
77499288 | 449 | my $cd = $sc->{$sid}; |
667670b2 TL |
450 | next if $cd->{state} eq 'ignored'; |
451 | ||
77499288 | 452 | $haenv->log('info', "adding new service '$sid' on node '$cd->{node}'"); |
f7ccd1b3 | 453 | # assume we are running to avoid relocate running service at add |
c2f2b9c6 | 454 | my $state = ($cd->{state} eq 'started') ? 'request_start' : 'request_stop'; |
83a84eb0 TL |
455 | $ss->{$sid} = { |
456 | state => $state, node => $cd->{node}, uid => compute_new_uuid('started'), | |
457 | }; | |
f7ccd1b3 DM |
458 | } |
459 | ||
667670b2 | 460 | # remove stale or ignored services from manager state |
4e5764af | 461 | foreach my $sid (keys %$ss) { |
667670b2 TL |
462 | next if $sc->{$sid} && $sc->{$sid}->{state} ne 'ignored'; |
463 | ||
464 | my $reason = defined($sc->{$sid}) ? 'ignored state requested' : 'no config'; | |
465 | $haenv->log('info', "removing stale service '$sid' ($reason)"); | |
466 | ||
46139211 | 467 | # remove all service related state information |
4e5764af DM |
468 | delete $ss->{$sid}; |
469 | } | |
5a28da91 | 470 | |
aa98a844 DM |
471 | $self->update_crm_commands(); |
472 | ||
c79442f2 DM |
473 | for (;;) { |
474 | my $repeat = 0; | |
289e4784 | 475 | |
270d4406 | 476 | $self->recompute_online_node_usage(); |
f7ccd1b3 | 477 | |
a5e4bef4 | 478 | foreach my $sid (sort keys %$ss) { |
c79442f2 DM |
479 | my $sd = $ss->{$sid}; |
480 | my $cd = $sc->{$sid} || { state => 'disabled' }; | |
f7ccd1b3 | 481 | |
9c7d068b | 482 | my $lrm_res = $sd->{uid} ? $lrm_results->{$sd->{uid}} : undef; |
a875fbe8 | 483 | |
c79442f2 DM |
484 | my $last_state = $sd->{state}; |
485 | ||
486 | if ($last_state eq 'stopped') { | |
487 | ||
abc920b4 | 488 | $self->next_state_stopped($sid, $cd, $sd, $lrm_res); |
f7ccd1b3 | 489 | |
c79442f2 | 490 | } elsif ($last_state eq 'started') { |
f7ccd1b3 | 491 | |
abc920b4 | 492 | $self->next_state_started($sid, $cd, $sd, $lrm_res); |
f7ccd1b3 | 493 | |
4931b586 TL |
494 | } elsif ($last_state eq 'request_start') { |
495 | ||
496 | $self->next_state_request_start($sid, $cd, $sd, $lrm_res); | |
497 | ||
b0fdf86a | 498 | } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') { |
f7ccd1b3 | 499 | |
8aaa0e36 | 500 | $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res); |
f7ccd1b3 | 501 | |
c79442f2 | 502 | } elsif ($last_state eq 'fence') { |
f7ccd1b3 | 503 | |
21e37ed4 | 504 | # do nothing here - wait until fenced |
f7ccd1b3 | 505 | |
c259b1a8 TL |
506 | } elsif ($last_state eq 'recovery') { |
507 | ||
508 | $self->next_state_recovery($sid, $cd, $sd, $lrm_res); | |
509 | ||
c79442f2 | 510 | } elsif ($last_state eq 'request_stop') { |
f7ccd1b3 | 511 | |
0df5b3dd | 512 | $self->next_state_request_stop($sid, $cd, $sd, $lrm_res); |
618fbeda | 513 | |
9c7d068b DM |
514 | } elsif ($last_state eq 'freeze') { |
515 | ||
516 | my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef; | |
83a84eb0 TL |
517 | if ($lrm_mode && $lrm_mode eq 'active') { # unfreeze if active again |
518 | my $state = ($cd->{state} eq 'started') ? 'started' : 'request_stop'; | |
519 | $change_service_state->($self, $sid, $state); | |
520 | } | |
9c7d068b | 521 | |
e88469ba DM |
522 | } elsif ($last_state eq 'error') { |
523 | ||
a2881965 | 524 | $self->next_state_error($sid, $cd, $sd, $lrm_res); |
e88469ba | 525 | |
a875fbe8 DM |
526 | } else { |
527 | ||
528 | die "unknown service state '$last_state'"; | |
618fbeda | 529 | } |
21e37ed4 | 530 | |
9c7d068b | 531 | my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef; |
