]>
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 DM |
6 | |
7 | use Data::Dumper; | |
c142ebc9 | 8 | use PVE::Tools; |
a89ff919 | 9 | use PVE::HA::Tools ':exit_codes'; |
c0bbd038 DM |
10 | use PVE::HA::NodeStatus; |
11 | ||
ceac1930 | 12 | my $fence_delay = 60; |
5385a606 | 13 | |
c0bbd038 | 14 | sub new { |
8f0bb968 | 15 | my ($this, $haenv) = @_; |
c0bbd038 DM |
16 | |
17 | my $class = ref($this) || $this; | |
18 | ||
8f0bb968 DM |
19 | my $ms = $haenv->read_manager_status(); |
20 | ||
21 | $ms->{master_node} = $haenv->nodename(); | |
22 | ||
23 | my $ns = PVE::HA::NodeStatus->new($haenv, $ms->{node_status} || {}); | |
24 | ||
59fd7207 DM |
25 | # fixme: use separate class PVE::HA::ServiceStatus |
26 | my $ss = $ms->{service_status} || {}; | |
27 | ||
c0bbd038 | 28 | my $self = bless { |
8f0bb968 DM |
29 | haenv => $haenv, |
30 | ms => $ms, # master status | |
31 | ns => $ns, # PVE::HA::NodeStatus | |
59fd7207 | 32 | ss => $ss, # service status |
c0bbd038 DM |
33 | }, $class; |
34 | ||
35 | return $self; | |
36 | } | |
37 | ||
d84da043 DM |
38 | sub cleanup { |
39 | my ($self) = @_; | |
40 | ||
41 | # todo: ? | |
42 | } | |
43 | ||
8f0bb968 | 44 | sub flush_master_status { |
c0bbd038 DM |
45 | my ($self) = @_; |
46 | ||
59fd7207 | 47 | my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss}); |
c0bbd038 | 48 | |
8f0bb968 | 49 | $ms->{node_status} = $ns->{status}; |
59fd7207 | 50 | $ms->{service_status} = $ss; |
d2f612cf DM |
51 | $ms->{timestamp} = $haenv->get_time(); |
52 | ||
8f0bb968 DM |
53 | $haenv->write_manager_status($ms); |
54 | } | |
c0bbd038 | 55 | |
f7ccd1b3 | 56 | sub select_service_node { |
270d4406 | 57 | my ($groups, $online_node_usage, $service_conf, $current_node, $try_next) = @_; |
f7ccd1b3 | 58 | |
e0a56314 | 59 | my $group = { 'nodes' => { $service_conf->{node} => 1 } }; # default group |
abc920b4 | 60 | |
270d4406 DM |
61 | $group = $groups->{ids}->{$service_conf->{group}} if $service_conf->{group} && |
62 | $groups->{ids}->{$service_conf->{group}}; | |
abc920b4 DM |
63 | |
64 | my $pri_groups = {}; | |
65 | my $group_members = {}; | |
e0a56314 | 66 | foreach my $entry (keys %{$group->{nodes}}) { |
abc920b4 DM |
67 | my ($node, $pri) = ($entry, 0); |
68 | if ($entry =~ m/^(\S+):(\d+)$/) { | |
69 | ($node, $pri) = ($1, $2); | |
70 | } | |
270d4406 | 71 | next if !defined($online_node_usage->{$node}); # offline |
abc920b4 DM |
72 | $pri_groups->{$pri}->{$node} = 1; |
73 | $group_members->{$node} = $pri; | |
74 | } | |
f7ccd1b3 | 75 | |
270d4406 | 76 | |
abc920b4 DM |
77 | # add non-group members to unrestricted groups (priority -1) |
78 | if (!$group->{restricted}) { | |
79 | my $pri = -1; | |
270d4406 | 80 | foreach my $node (keys %$online_node_usage) { |
abc920b4 DM |
81 | next if defined($group_members->{$node}); |
82 | $pri_groups->{$pri}->{$node} = 1; | |
83 | $group_members->{$node} = -1; | |
84 | } | |
85 | } | |
86 | ||
270d4406 | 87 | |
abc920b4 DM |
88 | my @pri_list = sort {$b <=> $a} keys %$pri_groups; |
89 | return undef if !scalar(@pri_list); | |
270d4406 | 90 | |
abc920b4 DM |
91 | if (!$try_next && $group->{nofailback} && defined($group_members->{$current_node})) { |
92 | return $current_node; | |
93 | } | |
94 | ||
95 | # select node from top priority node list | |
96 | ||
97 | my $top_pri = $pri_list[0]; | |
