]>
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 DM |
9 | use PVE::HA::NodeStatus; |
10 | ||
11 | sub new { | |
8f0bb968 | 12 | my ($this, $haenv) = @_; |
c0bbd038 DM |
13 | |
14 | my $class = ref($this) || $this; | |
15 | ||
6ee64cfc | 16 | my $self = bless { haenv => $haenv }, $class; |
8f0bb968 | 17 | |
6ee64cfc | 18 | my $old_ms = $haenv->read_manager_status(); |
8f0bb968 | 19 | |
6ee64cfc TL |
20 | # we only copy the state part of the manager which cannot be auto generated |
21 | ||
22 | $self->{ns} = PVE::HA::NodeStatus->new($haenv, $old_ms->{node_status} || {}); | |
8f0bb968 | 23 | |
59fd7207 | 24 | # fixme: use separate class PVE::HA::ServiceStatus |
6ee64cfc TL |
25 | $self->{ss} = $old_ms->{service_status} || {}; |
26 | ||
27 | $self->{ms} = { master_node => $haenv->nodename() }; | |
c0bbd038 DM |
28 | |
29 | return $self; | |
30 | } | |
31 | ||
d84da043 DM |
32 | sub cleanup { |
33 | my ($self) = @_; | |
34 | ||
35 | # todo: ? | |
36 | } | |
37 | ||
8f0bb968 | 38 | sub flush_master_status { |
c0bbd038 DM |
39 | my ($self) = @_; |
40 | ||
59fd7207 | 41 | my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss}); |
c0bbd038 | 42 | |
8f0bb968 | 43 | $ms->{node_status} = $ns->{status}; |
59fd7207 | 44 | $ms->{service_status} = $ss; |
d2f612cf | 45 | $ms->{timestamp} = $haenv->get_time(); |
289e4784 | 46 | |
8f0bb968 | 47 | $haenv->write_manager_status($ms); |
289e4784 | 48 | } |
c0bbd038 | 49 | |
48a6ba2a TL |
50 | sub get_service_group { |
51 | my ($groups, $online_node_usage, $service_conf) = @_; | |
f7ccd1b3 | 52 | |
09c5c4bf TL |
53 | my $group = {}; |
54 | # add all online nodes to default group to allow try_next when no group set | |
55 | foreach my $node (keys %$online_node_usage) { | |
56 | $group->{nodes}->{$node} = 1; | |
57 | } | |
abc920b4 | 58 | |
09c5c4bf | 59 | # overwrite default if service is bound to a specific group |
289e4784 | 60 | $group = $groups->{ids}->{$service_conf->{group}} if $service_conf->{group} && |
270d4406 | 61 | $groups->{ids}->{$service_conf->{group}}; |
abc920b4 | 62 | |
48a6ba2a TL |
63 | return $group; |
64 | } | |
65 | ||
66 | # groups available nodes with their priority as group index | |
67 | sub get_node_priority_groups { | |
68 | my ($group, $online_node_usage) = @_; | |
69 | ||
abc920b4 DM |
70 | my $pri_groups = {}; |
71 | my $group_members = {}; | |
e0a56314 | 72 | foreach my $entry (keys %{$group->{nodes}}) { |
abc920b4 DM |
73 | my ($node, $pri) = ($entry, 0); |
74 | if ($entry =~ m/^(\S+):(\d+)$/) { | |
75 | ($node, $pri) = ($1, $2); | |
76 | } | |
270d4406 | 77 | next if !defined($online_node_usage->{$node}); # offline |
abc920b4 DM |
78 | $pri_groups->{$pri}->{$node} = 1; |
79 | $group_members->{$node} = $pri; | |
80 | } | |
f7ccd1b3 | 81 | |
abc920b4 DM |
82 | # add non-group members to unrestricted groups (priority -1) |
83 | if (!$group->{restricted}) { | |
84 | my $pri = -1; | |
270d4406 | 85 | foreach my $node (keys %$online_node_usage) { |
abc920b4 DM |
86 | next if defined($group_members->{$node}); |
87 | $pri_groups->{$pri}->{$node} = 1; | |
88 | $group_members->{$node} = -1; | |
89 | } | |
90 | } | |
91 | ||
48a6ba2a TL |
92 | return ($pri_groups, $group_members); |
93 | } | |
94 | ||
95 | sub select_service_node { | |
96 | my ($groups, $online_node_usage, $service_conf, $current_node, $try_next, $tried_nodes) = @_; | |
97 | ||
98 | my $group = get_service_group($groups, $online_node_usage, $service_conf); | |
99 | ||
100 | my ($pri_groups, $group_members) = get_node_priority_groups($group, $online_node_usage); | |
101 | ||
abc920b4 DM |
102 | my @pri_list = sort {$b <=> $a} keys %$pri_groups; |
103 | return undef if !scalar(@pri_list); | |