07adc6a6 | 532 | if ($lrm_mode && $lrm_mode eq 'restart') { |
83a84eb0 TL |
533 | my $state = $sd->{state}; |
534 | if ($state eq 'started' || $state eq 'stopped'|| $state eq 'request_stop') { | |
535 | $change_service_state->($self, $sid, 'freeze'); | |
07adc6a6 | 536 | } |
9c7d068b | 537 | } |
07adc6a6 | 538 | |
c79442f2 | 539 | $repeat = 1 if $sd->{state} ne $last_state; |
f7ccd1b3 DM |
540 | } |
541 | ||
21e37ed4 DM |
542 | # handle fencing |
543 | my $fenced_nodes = {}; | |
9b2dbc2a | 544 | foreach my $sid (sort keys %$ss) { |
2deff1ae TL |
545 | my ($service_state, $service_node) = $ss->{$sid}->@{'state', 'node'}; |
546 | next if $service_state ne 'fence'; | |
0dcb6597 TL |
547 | |
548 | if (!defined($fenced_nodes->{$service_node})) { | |
2deff1ae | 549 | $fenced_nodes->{$service_node} = $ns->fence_node($service_node) || 0; |
21e37ed4 DM |
550 | } |
551 | ||
0dcb6597 | 552 | next if !$fenced_nodes->{$service_node}; |
21e37ed4 | 553 | |
9da84a0d | 554 | # node fence was successful - recover service |
c259b1a8 | 555 | $change_service_state->($self, $sid, 'recovery'); |
0dcb6597 | 556 | $repeat = 1; # for faster recovery execution |
21e37ed4 DM |
557 | } |
558 | ||
2deff1ae TL |
559 | # Avoid that a node without services in 'fence' state (e.g., removed |
560 | # manually by admin) is stuck with the 'fence' node state. | |
561 | for my $node (sort grep { !defined($fenced_nodes->{$_}) } keys $ns->{status}->%*) { | |
7dc92703 | 562 | next if $ns->get_node_state($node) ne 'fence'; |
7dc92703 | 563 | |
2deff1ae TL |
564 | $haenv->log('notice', "node '$node' in fence state but no services to-fence! admin interference?!"); |
565 | $repeat = 1 if $ns->fence_node($node); | |
7dc92703 FE |
566 | } |
567 | ||
c79442f2 | 568 | last if !$repeat; |
f7ccd1b3 | 569 | } |
f7ccd1b3 | 570 | |
8f0bb968 | 571 | $self->flush_master_status(); |
c0bbd038 DM |
572 | } |
573 | ||
a875fbe8 DM |
574 | # functions to compute next service states |
575 | # $cd: service configuration data (read only) | |
576 | # $sd: service status data (read only) | |
577 | # | |
578 | # Note: use change_service_state() to alter state | |
579 | # | |
580 | ||
0df5b3dd DM |
581 | sub next_state_request_stop { |
582 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; | |
583 | ||
584 | my $haenv = $self->{haenv}; | |
585 | my $ns = $self->{ns}; | |
586 | ||
587 | # check result from LRM daemon | |
588 | if ($lrm_res) { | |
589 | my $exit_code = $lrm_res->{exit_code}; | |
a89ff919 | 590 | if ($exit_code == SUCCESS) { |
0df5b3dd DM |
591 | &$change_service_state($self, $sid, 'stopped'); |
592 | return; | |
593 | } else { | |
33f01524 | 594 | $haenv->log('err', "service '$sid' stop failed (exit code $exit_code)"); |
0df5b3dd DM |
595 | &$change_service_state($self, $sid, 'error'); # fixme: what state? |
596 | return; | |
597 | } | |
598 | } | |
599 | ||
ce3d7003 | 600 | if ($ns->node_is_offline_delayed($sd->{node})) { |
0df5b3dd DM |
601 | &$change_service_state($self, $sid, 'fence'); |
602 | return; | |
603 | } | |
604 | } | |
605 | ||
8aaa0e36 DM |
606 | sub next_state_migrate_relocate { |
607 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; | |
608 | ||
609 | my $haenv = $self->{haenv}; | |
610 | my $ns = $self->{ns}; | |
611 | ||
612 | # check result from LRM daemon | |
613 | if ($lrm_res) { | |
614 | my $exit_code = $lrm_res->{exit_code}; | |
bb07bd2c | 615 | my $req_state = $cd->{state} eq 'started' ? 'started' : 'request_stop'; |
a89ff919 | 616 | if ($exit_code == SUCCESS) { |