98 | ||
c142ebc9 DM |
99 | my @nodes = sort { |
100 | $online_node_usage->{$a} <=> $online_node_usage->{$b} || $a cmp $b | |
101 | } keys %{$pri_groups->{$top_pri}}; | |
abc920b4 DM |
102 | |
103 | my $found; | |
104 | for (my $i = scalar(@nodes) - 1; $i >= 0; $i--) { | |
105 | my $node = $nodes[$i]; | |
106 | if ($node eq $current_node) { | |
107 | $found = $i; | |
108 | last; | |
109 | } | |
110 | } | |
111 | ||
abc920b4 DM |
112 | if ($try_next) { |
113 | ||
114 | if (defined($found) && ($found < (scalar(@nodes) - 1))) { | |
115 | return $nodes[$found + 1]; | |
116 | } else { | |
117 | return $nodes[0]; | |
118 | } | |
119 | ||
120 | } else { | |
121 | ||
122 | return $nodes[$found] if defined($found); | |
123 | ||
124 | return $nodes[0]; | |
125 | ||
126 | } | |
f7ccd1b3 DM |
127 | } |
128 | ||
c4a221bc DM |
129 | my $uid_counter = 0; |
130 | ||
d55aa611 DM |
131 | sub compute_new_uuid { |
132 | my ($state) = @_; | |
133 | ||
134 | $uid_counter++; | |
135 | return md5_base64($state . $$ . time() . $uid_counter); | |
136 | } | |
137 | ||
618fbeda DM |
138 | my $valid_service_states = { |
139 | stopped => 1, | |
140 | request_stop => 1, | |
141 | started => 1, | |
142 | fence => 1, | |
143 | migrate => 1, | |
b0fdf86a | 144 | relocate => 1, |
9c7d068b | 145 | freeze => 1, |
618fbeda DM |
146 | error => 1, |
147 | }; | |
148 | ||
270d4406 DM |
149 | sub recompute_online_node_usage { |
150 | my ($self) = @_; | |
151 | ||
152 | my $online_node_usage = {}; | |
153 | ||
154 | my $online_nodes = $self->{ns}->list_online_nodes(); | |
155 | ||
156 | foreach my $node (@$online_nodes) { | |
157 | $online_node_usage->{$node} = 0; | |
158 | } | |
159 | ||
160 | foreach my $sid (keys %{$self->{ss}}) { | |
161 | my $sd = $self->{ss}->{$sid}; | |
162 | my $state = $sd->{state}; | |
163 | if (defined($online_node_usage->{$sd->{node}})) { | |
164 | if (($state eq 'started') || ($state eq 'request_stop') || | |
9c7d068b | 165 | ($state eq 'fence') || ($state eq 'freeze') || ($state eq 'error')) { |
270d4406 DM |
166 | $online_node_usage->{$sd->{node}}++; |
167 | } elsif (($state eq 'migrate') || ($state eq 'relocate')) { | |
168 | $online_node_usage->{$sd->{target}}++; | |
169 | } elsif ($state eq 'stopped') { | |
170 | # do nothing | |
171 | } else { | |
172 | die "should not be reached"; | |
173 | } | |
174 | } | |
175 | } | |
176 | ||
177 | $self->{online_node_usage} = $online_node_usage; | |
178 | } | |
179 | ||
4e01bc86 DM |
180 | my $change_service_state = sub { |
181 | my ($self, $sid, $new_state, %params) = @_; | |
182 | ||
183 | my ($haenv, $ss) = ($self->{haenv}, $self->{ss}); | |
184 | ||
185 | my $sd = $ss->{$sid} || die "no such service '$sid"; | |
186 | ||
187 | my $old_state = $sd->{state}; | |
e4ffb299 | 188 | my $old_node = $sd->{node}; |
4e01bc86 DM |
189 | |
190 | die "no state change" if $old_state eq $new_state; # just to be sure | |
191 | ||
618fbeda DM |
192 | die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state}; |
193 | ||
e4ffb299 DM |
194 | foreach my $k (keys %$sd) { delete $sd->{$k}; }; |
195 | ||
196 | $sd->{state} = $new_state; | |
197 | $sd->{node} = $old_node; | |
198 | ||
199 | my $text_state = ''; | |
ba623362 | 200 | foreach my $k (sort keys %params) { |
4e01bc86 | 201 | my $v = $params{$k}; |