09c5c4bf TL |
104 | |
105 | # stay on current node if possible (avoids random migrations) | |
abc920b4 DM |
106 | if (!$try_next && $group->{nofailback} && defined($group_members->{$current_node})) { |
107 | return $current_node; | |
108 | } | |
109 | ||
110 | # select node from top priority node list | |
111 | ||
112 | my $top_pri = $pri_list[0]; | |
113 | ||
e6eeb7dc TL |
114 | # try to avoid nodes where the service failed already if we want to relocate |
115 | if ($try_next) { | |
116 | foreach my $node (@$tried_nodes) { | |
117 | delete $pri_groups->{$top_pri}->{$node}; | |
118 | } | |
119 | } | |
120 | ||
289e4784 | 121 | my @nodes = sort { |
c142ebc9 DM |
122 | $online_node_usage->{$a} <=> $online_node_usage->{$b} || $a cmp $b |
123 | } keys %{$pri_groups->{$top_pri}}; | |
abc920b4 DM |
124 | |
125 | my $found; | |
126 | for (my $i = scalar(@nodes) - 1; $i >= 0; $i--) { | |
127 | my $node = $nodes[$i]; | |
128 | if ($node eq $current_node) { | |
129 | $found = $i; | |
130 | last; | |
131 | } | |
132 | } | |
133 | ||
abc920b4 DM |
134 | if ($try_next) { |
135 | ||
136 | if (defined($found) && ($found < (scalar(@nodes) - 1))) { | |
137 | return $nodes[$found + 1]; | |
138 | } else { | |
139 | return $nodes[0]; | |
140 | } | |
141 | ||
142 | } else { | |
143 | ||
144 | return $nodes[$found] if defined($found); | |
145 | ||
146 | return $nodes[0]; | |
147 | ||
148 | } | |
f7ccd1b3 DM |
149 | } |
150 | ||
c4a221bc DM |
151 | my $uid_counter = 0; |
152 | ||
d55aa611 DM |
153 | sub compute_new_uuid { |
154 | my ($state) = @_; | |
289e4784 | 155 | |
d55aa611 DM |
156 | $uid_counter++; |
157 | return md5_base64($state . $$ . time() . $uid_counter); | |
158 | } | |
159 | ||
618fbeda DM |
160 | my $valid_service_states = { |
161 | stopped => 1, | |
162 | request_stop => 1, | |
163 | started => 1, | |
164 | fence => 1, | |
165 | migrate => 1, | |
b0fdf86a | 166 | relocate => 1, |
9c7d068b | 167 | freeze => 1, |
618fbeda DM |
168 | error => 1, |
169 | }; | |
170 | ||
270d4406 DM |
171 | sub recompute_online_node_usage { |
172 | my ($self) = @_; | |
173 | ||
174 | my $online_node_usage = {}; | |
175 | ||
176 | my $online_nodes = $self->{ns}->list_online_nodes(); | |
177 | ||
178 | foreach my $node (@$online_nodes) { | |
179 | $online_node_usage->{$node} = 0; | |
180 | } | |
181 | ||
182 | foreach my $sid (keys %{$self->{ss}}) { | |
183 | my $sd = $self->{ss}->{$sid}; | |
184 | my $state = $sd->{state}; | |
185 | if (defined($online_node_usage->{$sd->{node}})) { | |
289e4784 | 186 | if (($state eq 'started') || ($state eq 'request_stop') || |
9c7d068b | 187 | ($state eq 'fence') || ($state eq 'freeze') || ($state eq 'error')) { |
270d4406 DM |
188 | $online_node_usage->{$sd->{node}}++; |
189 | } elsif (($state eq 'migrate') || ($state eq 'relocate')) { | |
190 | $online_node_usage->{$sd->{target}}++; | |
191 | } elsif ($state eq 'stopped') { | |
192 | # do nothing | |
193 | } else { | |
194 | die "should not be reached"; | |
195 | } | |
196 | } | |
197 | } | |
198 | ||
199 | $self->{online_node_usage} = $online_node_usage; | |
200 | } | |
201 | ||
4e01bc86 DM |
202 | my $change_service_state = sub { |
203 | my ($self, $sid, $new_state, %params) = @_; | |
204 | ||
205 | my ($haenv, $ss) = ($self->{haenv}, $self->{ss}); | |
206 | ||
207 | my $sd = $ss->{$sid} || die "no such service '$sid"; | |
208 | ||
209 | my $old_state = $sd->{state}; | |
e4ffb299 | 210 | my $old_node = $sd->{node}; |
46139211 | 211 | my $old_failed_nodes = $sd->{failed_nodes}; |
4e01bc86 DM |
212 | |
213 | die "no state change" if $old_state eq $new_state; # just to be sure | |
214 | ||
618fbeda DM |