542a9902 | 617 | &$change_service_state($self, $sid, $req_state, node => $sd->{target}); |
8aaa0e36 | 618 | return; |
660596ce TL |
619 | } elsif ($exit_code == EWRONG_NODE) { |
620 | $haenv->log('err', "service '$sid' - migration failed: service" . | |
621 | " registered on wrong node!"); | |
622 | &$change_service_state($self, $sid, 'error'); | |
8aaa0e36 DM |
623 | } else { |
624 | $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)"); | |
542a9902 | 625 | &$change_service_state($self, $sid, $req_state, node => $sd->{node}); |
8aaa0e36 DM |
626 | return; |
627 | } | |
628 | } | |
629 | ||
ce3d7003 | 630 | if ($ns->node_is_offline_delayed($sd->{node})) { |
8aaa0e36 DM |
631 | &$change_service_state($self, $sid, 'fence'); |
632 | return; | |
633 | } | |
634 | } | |
635 | ||
a875fbe8 | 636 | sub next_state_stopped { |
abc920b4 | 637 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; |
a875fbe8 DM |
638 | |
639 | my $haenv = $self->{haenv}; | |
e88469ba | 640 | my $ns = $self->{ns}; |
a875fbe8 | 641 | |
ff6f1c5c DM |
642 | if ($sd->{node} ne $cd->{node}) { |
643 | # this can happen if we fence a node with active migrations | |
644 | # hack: modify $sd (normally this should be considered read-only) | |
24678a59 | 645 | $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node})"); |
289e4784 | 646 | $sd->{node} = $cd->{node}; |
ff6f1c5c DM |
647 | } |
648 | ||
94b7ebe2 | 649 | if ($sd->{cmd}) { |
21caf0db | 650 | my $cmd = shift @{$sd->{cmd}}; |
94b7ebe2 | 651 | |
b0fdf86a | 652 | if ($cmd eq 'migrate' || $cmd eq 'relocate') { |
21caf0db | 653 | my $target = shift @{$sd->{cmd}}; |
94b7ebe2 | 654 | if (!$ns->node_is_online($target)) { |
b0fdf86a | 655 | $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online"); |
e88469ba | 656 | } elsif ($sd->{node} eq $target) { |
b0fdf86a | 657 | $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'"); |
94b7ebe2 | 658 | } else { |
4931b586 | 659 | &$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target); |
9da84a0d | 660 | return; |
94b7ebe2 | 661 | } |
21caf0db FE |
662 | } elsif ($cmd eq 'stop') { |
663 | $haenv->log('info', "ignore service '$sid' $cmd request - service already stopped"); | |
94b7ebe2 | 664 | } else { |
289e4784 | 665 | $haenv->log('err', "unknown command '$cmd' for service '$sid'"); |
94b7ebe2 | 666 | } |
21caf0db | 667 | delete $sd->{cmd}; |
35cbb764 | 668 | } |
94b7ebe2 | 669 | |
a875fbe8 | 670 | if ($cd->{state} eq 'disabled') { |
35cbb764 TL |
671 | # NOTE: do nothing here, the stop state is an exception as we do not |
672 | # process the LRM result here, thus the LRM always tries to stop the | |
673 | # service (protection for the case no CRM is active) | |
e88469ba | 674 | return; |
35cbb764 | 675 | } |
e88469ba | 676 | |
84c945e4 | 677 | if ($ns->node_is_offline_delayed($sd->{node}) && $ns->get_node_state($sd->{node}) ne 'maintenance') { |
af14d5f3 TL |
678 | &$change_service_state($self, $sid, 'fence'); |
679 | return; | |
680 | } | |
681 | ||
682 | if ($cd->{state} eq 'stopped') { | |
683 | # almost the same as 'disabled' state but the service will also get recovered | |
684 | return; | |
685 | } | |
686 | ||
bb07bd2c | 687 | if ($cd->{state} eq 'started') { |
4931b586 TL |
688 | # simply mark it started, if it's on the wrong node next_state_started will fix that for us |
689 | $change_service_state->($self, $sid, 'request_start', node => $sd->{node}); | |
e88469ba | 690 | return; |
a875fbe8 | 691 | } |
e88469ba DM |
692 | |
693 | $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration"); | |
