]> git.proxmox.com Git - pve-ha-manager.git/blob - src/PVE/HA/Manager.pm
restructure directory layout for dpkg-buildpackage
[pve-ha-manager.git] / src / PVE / HA / Manager.pm
1 package PVE::HA::Manager;
2
3 use strict;
4 use warnings;
5 use Digest::MD5 qw(md5_base64);
6
7 use Data::Dumper;
8
9 use PVE::HA::NodeStatus;
10
11 sub new {
12 my ($this, $haenv) = @_;
13
14 my $class = ref($this) || $this;
15
16 my $ms = $haenv->read_manager_status();
17
18 $ms->{master_node} = $haenv->nodename();
19
20 my $ns = PVE::HA::NodeStatus->new($haenv, $ms->{node_status} || {});
21
22 # fixme: use separate class PVE::HA::ServiceStatus
23 my $ss = $ms->{service_status} || {};
24
25 my $self = bless {
26 haenv => $haenv,
27 ms => $ms, # master status
28 ns => $ns, # PVE::HA::NodeStatus
29 ss => $ss, # service status
30 }, $class;
31
32 return $self;
33 }
34
35 sub cleanup {
36 my ($self) = @_;
37
38 # todo: ?
39 }
40
41 sub flush_master_status {
42 my ($self) = @_;
43
44 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
45
46 $ms->{node_status} = $ns->{status};
47 $ms->{service_status} = $ss;
48
49 $haenv->write_manager_status($ms);
50 }
51
52 sub select_service_node {
53 my ($groups, $online_node_usage, $service_conf, $current_node, $try_next) = @_;
54
55 my $group = { 'nodes' => $service_conf->{node} }; # default group
56
57 $group = $groups->{ids}->{$service_conf->{group}} if $service_conf->{group} &&
58 $groups->{ids}->{$service_conf->{group}};
59
60 my $pri_groups = {};
61 my $group_members = {};
62 foreach my $entry (PVE::Tools::split_list($group->{nodes})) {
63 my ($node, $pri) = ($entry, 0);
64 if ($entry =~ m/^(\S+):(\d+)$/) {
65 ($node, $pri) = ($1, $2);
66 }
67 next if !defined($online_node_usage->{$node}); # offline
68 $pri_groups->{$pri}->{$node} = 1;
69 $group_members->{$node} = $pri;
70 }
71
72
73 # add non-group members to unrestricted groups (priority -1)
74 if (!$group->{restricted}) {
75 my $pri = -1;
76 foreach my $node (keys %$online_node_usage) {
77 next if defined($group_members->{$node});
78 $pri_groups->{$pri}->{$node} = 1;
79 $group_members->{$node} = -1;
80 }
81 }
82
83
84 my @pri_list = sort {$b <=> $a} keys %$pri_groups;
85 return undef if !scalar(@pri_list);
86
87 if (!$try_next && $group->{nofailback} && defined($group_members->{$current_node})) {
88 return $current_node;
89 }
90
91 # select node from top priority node list
92
93 my $top_pri = $pri_list[0];
94
95 my @nodes = sort { $online_node_usage->{$a} <=> $online_node_usage->{$b} } keys %{$pri_groups->{$top_pri}};
96
97 my $found;
98 for (my $i = scalar(@nodes) - 1; $i >= 0; $i--) {
99 my $node = $nodes[$i];
100 if ($node eq $current_node) {
101 $found = $i;
102 last;
103 }
104 }
105
106 if ($try_next) {
107
108 if (defined($found) && ($found < (scalar(@nodes) - 1))) {
109 return $nodes[$found + 1];
110 } else {
111 return $nodes[0];
112 }
113
114 } else {
115
116 return $nodes[$found] if defined($found);
117
118 return $nodes[0];
119
120 }
121 }
122
123 my $uid_counter = 0;
124
125 my $valid_service_states = {
126 stopped => 1,
127 request_stop => 1,
128 started => 1,
129 fence => 1,
130 migrate => 1,
131 relocate => 1,
132 error => 1,
133 };
134
135 sub recompute_online_node_usage {
136 my ($self) = @_;
137
138 my $online_node_usage = {};
139
140 my $online_nodes = $self->{ns}->list_online_nodes();
141
142 foreach my $node (@$online_nodes) {
143 $online_node_usage->{$node} = 0;