215 | die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state}; |
216 | ||
e4ffb299 DM |
217 | foreach my $k (keys %$sd) { delete $sd->{$k}; }; |
218 | ||
219 | $sd->{state} = $new_state; | |
220 | $sd->{node} = $old_node; | |
ea998b07 | 221 | $sd->{failed_nodes} = $old_failed_nodes if defined($old_failed_nodes); |
e4ffb299 DM |
222 | |
223 | my $text_state = ''; | |
ba623362 | 224 | foreach my $k (sort keys %params) { |
4e01bc86 | 225 | my $v = $params{$k}; |
e4ffb299 DM |
226 | $text_state .= ", " if $text_state; |
227 | $text_state .= "$k = $v"; | |
4e01bc86 DM |
228 | $sd->{$k} = $v; |
229 | } | |
270d4406 DM |
230 | |
231 | $self->recompute_online_node_usage(); | |
232 | ||
d55aa611 | 233 | $sd->{uid} = compute_new_uuid($new_state); |
4e01bc86 | 234 | |
24678a59 TL |
235 | $text_state = " ($text_state)" if $text_state; |
236 | $haenv->log('info', "service '$sid': state changed from '${old_state}'" . | |
237 | " to '${new_state}'$text_state"); | |
4e01bc86 DM |
238 | }; |
239 | ||
5dd3ed86 TL |
240 | # clean up a possible bad state from a recovered service to allow its start |
241 | my $fence_recovery_cleanup = sub { | |
242 | my ($self, $sid, $fenced_node) = @_; | |
243 | ||
244 | my $haenv = $self->{haenv}; | |
245 | ||
0087839a | 246 | my (undef, $type, $id) = $haenv->parse_sid($sid); |
5dd3ed86 TL |
247 | my $plugin = PVE::HA::Resources->lookup($type); |
248 | ||
249 | # should not happen | |
250 | die "unknown resource type '$type'" if !$plugin; | |
251 | ||
252 | # locks may block recovery, cleanup those which are safe to remove after fencing | |
253 | my $removable_locks = ['backup', 'mounted']; | |
254 | if (my $removed_lock = $plugin->remove_locks($haenv, $id, $removable_locks, $fenced_node)) { | |
255 | $haenv->log('warning', "removed leftover lock '$removed_lock' from recovered " . | |
256 | "service '$sid' to allow its start."); | |
257 | } | |
258 | }; | |
259 | ||
9da84a0d TL |
260 | # after a node was fenced this recovers the service to a new node |
261 | my $recover_fenced_service = sub { | |
262 | my ($self, $sid, $cd) = @_; | |
263 | ||
264 | my ($haenv, $ss) = ($self->{haenv}, $self->{ss}); | |
265 | ||
266 | my $sd = $ss->{$sid}; | |
267 | ||
268 | if ($sd->{state} ne 'fence') { # should not happen | |
269 | $haenv->log('err', "cannot recover service '$sid' from fencing," . | |
270 | " wrong state '$sd->{state}'"); | |
271 | return; | |
272 | } | |
273 | ||
274 | my $fenced_node = $sd->{node}; # for logging purpose | |
275 | ||
276 | $self->recompute_online_node_usage(); # we want the most current node state | |
277 | ||
278 | my $recovery_node = select_service_node($self->{groups}, | |
279 | $self->{online_node_usage}, | |
280 | $cd, $sd->{node}); | |
281 | ||
282 | if ($recovery_node) { | |
283 | $haenv->log('info', "recover service '$sid' from fenced node " . | |
284 | "'$fenced_node' to node '$recovery_node'"); | |
285 | ||
5dd3ed86 TL |
286 | &$fence_recovery_cleanup($self, $sid, $fenced_node); |
287 | ||
9da84a0d TL |
288 | $haenv->steal_service($sid, $sd->{node}, $recovery_node); |
289 | ||
290 | # $sd *is normally read-only*, fencing is the exception | |
291 | $cd->{node} = $sd->{node} = $recovery_node; | |
bb07bd2c | 292 | my $new_state = ($cd->{state} eq 'started') ? 'started' : 'request_stop'; |
af14d5f3 | 293 | &$change_service_state($self, $sid, $new_state, node => $recovery_node); |
9da84a0d | 294 | } else { |
08aa53c2 | 295 | # no possible node found, cannot recover |
9da84a0d TL |
296 | $haenv->log('err', "recovering service '$sid' from fenced node " . |
297 | "'$fenced_node' failed, no recovery node found"); | |
08aa53c2 | 298 | &$change_service_state($self, $sid, 'error'); |
9da84a0d TL |