e4ffb299 DM |
202 | $text_state .= ", " if $text_state; |
203 | $text_state .= "$k = $v"; | |
4e01bc86 DM |
204 | $sd->{$k} = $v; |
205 | } | |
270d4406 DM |
206 | |
207 | $self->recompute_online_node_usage(); | |
208 | ||
d55aa611 DM |
209 | $sd->{uid} = compute_new_uuid($new_state); |
210 | ||
4e01bc86 | 211 | |
e4ffb299 | 212 | $text_state = " ($text_state)" if $text_state; |
e5986717 | 213 | $haenv->log('info', "service '$sid': state changed from '${old_state}' to '${new_state}' $text_state"); |
4e01bc86 DM |
214 | }; |
215 | ||
9c7d068b | 216 | # read LRM status for all nodes |
c4a221bc | 217 | sub read_lrm_status { |
332170bd | 218 | my ($self) = @_; |
c4a221bc | 219 | |
9c7d068b | 220 | my $nodes = $self->{ns}->list_nodes(); |
c4a221bc DM |
221 | my $haenv = $self->{haenv}; |
222 | ||
9c7d068b DM |
223 | my $results = {}; |
224 | my $modes = {}; | |
332170bd | 225 | foreach my $node (@$nodes) { |
9c7d068b | 226 | my $lrm_status = $haenv->read_lrm_status($node); |
02ffd753 | 227 | $modes->{$node} = $lrm_status->{mode} || 'active'; |
9c7d068b DM |
228 | foreach my $uid (keys %{$lrm_status->{results}}) { |
229 | next if $results->{$uid}; # should not happen | |
230 | $results->{$uid} = $lrm_status->{results}->{$uid}; | |
c4a221bc DM |
231 | } |
232 | } | |
233 | ||
9c7d068b DM |
234 | |
235 | return ($results, $modes); | |
c4a221bc DM |
236 | } |
237 | ||
aa98a844 DM |
238 | # read new crm commands and save them into crm master status |
239 | sub update_crm_commands { | |
240 | my ($self) = @_; | |
241 | ||
242 | my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss}); | |
243 | ||
244 | my $cmdlist = $haenv->read_crm_commands(); | |
245 | ||
246 | foreach my $cmd (split(/\n/, $cmdlist)) { | |
247 | chomp $cmd; | |
248 | ||
b0fdf86a DM |
249 | if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) { |
250 | my ($task, $sid, $node) = ($1, $2, $3); | |
aa98a844 DM |
251 | if (my $sd = $ss->{$sid}) { |
252 | if (!$ns->node_is_online($node)) { | |
253 | $haenv->log('err', "crm command error - node not online: $cmd"); | |
254 | } else { | |
255 | if ($node eq $sd->{node}) { | |
256 | $haenv->log('info', "ignore crm command - service already on target node: $cmd"); | |
257 | } else { | |
258 | $haenv->log('info', "got crm command: $cmd"); | |
b0fdf86a | 259 | $ss->{$sid}->{cmd} = [ $task, $node]; |
aa98a844 DM |
260 | } |
261 | } | |
262 | } else { | |
263 | $haenv->log('err', "crm command error - no such service: $cmd"); | |
264 | } | |
265 | ||
266 | } else { | |
267 | $haenv->log('err', "unable to parse crm command: $cmd"); | |
268 | } | |
269 | } | |
270 | ||
271 | } | |
272 | ||
8f0bb968 DM |
273 | sub manage { |
274 | my ($self) = @_; | |
c0bbd038 | 275 | |
59fd7207 | 276 | my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss}); |
c0bbd038 | 277 | |
332170bd | 278 | $ns->update($haenv->get_node_info()); |
c79442f2 DM |
279 | |
280 | if (!$ns->node_is_online($haenv->nodename())) { | |
e5986717 | 281 | $haenv->log('info', "master seems offline"); |
c79442f2 DM |
282 | return; |
283 | } | |
284 | ||
9c7d068b | 285 | my ($lrm_results, $lrm_modes) = $self->read_lrm_status(); |
c4a221bc | 286 | |
f7ccd1b3 DM |
287 | my $sc = $haenv->read_service_config(); |
288 | ||
abc920b4 DM |
289 | $self->{groups} = $haenv->read_group_config(); # update |