a875fbe8 DM |
694 | } |
695 | ||
4931b586 TL |
696 | sub next_state_request_start { |
697 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; | |
698 | ||
699 | $change_service_state->($self, $sid, 'started', node => $sd->{node}); | |
700 | } | |
701 | ||
46139211 | 702 | sub record_service_failed_on_node { |
57fe8e87 | 703 | my ($self, $sid, $node) = @_; |
46139211 | 704 | |
57fe8e87 DM |
705 | if (!defined($self->{ss}->{$sid}->{failed_nodes})) { |
706 | $self->{ss}->{$sid}->{failed_nodes} = []; | |
707 | } | |
46139211 | 708 | |
57fe8e87 | 709 | push @{$self->{ss}->{$sid}->{failed_nodes}}, $node; |
46139211 TL |
710 | } |
711 | ||
a875fbe8 | 712 | sub next_state_started { |
abc920b4 | 713 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; |
a875fbe8 DM |
714 | |
715 | my $haenv = $self->{haenv}; | |
ea4443cc | 716 | my $master_status = $self->{ms}; |
a875fbe8 DM |
717 | my $ns = $self->{ns}; |
718 | ||
719 | if (!$ns->node_is_online($sd->{node})) { | |
b0e9158d | 720 | if ($ns->node_is_offline_delayed($sd->{node})) { |
5385a606 DM |
721 | &$change_service_state($self, $sid, 'fence'); |
722 | } | |
99278e06 TL |
723 | if ($ns->get_node_state($sd->{node}) ne 'maintenance') { |
724 | return; | |
2167dd1e TL |
725 | } else { |
726 | # save current node as fallback for when it comes out of | |
727 | # maintenance | |
728 | $sd->{maintenance_node} = $sd->{node}; | |
99278e06 | 729 | } |
e88469ba | 730 | } |
289e4784 | 731 | |
af14d5f3 | 732 | if ($cd->{state} eq 'disabled' || $cd->{state} eq 'stopped') { |
e88469ba DM |
733 | &$change_service_state($self, $sid, 'request_stop'); |
734 | return; | |
735 | } | |
736 | ||
bb07bd2c | 737 | if ($cd->{state} eq 'started') { |
e88469ba DM |
738 | |
739 | if ($sd->{cmd}) { | |
21caf0db | 740 | my $cmd = shift @{$sd->{cmd}}; |
e88469ba | 741 | |
b0fdf86a | 742 | if ($cmd eq 'migrate' || $cmd eq 'relocate') { |
21caf0db | 743 | my $target = shift @{$sd->{cmd}}; |
e88469ba | 744 | if (!$ns->node_is_online($target)) { |
b0fdf86a | 745 | $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online"); |
e88469ba | 746 | } elsif ($sd->{node} eq $target) { |
b0fdf86a | 747 | $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'"); |
e88469ba | 748 | } else { |
a3cb8dcb | 749 | $haenv->log('info', "$cmd service '$sid' to node '$target'"); |
b0fdf86a | 750 | &$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target); |
e88469ba | 751 | } |
21caf0db FE |
752 | } elsif ($cmd eq 'stop') { |
753 | my $timeout = shift @{$sd->{cmd}}; | |
396eb6f0 TL |
754 | if ($timeout == 0) { |
755 | $haenv->log('info', "request immediate service hard-stop for service '$sid'"); | |
756 | } else { | |
757 | $haenv->log('info', "request graceful stop with timeout '$timeout' for service '$sid'"); | |
758 | } | |
21caf0db FE |
759 | &$change_service_state($self, $sid, 'request_stop', timeout => $timeout); |
760 | $haenv->update_service_config($sid, {'state' => 'stopped'}); | |
a875fbe8 | 761 | } else { |
289e4784 | 762 | $haenv->log('err', "unknown command '$cmd' for service '$sid'"); |
a875fbe8 | 763 | } |
21caf0db FE |
764 | |
765 | delete $sd->{cmd}; | |
766 | ||
a875fbe8 | 767 | } else { |
b0fdf86a | 768 | |
abc920b4 | 769 | my $try_next = 0; |
46139211 | 770 | |
ea4443cc | 771 | if ($lrm_res) { |
46139211 | 772 | |
e9e1cd68 TL |
773 | my $ec = $lrm_res->{exit_code}; |
774 | if ($ec == SUCCESS) { | |
775 | ||
46139211 | 776 | if (defined($sd->{failed_nodes})) { |