299 | } |
300 | }; | |
301 | ||
289e4784 | 302 | # read LRM status for all nodes |
c4a221bc | 303 | sub read_lrm_status { |
332170bd | 304 | my ($self) = @_; |
c4a221bc | 305 | |
9c7d068b | 306 | my $nodes = $self->{ns}->list_nodes(); |
c4a221bc DM |
307 | my $haenv = $self->{haenv}; |
308 | ||
9c7d068b DM |
309 | my $results = {}; |
310 | my $modes = {}; | |
332170bd | 311 | foreach my $node (@$nodes) { |
9c7d068b | 312 | my $lrm_status = $haenv->read_lrm_status($node); |
02ffd753 | 313 | $modes->{$node} = $lrm_status->{mode} || 'active'; |
9c7d068b DM |
314 | foreach my $uid (keys %{$lrm_status->{results}}) { |
315 | next if $results->{$uid}; # should not happen | |
316 | $results->{$uid} = $lrm_status->{results}->{$uid}; | |
c4a221bc DM |
317 | } |
318 | } | |
319 | ||
289e4784 | 320 | |
9c7d068b | 321 | return ($results, $modes); |
c4a221bc DM |
322 | } |
323 | ||
aa98a844 DM |
324 | # read new crm commands and save them into crm master status |
325 | sub update_crm_commands { | |
326 | my ($self) = @_; | |
327 | ||
328 | my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss}); | |
329 | ||
330 | my $cmdlist = $haenv->read_crm_commands(); | |
bf7febe3 | 331 | |
aa98a844 DM |
332 | foreach my $cmd (split(/\n/, $cmdlist)) { |
333 | chomp $cmd; | |
334 | ||
b0fdf86a | 335 | if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) { |
289e4784 | 336 | my ($task, $sid, $node) = ($1, $2, $3); |
aa98a844 DM |
337 | if (my $sd = $ss->{$sid}) { |
338 | if (!$ns->node_is_online($node)) { | |
339 | $haenv->log('err', "crm command error - node not online: $cmd"); | |
340 | } else { | |
341 | if ($node eq $sd->{node}) { | |
342 | $haenv->log('info', "ignore crm command - service already on target node: $cmd"); | |
289e4784 | 343 | } else { |
aa98a844 | 344 | $haenv->log('info', "got crm command: $cmd"); |
3d42b01b | 345 | $ss->{$sid}->{cmd} = [ $task, $node ]; |
aa98a844 DM |
346 | } |
347 | } | |
348 | } else { | |
349 | $haenv->log('err', "crm command error - no such service: $cmd"); | |
350 | } | |
351 | ||
352 | } else { | |
353 | $haenv->log('err', "unable to parse crm command: $cmd"); | |
354 | } | |
355 | } | |
356 | ||
357 | } | |
358 | ||
8f0bb968 DM |
359 | sub manage { |
360 | my ($self) = @_; | |
c0bbd038 | 361 | |
59fd7207 | 362 | my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss}); |
c0bbd038 | 363 | |
332170bd | 364 | $ns->update($haenv->get_node_info()); |
c79442f2 DM |
365 | |
366 | if (!$ns->node_is_online($haenv->nodename())) { | |
e5986717 | 367 | $haenv->log('info', "master seems offline"); |
c79442f2 DM |
368 | return; |
369 | } | |
370 | ||
9c7d068b | 371 | my ($lrm_results, $lrm_modes) = $self->read_lrm_status(); |
c4a221bc | 372 | |
f7ccd1b3 DM |
373 | my $sc = $haenv->read_service_config(); |
374 | ||
abc920b4 DM |
375 | $self->{groups} = $haenv->read_group_config(); # update |
376 | ||
f7ccd1b3 DM |
377 | # compute new service status |
378 | ||
379 | # add new service | |
cc32a8f3 | 380 | foreach my $sid (sort keys %$sc) { |
f7ccd1b3 | 381 | next if $ss->{$sid}; # already there |
77499288 | 382 | my $cd = $sc->{$sid}; |
667670b2 TL |
383 | next if $cd->{state} eq 'ignored'; |
384 | ||
77499288 | 385 | $haenv->log('info', "adding new service '$sid' on node '$cd->{node}'"); |
f7ccd1b3 | 386 | # assume we are running to avoid relocate running service at add |
bb07bd2c | 387 | my $state = ($cd->{state} eq 'started') ? 'started' : 'request_stop'; |
77499288 | 388 | $ss->{$sid} = { state => $state, node => $cd->{node}, |
d55aa611 | 389 | uid => compute_new_uuid('started') }; |
f7ccd1b3 DM |
390 | } |
391 | ||
667670b2 | 392 | # remove stale or ignored services from manager state |