290 | ||
f7ccd1b3 DM |
291 | # compute new service status |
292 | ||
293 | # add new service | |
cc32a8f3 | 294 | foreach my $sid (sort keys %$sc) { |
f7ccd1b3 | 295 | next if $ss->{$sid}; # already there |
cc32a8f3 | 296 | $haenv->log('info', "adding new service '$sid' on node '$sc->{$sid}->{node}'"); |
f7ccd1b3 | 297 | # assume we are running to avoid relocate running service at add |
d55aa611 DM |
298 | $ss->{$sid} = { state => 'started', node => $sc->{$sid}->{node}, |
299 | uid => compute_new_uuid('started') }; | |
f7ccd1b3 DM |
300 | } |
301 | ||
4e5764af DM |
302 | # remove stale service from manager state |
303 | foreach my $sid (keys %$ss) { | |
304 | next if $sc->{$sid}; | |
305 | $haenv->log('info', "removing stale service '$sid' (no config)"); | |
306 | delete $ss->{$sid}; | |
307 | } | |
308 | ||
aa98a844 DM |
309 | $self->update_crm_commands(); |
310 | ||
c79442f2 DM |
311 | for (;;) { |
312 | my $repeat = 0; | |
270d4406 DM |
313 | |
314 | $self->recompute_online_node_usage(); | |
f7ccd1b3 | 315 | |
c79442f2 DM |
316 | foreach my $sid (keys %$ss) { |
317 | my $sd = $ss->{$sid}; | |
318 | my $cd = $sc->{$sid} || { state => 'disabled' }; | |
f7ccd1b3 | 319 | |
9c7d068b | 320 | my $lrm_res = $sd->{uid} ? $lrm_results->{$sd->{uid}} : undef; |
a875fbe8 | 321 | |
c79442f2 DM |
322 | my $last_state = $sd->{state}; |
323 | ||
324 | if ($last_state eq 'stopped') { | |
325 | ||
abc920b4 | 326 | $self->next_state_stopped($sid, $cd, $sd, $lrm_res); |
f7ccd1b3 | 327 | |
c79442f2 | 328 | } elsif ($last_state eq 'started') { |
f7ccd1b3 | 329 | |
abc920b4 | 330 | $self->next_state_started($sid, $cd, $sd, $lrm_res); |
f7ccd1b3 | 331 | |
b0fdf86a | 332 | } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') { |
f7ccd1b3 | 333 | |
8aaa0e36 | 334 | $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res); |
f7ccd1b3 | 335 | |
c79442f2 | 336 | } elsif ($last_state eq 'fence') { |
f7ccd1b3 | 337 | |
21e37ed4 | 338 | # do nothing here - wait until fenced |
f7ccd1b3 | 339 | |
c79442f2 | 340 | } elsif ($last_state eq 'request_stop') { |
f7ccd1b3 | 341 | |
0df5b3dd | 342 | $self->next_state_request_stop($sid, $cd, $sd, $lrm_res); |
618fbeda | 343 | |
9c7d068b DM |
344 | } elsif ($last_state eq 'freeze') { |
345 | ||
346 | my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef; | |
9c7d068b DM |
347 | # unfreeze |
348 | &$change_service_state($self, $sid, 'started') | |
02ffd753 | 349 | if $lrm_mode && $lrm_mode eq 'active'; |
9c7d068b | 350 | |
e88469ba DM |
351 | } elsif ($last_state eq 'error') { |
352 | ||
a2881965 | 353 | $self->next_state_error($sid, $cd, $sd, $lrm_res); |
e88469ba | 354 | |
a875fbe8 DM |
355 | } else { |
356 | ||
357 | die "unknown service state '$last_state'"; | |
618fbeda | 358 | } |
21e37ed4 | 359 | |
9c7d068b | 360 | my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef; |
07adc6a6 DM |
361 | if ($lrm_mode && $lrm_mode eq 'restart') { |
362 | if (($sd->{state} eq 'started' || $sd->{state} eq 'stopped' || | |
363 | $sd->{state} eq 'request_stop')) { | |
364 | &$change_service_state($self, $sid, 'freeze'); | |
365 | } | |
9c7d068b | 366 | } |
07adc6a6 | 367 | |
c79442f2 | 368 | $repeat = 1 if $sd->{state} ne $last_state; |
f7ccd1b3 DM |
369 | } |
370 | ||
21e37ed4 DM |
371 | # handle fencing |
372 | my $fenced_nodes = {}; | |