81449997 | 777 | $haenv->log('info', "relocation policy successful for '$sid' on node '$sd->{node}'," . |
46139211 TL |
778 | " failed nodes: " . join(', ', @{$sd->{failed_nodes}}) ); |
779 | } | |
780 | ||
781 | delete $sd->{failed_nodes}; | |
e9e1cd68 | 782 | |
b47920fd DM |
783 | # store flag to indicate successful start - only valid while state == 'started' |
784 | $sd->{running} = 1; | |
785 | ||
b159176a | 786 | } elsif ($ec == ERROR || $ec == EWRONG_NODE) { |
b47920fd DM |
787 | |
788 | delete $sd->{running}; | |
789 | ||
e9e1cd68 | 790 | # apply our relocate policy if we got ERROR from the LRM |
46139211 | 791 | $self->record_service_failed_on_node($sid, $sd->{node}); |
ea4443cc | 792 | |
46139211 | 793 | if (scalar(@{$sd->{failed_nodes}}) <= $cd->{max_relocate}) { |
ea4443cc | 794 | |
e9e1cd68 TL |
795 | # tell select_service_node to relocate if possible |
796 | $try_next = 1; | |
ea4443cc TL |
797 | |
798 | $haenv->log('warning', "starting service $sid on node". | |
799 | " '$sd->{node}' failed, relocating service."); | |
ea4443cc TL |
800 | |
801 | } else { | |
802 | ||
46139211 TL |
803 | $haenv->log('err', "recovery policy for service $sid " . |
804 | "failed, entering error state. Failed nodes: ". | |
805 | join(', ', @{$sd->{failed_nodes}})); | |
ea4443cc TL |
806 | &$change_service_state($self, $sid, 'error'); |
807 | return; | |
808 | ||
809 | } | |
e9e1cd68 | 810 | } else { |
46139211 TL |
811 | $self->record_service_failed_on_node($sid, $sd->{node}); |
812 | ||
e9e1cd68 TL |
813 | $haenv->log('err', "service '$sid' got unrecoverable error" . |
814 | " (exit code $ec))"); | |
815 | # we have no save way out (yet) for other errors | |
816 | &$change_service_state($self, $sid, 'error'); | |
35cbb764 | 817 | return; |
ea4443cc | 818 | } |
abc920b4 DM |
819 | } |
820 | ||
2167dd1e TL |
821 | my $node = select_service_node( |
822 | $self->{groups}, | |
823 | $self->{online_node_usage}, | |
b2598576 | 824 | $sid, |
2167dd1e TL |
825 | $cd, |
826 | $sd->{node}, | |
827 | $try_next, | |
828 | $sd->{failed_nodes}, | |
829 | $sd->{maintenance_node}, | |
830 | ); | |
abc920b4 | 831 | |
b0fdf86a | 832 | if ($node && ($sd->{node} ne $node)) { |
5d724d4d | 833 | $self->{online_node_usage}->add_service_usage_to_node($node, $sid, $sd->{node}); |
2167dd1e TL |
834 | |
835 | if (defined(my $fallback = $sd->{maintenance_node})) { | |
836 | if ($node eq $fallback) { | |
837 | $haenv->log('info', "moving service '$sid' back to '$fallback', node came back from maintenance."); | |
838 | delete $sd->{maintenance_node}; | |
839 | } elsif ($sd->{node} ne $fallback) { | |
840 | $haenv->log('info', "dropping maintenance fallback node '$fallback' for '$sid'"); | |
841 | delete $sd->{maintenance_node}; | |
842 | } | |
843 | } | |
844 | ||
c0255b2c TL |
845 | if ($cd->{type} eq 'vm') { |
846 | $haenv->log('info', "migrate service '$sid' to node '$node' (running)"); | |
847 | &$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node); | |
848 | } else { | |
849 | $haenv->log('info', "relocate service '$sid' to node '$node'"); | |
850 | &$change_service_state($self, $sid, 'relocate', node => $sd->{node}, target => $node); | |
851 | } | |
b0fdf86a | 852 | } else { |
e6eeb7dc TL |
853 | if ($try_next && !defined($node)) { |
854 | $haenv->log('warning', "Start Error Recovery: Tried all available " . | |
855 | " nodes for service '$sid', retry start on current node. " . | |
856 | "Tried nodes: " . join(', ', @{$sd->{failed_nodes}})); | |
857 | } | |