4e5764af | 393 | foreach my $sid (keys %$ss) { |
667670b2 TL |
394 | next if $sc->{$sid} && $sc->{$sid}->{state} ne 'ignored'; |
395 | ||
396 | my $reason = defined($sc->{$sid}) ? 'ignored state requested' : 'no config'; | |
397 | $haenv->log('info', "removing stale service '$sid' ($reason)"); | |
398 | ||
46139211 | 399 | # remove all service related state information |
4e5764af DM |
400 | delete $ss->{$sid}; |
401 | } | |
5a28da91 | 402 | |
aa98a844 DM |
403 | $self->update_crm_commands(); |
404 | ||
c79442f2 DM |
405 | for (;;) { |
406 | my $repeat = 0; | |
289e4784 | 407 | |
270d4406 | 408 | $self->recompute_online_node_usage(); |
f7ccd1b3 | 409 | |
a5e4bef4 | 410 | foreach my $sid (sort keys %$ss) { |
c79442f2 DM |
411 | my $sd = $ss->{$sid}; |
412 | my $cd = $sc->{$sid} || { state => 'disabled' }; | |
f7ccd1b3 | 413 | |
9c7d068b | 414 | my $lrm_res = $sd->{uid} ? $lrm_results->{$sd->{uid}} : undef; |
a875fbe8 | 415 | |
c79442f2 DM |
416 | my $last_state = $sd->{state}; |
417 | ||
418 | if ($last_state eq 'stopped') { | |
419 | ||
abc920b4 | 420 | $self->next_state_stopped($sid, $cd, $sd, $lrm_res); |
f7ccd1b3 | 421 | |
c79442f2 | 422 | } elsif ($last_state eq 'started') { |
f7ccd1b3 | 423 | |
abc920b4 | 424 | $self->next_state_started($sid, $cd, $sd, $lrm_res); |
f7ccd1b3 | 425 | |
b0fdf86a | 426 | } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') { |
f7ccd1b3 | 427 | |
8aaa0e36 | 428 | $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res); |
f7ccd1b3 | 429 | |
c79442f2 | 430 | } elsif ($last_state eq 'fence') { |
f7ccd1b3 | 431 | |
21e37ed4 | 432 | # do nothing here - wait until fenced |
f7ccd1b3 | 433 | |
c79442f2 | 434 | } elsif ($last_state eq 'request_stop') { |
f7ccd1b3 | 435 | |
0df5b3dd | 436 | $self->next_state_request_stop($sid, $cd, $sd, $lrm_res); |
618fbeda | 437 | |
9c7d068b DM |
438 | } elsif ($last_state eq 'freeze') { |
439 | ||
440 | my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef; | |
9c7d068b | 441 | # unfreeze |
bb07bd2c | 442 | my $state = ($cd->{state} eq 'started') ? 'started' : 'request_stop'; |
af14d5f3 | 443 | &$change_service_state($self, $sid, $state) |
02ffd753 | 444 | if $lrm_mode && $lrm_mode eq 'active'; |
9c7d068b | 445 | |
e88469ba DM |
446 | } elsif ($last_state eq 'error') { |
447 | ||
a2881965 | 448 | $self->next_state_error($sid, $cd, $sd, $lrm_res); |
e88469ba | 449 | |
a875fbe8 DM |
450 | } else { |
451 | ||
452 | die "unknown service state '$last_state'"; | |
618fbeda | 453 | } |
21e37ed4 | 454 | |
9c7d068b | 455 | my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef; |
07adc6a6 DM |
456 | if ($lrm_mode && $lrm_mode eq 'restart') { |
457 | if (($sd->{state} eq 'started' || $sd->{state} eq 'stopped' || | |
458 | $sd->{state} eq 'request_stop')) { | |
459 | &$change_service_state($self, $sid, 'freeze'); | |
460 | } | |
9c7d068b | 461 | } |
07adc6a6 | 462 | |
c79442f2 | 463 | $repeat = 1 if $sd->{state} ne $last_state; |
f7ccd1b3 DM |
464 | } |
465 | ||
21e37ed4 DM |
466 | # handle fencing |
467 | my $fenced_nodes = {}; | |
9b2dbc2a | 468 | foreach my $sid (sort keys %$ss) { |
21e37ed4 DM |
469 | my $sd = $ss->{$sid}; |
470 | next if $sd->{state} ne 'fence'; | |
471 | ||
472 | if (!defined($fenced_nodes->{$sd->{node}})) { | |
473 | $fenced_nodes->{$sd->{node}} = $ns->fence_node($sd->{node}) || 0; | |
474 | } | |
475 | ||
476 | next if !$fenced_nodes->{$sd->{node}}; | |
477 | ||
9da84a0d TL |
478 | # node fence was successful - recover service |
479 | &$recover_fenced_service($self, $sid, $sc->{$sid}); | |
21e37ed4 DM |
480 | } |
481 | ||
c79442f2 | 482 | last if !$repeat; |
f7ccd1b3 | 483 | } |
f7ccd1b3 | 484 | |
8f0bb968 | 485 | $self->flush_master_status(); |