373 | foreach my $sid (keys %$ss) { | |
374 | my $sd = $ss->{$sid}; | |
375 | next if $sd->{state} ne 'fence'; | |
376 | ||
377 | if (!defined($fenced_nodes->{$sd->{node}})) { | |
378 | $fenced_nodes->{$sd->{node}} = $ns->fence_node($sd->{node}) || 0; | |
379 | } | |
380 | ||
381 | next if !$fenced_nodes->{$sd->{node}}; | |
382 | ||
63f6a08c | 383 | # node fence was successful - mark service as stopped |
21e37ed4 DM |
384 | &$change_service_state($self, $sid, 'stopped'); |
385 | } | |
386 | ||
c79442f2 | 387 | last if !$repeat; |
f7ccd1b3 | 388 | } |
f7ccd1b3 | 389 | |
8f0bb968 | 390 | $self->flush_master_status(); |
c0bbd038 DM |
391 | } |
392 | ||
a875fbe8 DM |
393 | # functions to compute next service states |
394 | # $cd: service configuration data (read only) | |
395 | # $sd: service status data (read only) | |
396 | # | |
397 | # Note: use change_service_state() to alter state | |
398 | # | |
399 | ||
0df5b3dd DM |
400 | sub next_state_request_stop { |
401 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; | |
402 | ||
403 | my $haenv = $self->{haenv}; | |
404 | my $ns = $self->{ns}; | |
405 | ||
406 | # check result from LRM daemon | |
407 | if ($lrm_res) { | |
408 | my $exit_code = $lrm_res->{exit_code}; | |
a89ff919 | 409 | if ($exit_code == SUCCESS) { |
0df5b3dd DM |
410 | &$change_service_state($self, $sid, 'stopped'); |
411 | return; | |
412 | } else { | |
33f01524 | 413 | $haenv->log('err', "service '$sid' stop failed (exit code $exit_code)"); |
0df5b3dd DM |
414 | &$change_service_state($self, $sid, 'error'); # fixme: what state? |
415 | return; | |
416 | } | |
417 | } | |
418 | ||
5385a606 | 419 | if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) { |
0df5b3dd DM |
420 | &$change_service_state($self, $sid, 'fence'); |
421 | return; | |
422 | } | |
423 | } | |
424 | ||
8aaa0e36 DM |
425 | sub next_state_migrate_relocate { |
426 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; | |
427 | ||
428 | my $haenv = $self->{haenv}; | |
429 | my $ns = $self->{ns}; | |
430 | ||
431 | # check result from LRM daemon | |
432 | if ($lrm_res) { | |
433 | my $exit_code = $lrm_res->{exit_code}; | |
a89ff919 | 434 | if ($exit_code == SUCCESS) { |
8aaa0e36 DM |
435 | &$change_service_state($self, $sid, 'started', node => $sd->{target}); |
436 | return; | |
437 | } else { | |
438 | $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)"); | |
439 | &$change_service_state($self, $sid, 'started', node => $sd->{node}); | |
440 | return; | |
441 | } | |
442 | } | |
443 | ||
5385a606 | 444 | if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) { |
8aaa0e36 DM |
445 | &$change_service_state($self, $sid, 'fence'); |
446 | return; | |
447 | } | |
448 | } | |
449 | ||
450 | ||
a875fbe8 | 451 | sub next_state_stopped { |
abc920b4 | 452 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; |
a875fbe8 DM |
453 | |
454 | my $haenv = $self->{haenv}; | |
e88469ba | 455 | my $ns = $self->{ns}; |
a875fbe8 | 456 | |
ff6f1c5c DM |
457 | if ($sd->{node} ne $cd->{node}) { |
458 | # this can happen if we fence a node with active migrations | |
459 | # hack: modify $sd (normally this should be considered read-only) | |
460 | $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node}"); | |
461 | $sd->{node} = $cd->{node}; | |
462 | } | |