35cbb764 | 858 | # ensure service get started again if it went unexpected down |
bf2d8d74 TL |
859 | # but ensure also no LRM result gets lost |
860 | $sd->{uid} = compute_new_uuid($sd->{state}) if defined($lrm_res); | |
b0fdf86a | 861 | } |
a875fbe8 | 862 | } |
e88469ba DM |
863 | |
864 | return; | |
35cbb764 | 865 | } |
e88469ba DM |
866 | |
867 | $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration"); | |
a875fbe8 | 868 | } |
c0bbd038 | 869 | |
a2881965 TL |
870 | sub next_state_error { |
871 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; | |
872 | ||
873 | my $ns = $self->{ns}; | |
46139211 | 874 | my $ms = $self->{ms}; |
a2881965 TL |
875 | |
876 | if ($cd->{state} eq 'disabled') { | |
46139211 TL |
877 | # clean up on error recovery |
878 | delete $sd->{failed_nodes}; | |
879 | ||
a2881965 TL |
880 | &$change_service_state($self, $sid, 'stopped'); |
881 | return; | |
882 | } | |
883 | ||
a2881965 TL |
884 | } |
885 | ||
c259b1a8 TL |
886 | # after a node was fenced this recovers the service to a new node |
887 | sub next_state_recovery { | |
888 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; | |
889 | ||
890 | my ($haenv, $ss) = ($self->{haenv}, $self->{ss}); | |
891 | my $ns = $self->{ns}; | |
892 | my $ms = $self->{ms}; | |
893 | ||
894 | if ($sd->{state} ne 'recovery') { # should not happen | |
895 | $haenv->log('err', "cannot recover service '$sid' from fencing, wrong state '$sd->{state}'"); | |
896 | return; | |
897 | } | |
898 | ||
899 | my $fenced_node = $sd->{node}; # for logging purpose | |
900 | ||
901 | $self->recompute_online_node_usage(); # we want the most current node state | |
902 | ||
903 | my $recovery_node = select_service_node( | |
904 | $self->{groups}, | |
905 | $self->{online_node_usage}, | |
b2598576 | 906 | $sid, |
c259b1a8 TL |
907 | $cd, |
908 | $sd->{node}, | |
909 | ); | |
910 | ||
911 | if ($recovery_node) { | |
90a24755 TL |
912 | my $msg = "recover service '$sid' from fenced node '$fenced_node' to node '$recovery_node'"; |
913 | if ($recovery_node eq $fenced_node) { | |
914 | # can happen if restriced groups and the node came up again OK | |
915 | $msg = "recover service '$sid' to previous failed and fenced node '$fenced_node' again"; | |
916 | } | |
917 | $haenv->log('info', "$msg"); | |
c259b1a8 TL |
918 | |
919 | $fence_recovery_cleanup->($self, $sid, $fenced_node); | |
920 | ||
921 | $haenv->steal_service($sid, $sd->{node}, $recovery_node); | |
5d724d4d | 922 | $self->{online_node_usage}->add_service_usage_to_node($recovery_node, $sid, $recovery_node); |
c259b1a8 TL |
923 | |
924 | # NOTE: $sd *is normally read-only*, fencing is the exception | |
925 | $cd->{node} = $sd->{node} = $recovery_node; | |
926 | my $new_state = ($cd->{state} eq 'started') ? 'started' : 'request_stop'; | |
927 | $change_service_state->($self, $sid, $new_state, node => $recovery_node); | |
928 | } else { | |
929 | # no possible node found, cannot recover - but retry later, as we always try to make it available | |
930 | $haenv->log('err', "recovering service '$sid' from fenced node '$fenced_node' failed, no recovery node found"); | |
719883e9 TL |
931 | |
932 | if ($cd->{state} eq 'disabled') { | |
933 | # allow getting a service out of recovery manually if an admin disables it. | |
934 | delete $sd->{failed_nodes}; # clean up on recovery to stopped | |
935 | $change_service_state->($self, $sid, 'stopped'); # must NOT go through request_stop | |
936 | return; | |
937 | } | |
c259b1a8 TL |
938 | } |
939 | } | |
940 | ||
c0bbd038 | 941 | 1; |