c0bbd038 DM |
486 | } |
487 | ||
a875fbe8 DM |
488 | # functions to compute next service states |
489 | # $cd: service configuration data (read only) | |
490 | # $sd: service status data (read only) | |
491 | # | |
492 | # Note: use change_service_state() to alter state | |
493 | # | |
494 | ||
0df5b3dd DM |
495 | sub next_state_request_stop { |
496 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; | |
497 | ||
498 | my $haenv = $self->{haenv}; | |
499 | my $ns = $self->{ns}; | |
500 | ||
501 | # check result from LRM daemon | |
502 | if ($lrm_res) { | |
503 | my $exit_code = $lrm_res->{exit_code}; | |
a89ff919 | 504 | if ($exit_code == SUCCESS) { |
0df5b3dd DM |
505 | &$change_service_state($self, $sid, 'stopped'); |
506 | return; | |
507 | } else { | |
33f01524 | 508 | $haenv->log('err', "service '$sid' stop failed (exit code $exit_code)"); |
0df5b3dd DM |
509 | &$change_service_state($self, $sid, 'error'); # fixme: what state? |
510 | return; | |
511 | } | |
512 | } | |
513 | ||
ce3d7003 | 514 | if ($ns->node_is_offline_delayed($sd->{node})) { |
0df5b3dd DM |
515 | &$change_service_state($self, $sid, 'fence'); |
516 | return; | |
517 | } | |
518 | } | |
519 | ||
8aaa0e36 DM |
520 | sub next_state_migrate_relocate { |
521 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; | |
522 | ||
523 | my $haenv = $self->{haenv}; | |
524 | my $ns = $self->{ns}; | |
525 | ||
526 | # check result from LRM daemon | |
527 | if ($lrm_res) { | |
528 | my $exit_code = $lrm_res->{exit_code}; | |
bb07bd2c | 529 | my $req_state = $cd->{state} eq 'started' ? 'started' : 'request_stop'; |
a89ff919 | 530 | if ($exit_code == SUCCESS) { |
542a9902 | 531 | &$change_service_state($self, $sid, $req_state, node => $sd->{target}); |
8aaa0e36 | 532 | return; |
660596ce TL |
533 | } elsif ($exit_code == EWRONG_NODE) { |
534 | $haenv->log('err', "service '$sid' - migration failed: service" . | |
535 | " registered on wrong node!"); | |
536 | &$change_service_state($self, $sid, 'error'); | |
8aaa0e36 DM |
537 | } else { |
538 | $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)"); | |
542a9902 | 539 | &$change_service_state($self, $sid, $req_state, node => $sd->{node}); |
8aaa0e36 DM |
540 | return; |
541 | } | |
542 | } | |
543 | ||
ce3d7003 | 544 | if ($ns->node_is_offline_delayed($sd->{node})) { |
8aaa0e36 DM |
545 | &$change_service_state($self, $sid, 'fence'); |
546 | return; | |
547 | } | |
548 | } | |
549 | ||
a875fbe8 | 550 | sub next_state_stopped { |
abc920b4 | 551 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; |
a875fbe8 DM |
552 | |
553 | my $haenv = $self->{haenv}; | |
e88469ba | 554 | my $ns = $self->{ns}; |
a875fbe8 | 555 | |
ff6f1c5c DM |
556 | if ($sd->{node} ne $cd->{node}) { |
557 | # this can happen if we fence a node with active migrations | |
558 | # hack: modify $sd (normally this should be considered read-only) | |
24678a59 | 559 | $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node})"); |
289e4784 | 560 | $sd->{node} = $cd->{node}; |
ff6f1c5c DM |
561 | } |
562 | ||
94b7ebe2 DM |
563 | if ($sd->{cmd}) { |
564 | my ($cmd, $target) = @{$sd->{cmd}}; | |
565 | delete $sd->{cmd}; | |
566 | ||
b0fdf86a | 567 | if ($cmd eq 'migrate' || $cmd eq 'relocate') { |
94b7ebe2 | 568 | if (!$ns->node_is_online($target)) { |
b0fdf86a | 569 | $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online"); |
e88469ba | 570 | } elsif ($sd->{node} eq $target) { |
b0fdf86a | 571 | $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'"); |
94b7ebe2 | 572 | } else { |
9dad9c88 TL |
573 | &$change_service_state($self, $sid, $cmd, node => $sd->{node}, |
574 | target => $target); | |