463 | ||
94b7ebe2 DM |
464 | if ($sd->{cmd}) { |
465 | my ($cmd, $target) = @{$sd->{cmd}}; | |
466 | delete $sd->{cmd}; | |
467 | ||
b0fdf86a | 468 | if ($cmd eq 'migrate' || $cmd eq 'relocate') { |
94b7ebe2 | 469 | if (!$ns->node_is_online($target)) { |
b0fdf86a | 470 | $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online"); |
e88469ba | 471 | } elsif ($sd->{node} eq $target) { |
b0fdf86a | 472 | $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'"); |
94b7ebe2 | 473 | } else { |
6da27e23 DM |
474 | eval { |
475 | $haenv->change_service_location($sid, $sd->{node}, $target); | |
476 | $cd->{node} = $sd->{node} = $target; # fixme: $sd is read-only??!! | |
477 | $haenv->log('info', "$cmd service '$sid' to node '$target' (stopped)"); | |
478 | }; | |
479 | if (my $err = $@) { | |
480 | $haenv->log('err', "$cmd service '$sid' to node '$target' failed - $err"); | |
481 | } | |
94b7ebe2 DM |
482 | } |
483 | } else { | |
484 | $haenv->log('err', "unknown command '$cmd' for service '$sid'"); | |
485 | } | |
486 | } | |
487 | ||
a875fbe8 DM |
488 | if ($cd->{state} eq 'disabled') { |
489 | # do nothing | |
e88469ba DM |
490 | return; |
491 | } | |
492 | ||
493 | if ($cd->{state} eq 'enabled') { | |
270d4406 | 494 | if (my $node = select_service_node($self->{groups}, $self->{online_node_usage}, $cd, $sd->{node})) { |
a875fbe8 | 495 | if ($node && ($sd->{node} ne $node)) { |
6da27e23 DM |
496 | eval { |
497 | $haenv->change_service_location($sid, $sd->{node}, $node); | |
498 | $cd->{node} = $sd->{node} = $node; # fixme: $sd is read-only??!! | |
499 | }; | |
500 | if (my $err = $@) { | |
501 | $haenv->log('err', "move service '$sid' to node '$node' failed - $err"); | |
502 | } else { | |
503 | &$change_service_state($self, $sid, 'started', node => $node); | |
504 | } | |
505 | } else { | |
506 | &$change_service_state($self, $sid, 'started', node => $node); | |
a875fbe8 | 507 | } |
a875fbe8 DM |
508 | } else { |
509 | # fixme: warn | |
510 | } | |
e88469ba DM |
511 | |
512 | return; | |
a875fbe8 | 513 | } |
e88469ba DM |
514 | |
515 | $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration"); | |
a875fbe8 DM |
516 | } |
517 | ||
518 | sub next_state_started { | |
abc920b4 | 519 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; |
a875fbe8 DM |
520 | |
521 | my $haenv = $self->{haenv}; | |
ea4443cc | 522 | my $master_status = $self->{ms}; |
a875fbe8 DM |
523 | my $ns = $self->{ns}; |
524 | ||
525 | if (!$ns->node_is_online($sd->{node})) { | |
5385a606 DM |
526 | if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) { |
527 | &$change_service_state($self, $sid, 'fence'); | |
528 | } | |
e88469ba DM |
529 | return; |
530 | } | |
a875fbe8 | 531 | |
e88469ba DM |
532 | if ($cd->{state} eq 'disabled') { |
533 | &$change_service_state($self, $sid, 'request_stop'); | |
534 | return; | |
535 | } | |
536 | ||
537 | if ($cd->{state} eq 'enabled') { | |
e88469ba DM |
538 | |
539 | if ($sd->{cmd}) { | |
540 | my ($cmd, $target) = @{$sd->{cmd}}; | |
541 | delete $sd->{cmd}; | |
542 | ||
b0fdf86a | 543 | if ($cmd eq 'migrate' || $cmd eq 'relocate') { |
e88469ba | 544 | if (!$ns->node_is_online($target)) { |
b0fdf86a | 545 | $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online"); |
e88469ba | 546 | } elsif ($sd->{node} eq $target) { |
b0fdf86a | 547 | $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'"); |
e88469ba | 548 | } else { |
a3cb8dcb | 549 | $haenv->log('info', "$cmd service '$sid' to node '$target'"); |
b0fdf86a | 550 | &$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target); |
e88469ba | 551 | } |
a875fbe8 | 552 | } else { |
e88469ba | 553 | $haenv->log('err', "unknown command '$cmd' for service '$sid'"); |
a875fbe8 DM |
554 | } |
555 | } else { | |
b0fdf86a | 556 | |
abc920b4 | 557 | my $try_next = 0; |
ea4443cc | 558 | if ($lrm_res) { |
e9e1cd68 TL |
559 | my $ec = $lrm_res->{exit_code}; |
560 | if ($ec == SUCCESS) { | |
561 | ||
562 | $master_status->{relocate_trial}->{$sid} = 0; | |
563 | ||
564 | } elsif ($ec == ETRY_AGAIN) { | |
565 | ||
566 | # do nothing, the LRM wants to try again | |
567 | ||
568 | } elsif ($ec == ERROR) { | |
569 | # apply our relocate policy if we got ERROR from the LRM | |
ea4443cc TL |
570 | |
571 | my $try = $master_status->{relocate_trial}->{$sid} || 0; | |
572 | ||
573 | if ($try < $cd->{max_relocate}) { | |
574 | ||
575 | $try++; | |
e9e1cd68 TL |
576 | # tell select_service_node to relocate if possible |
577 | $try_next = 1; | |
ea4443cc TL |
578 | |
579 | $haenv->log('warning', "starting service $sid on node". | |
580 | " '$sd->{node}' failed, relocating service."); | |
581 | $master_status->{relocate_trial}->{$sid} = $try; | |
582 | ||
583 | } else { | |
584 | ||
585 | $haenv->log('err', "recovery policy for service". | |
586 | " $sid failed, entering error state!"); | |
587 | &$change_service_state($self, $sid, 'error'); | |
588 | return; | |
589 | ||
590 | } | |
e9e1cd68 TL |
591 | } else { |
592 | $haenv->log('err', "service '$sid' got unrecoverable error" . | |
593 | " (exit code $ec))"); | |
594 | # we have no save way out (yet) for other errors | |
595 | &$change_service_state($self, $sid, 'error'); | |
ea4443cc | 596 | } |
abc920b4 DM |
597 | } |
598 | ||
270d4406 DM |
599 | my $node = select_service_node($self->{groups}, $self->{online_node_usage}, |
600 | $cd, $sd->{node}, $try_next); | |
abc920b4 | 601 | |
b0fdf86a | 602 | if ($node && ($sd->{node} ne $node)) { |
c0255b2c TL |
603 | if ($cd->{type} eq 'vm') { |
604 | $haenv->log('info', "migrate service '$sid' to node '$node' (running)"); | |
605 | &$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node); | |
606 | } else { | |
607 | $haenv->log('info', "relocate service '$sid' to node '$node'"); | |
608 | &$change_service_state($self, $sid, 'relocate', node => $sd->{node}, target => $node); | |
609 | } | |
b0fdf86a DM |
610 | } else { |
611 | # do nothing | |
612 | } | |
a875fbe8 | 613 | } |
e88469ba DM |
614 | |
615 | return; | |
616 | } | |
617 | ||
618 | $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration"); | |
a875fbe8 | 619 | } |
c0bbd038 | 620 | |
a2881965 TL |
621 | sub next_state_error { |
622 | my ($self, $sid, $cd, $sd, $lrm_res) = @_; | |
623 | ||
624 | my $ns = $self->{ns}; | |
625 | ||
626 | if ($cd->{state} eq 'disabled') { | |
627 | &$change_service_state($self, $sid, 'stopped'); | |
628 | return; | |
629 | } | |
630 | ||
631 | if ($ns->node_is_offline_delayed($sd->{node}, $fence_delay)) { | |
632 | &$change_service_state($self, $sid, 'fence'); | |
633 | return; | |
634 | } | |
635 | ||
636 | } | |
637 | ||
c0bbd038 | 638 | 1; |