9da84a0d | 575 | return; |
94b7ebe2 DM |
576 | } |
577 | } else { | |
289e4784 | 578 | $haenv->log('err', "unknown command '$cmd' for service '$sid'"); |
94b7ebe2 | 579 | } |
35cbb764 | 580 | } |
94b7ebe2 | 581 | |
a875fbe8 | 582 | if ($cd->{state} eq 'disabled') { |
35cbb764 TL |
583 | # NOTE: do nothing here, the stop state is an exception as we do not |
584 | # process the LRM result here, thus the LRM always tries to stop the | |
585 | # service (protection for the case no CRM is active) | |
e88469ba | 586 | return; |
35cbb764 | 587 | } |
e88469ba | 588 | |
af14d5f3 TL |
589 | if ($ns->node_is_offline_delayed($sd->{node})) { |
590 | &$change_service_state($self, $sid, 'fence'); | |
591 | return; | |
592 | } | |
593 | ||
594 | if ($cd->{state} eq 'stopped') { | |
595 | # almost the same as 'disabled' state but the service will also get recovered | |
596 | return; | |
597 | } | |
598 | ||
bb07bd2c | 599 | if ($cd->{state} eq 'started') { |
9da84a0d TL |
600 | # simply mark it started, if it's on the wrong node |
601 | # next_state_started will fix that for us | |
602 | &$change_service_state($self, $sid, 'started', node => $sd->{node}); | |
e88469ba | 603 | return; |
a875fbe8 | 604 | } |
e88469ba DM |
605 | |
606 | $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration"); | |
a875fbe8 DM |
607 | } |
608 | ||
46139211 | 609 | sub record_service_failed_on_node { |
57fe8e87 | 610 | my ($self, $sid, $node) = @_; |
46139211 | 611 | |
57fe8e87 DM |
612 | if (!defined($self->{ss}->{$sid}->{failed_nodes})) { |
613 | $self->{ss}->{$sid}->{failed_nodes} = []; | |
614 | } | |
46139211 | 615 | |
57fe8e87 | 616 | push @{$self->{ss}->{$sid}->{failed_nodes}}, $node; |
46139211 TL |
617 | } |
618 | ||
a875fbe8 | 619 | sub next_state_started { |
abc920b4 | 620 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; |
a875fbe8 DM |
621 | |
622 | my $haenv = $self->{haenv}; | |
ea4443cc | 623 | my $master_status = $self->{ms}; |
a875fbe8 DM |
624 | my $ns = $self->{ns}; |
625 | ||
626 | if (!$ns->node_is_online($sd->{node})) { | |
b0e9158d | 627 | if ($ns->node_is_offline_delayed($sd->{node})) { |
5385a606 DM |
628 | &$change_service_state($self, $sid, 'fence'); |
629 | } | |
e88469ba DM |
630 | return; |
631 | } | |
289e4784 | 632 | |
af14d5f3 | 633 | if ($cd->{state} eq 'disabled' || $cd->{state} eq 'stopped') { |
e88469ba DM |
634 | &$change_service_state($self, $sid, 'request_stop'); |
635 | return; | |
636 | } | |
637 | ||
bb07bd2c | 638 | if ($cd->{state} eq 'started') { |
e88469ba DM |
639 | |
640 | if ($sd->{cmd}) { | |
641 | my ($cmd, $target) = @{$sd->{cmd}}; | |
642 | delete $sd->{cmd}; | |
643 | ||
b0fdf86a | 644 | if ($cmd eq 'migrate' || $cmd eq 'relocate') { |
e88469ba | 645 | if (!$ns->node_is_online($target)) { |
b0fdf86a | 646 | $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online"); |
e88469ba | 647 | } elsif ($sd->{node} eq $target) { |
b0fdf86a | 648 | $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'"); |
e88469ba | 649 | } else { |
a3cb8dcb | 650 | $haenv->log('info', "$cmd service '$sid' to node '$target'"); |
b0fdf86a | 651 | &$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target); |
e88469ba | 652 | } |
a875fbe8 | 653 | } else { |
289e4784 | 654 | $haenv->log('err', "unknown command '$cmd' for service '$sid'"); |
a875fbe8 DM |
655 | } |
656 | } else { | |
b0fdf86a | 657 | |
abc920b4 | 658 | my $try_next = 0; |
46139211 | 659 | |
ea4443cc | 660 | if ($lrm_res) { |
46139211 | 661 | |
e9e1cd68 TL |
662 | my $ec = $lrm_res->{exit_code}; |
663 | if ($ec == SUCCESS) { | |
664 | ||
46139211 | 665 | if (defined($sd->{failed_nodes})) { |
81449997 | 666 | $haenv->log('info', "relocation policy successful for '$sid' on node '$sd->{node}'," . |
46139211 TL |
667 | " failed nodes: " . join(', ', @{$sd->{failed_nodes}}) ); |
668 | } | |
669 | ||
670 | delete $sd->{failed_nodes}; | |
e9e1cd68 | 671 | |
b47920fd DM |
672 | # store flag to indicate successful start - only valid while state == 'started' |
673 | $sd->{running} = 1; | |
674 | ||
e9e1cd68 | 675 | } elsif ($ec == ERROR) { |
b47920fd DM |
676 | |
677 | delete $sd->{running}; | |
678 | ||
e9e1cd68 | 679 | # apply our relocate policy if we got ERROR from the LRM |
46139211 | 680 | $self->record_service_failed_on_node($sid, $sd->{node}); |
ea4443cc | 681 | |
46139211 | 682 | if (scalar(@{$sd->{failed_nodes}}) <= $cd->{max_relocate}) { |
ea4443cc | 683 | |
e9e1cd68 TL |
684 | # tell select_service_node to relocate if possible |
685 | $try_next = 1; | |
ea4443cc TL |
686 | |
687 | $haenv->log('warning', "starting service $sid on node". | |
688 | " '$sd->{node}' failed, relocating service."); | |
ea4443cc TL |
689 | |
690 | } else { | |
691 | ||
46139211 TL |
692 | $haenv->log('err', "recovery policy for service $sid " . |
693 | "failed, entering error state. Failed nodes: ". | |
694 | join(', ', @{$sd->{failed_nodes}})); | |
ea4443cc TL |
695 | &$change_service_state($self, $sid, 'error'); |
696 | return; | |
697 | ||
698 | } | |
e9e1cd68 | 699 | } else { |
46139211 TL |
700 | $self->record_service_failed_on_node($sid, $sd->{node}); |
701 | ||
e9e1cd68 TL |
702 | $haenv->log('err', "service '$sid' got unrecoverable error" . |
703 | " (exit code $ec))"); | |
704 | # we have no save way out (yet) for other errors | |
705 | &$change_service_state($self, $sid, 'error'); | |
35cbb764 | 706 | return; |
ea4443cc | 707 | } |
abc920b4 DM |
708 | } |
709 | ||
e6eeb7dc TL |
710 | my $node = select_service_node($self->{groups}, $self->{online_node_usage}, |
711 | $cd, $sd->{node}, $try_next, $sd->{failed_nodes}); | |
abc920b4 | 712 | |
b0fdf86a | 713 | if ($node && ($sd->{node} ne $node)) { |
c0255b2c TL |
714 | if ($cd->{type} eq 'vm') { |
715 | $haenv->log('info', "migrate service '$sid' to node '$node' (running)"); | |
716 | &$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node); | |
717 | } else { | |
718 | $haenv->log('info', "relocate service '$sid' to node '$node'"); | |
719 | &$change_service_state($self, $sid, 'relocate', node => $sd->{node}, target => $node); | |
720 | } | |
b0fdf86a | 721 | } else { |
e6eeb7dc TL |
722 | if ($try_next && !defined($node)) { |
723 | $haenv->log('warning', "Start Error Recovery: Tried all available " . | |
724 | " nodes for service '$sid', retry start on current node. " . | |
725 | "Tried nodes: " . join(', ', @{$sd->{failed_nodes}})); | |
726 | } | |
35cbb764 | 727 | # ensure service get started again if it went unexpected down |
bf2d8d74 TL |
728 | # but ensure also no LRM result gets lost |
729 | $sd->{uid} = compute_new_uuid($sd->{state}) if defined($lrm_res); | |
b0fdf86a | 730 | } |
a875fbe8 | 731 | } |
e88469ba DM |
732 | |
733 | return; | |
35cbb764 | 734 | } |
e88469ba DM |
735 | |
736 | $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration"); | |
a875fbe8 | 737 | } |
c0bbd038 | 738 | |
a2881965 TL |
739 | sub next_state_error { |
740 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; | |
741 | ||
742 | my $ns = $self->{ns}; | |
46139211 | 743 | my $ms = $self->{ms}; |
a2881965 TL |
744 | |
745 | if ($cd->{state} eq 'disabled') { | |
46139211 TL |
746 | # clean up on error recovery |
747 | delete $sd->{failed_nodes}; | |
748 | ||
a2881965 TL |
749 | &$change_service_state($self, $sid, 'stopped'); |
750 | return; | |
751 | } | |
752 | ||
a2881965 TL |
753 | } |
754 | ||
c0bbd038 | 755 | 1; |