144 }
145
146 foreach my $sid (keys %{$self->{ss}}) {
147 my $sd = $self->{ss}->{$sid};
148 my $state = $sd->{state};
149 if (defined($online_node_usage->{$sd->{node}})) {
150 if (($state eq 'started') || ($state eq 'request_stop') ||
151 ($state eq 'fence') || ($state eq 'error')) {
152 $online_node_usage->{$sd->{node}}++;
153 } elsif (($state eq 'migrate') || ($state eq 'relocate')) {
154 $online_node_usage->{$sd->{target}}++;
155 } elsif ($state eq 'stopped') {
156 # do nothing
157 } else {
158 die "should not be reached";
159 }
160 }
161 }
162
163 $self->{online_node_usage} = $online_node_usage;
164 }
165
166 my $change_service_state = sub {
167 my ($self, $sid, $new_state, %params) = @_;
168
169 my ($haenv, $ss) = ($self->{haenv}, $self->{ss});
170
171 my $sd = $ss->{$sid} || die "no such service '$sid";
172
173 my $old_state = $sd->{state};
174 my $old_node = $sd->{node};
175
176 die "no state change" if $old_state eq $new_state; # just to be sure
177
178 die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state};
179
180 foreach my $k (keys %$sd) { delete $sd->{$k}; };
181
182 $sd->{state} = $new_state;
183 $sd->{node} = $old_node;
184
185 my $text_state = '';
186 foreach my $k (keys %params) {
187 my $v = $params{$k};
188 $text_state .= ", " if $text_state;
189 $text_state .= "$k = $v";
190 $sd->{$k} = $v;
191 }
192
193 $self->recompute_online_node_usage();
194
195 $uid_counter++;
196 $sd->{uid} = md5_base64($new_state . $$ . time() . $uid_counter);
197
198 $text_state = " ($text_state)" if $text_state;
199 $haenv->log('info', "service '$sid': state changed from '${old_state}' to '${new_state}' $text_state\n");
200 };
201
202 # read LRM status for all active nodes
203 sub read_lrm_status {
204 my ($self) = @_;
205
206 my $nodes = $self->{ns}->list_online_nodes();
207 my $haenv = $self->{haenv};
208
209 my $res = {};
210
211 foreach my $node (@$nodes) {
212 my $ls = $haenv->read_lrm_status($node);
213 foreach my $uid (keys %$ls) {
214 next if $res->{$uid}; # should not happen
215 $res->{$uid} = $ls->{$uid};
216 }
217 }
218
219 return $res;
220 }
221
222 # read new crm commands and save them into crm master status
223 sub update_crm_commands {
224 my ($self) = @_;
225
226 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
227
228 my $cmdlist = $haenv->read_crm_commands();
229
230 foreach my $cmd (split(/\n/, $cmdlist)) {
231 chomp $cmd;
232
233 if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) {
234 my ($task, $sid, $node) = ($1, $2, $3);
235 if (my $sd = $ss->{$sid}) {
236 if (!$ns->node_is_online($node)) {
237 $haenv->log('err', "crm command error - node not online: $cmd");
238 } else {
239 if ($node eq $sd->{node}) {
240 $haenv->log('info', "ignore crm command - service already on target node: $cmd");
241 } else {
242 $haenv->log('info', "got crm command: $cmd");
243 $ss->{$sid}->{cmd} = [ $task, $node];
244 }
245 }
246 } else {
247 $haenv->log('err', "crm command error - no such service: $cmd");
248 }
249
250 } else {
251 $haenv->log('err', "unable to parse crm command: $cmd");
252 }
253 }
254
255 }
256
257 sub manage {
258 my ($self) = @_;
259
260 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
261
262 $ns->update($haenv->get_node_info());
263
264 if (!$ns->node_is_online($haenv->nodename())) {
265 $haenv->log('info', "master seems offline\n");
266 return;
267 }
268
269 my $lrm_status = $self->read_lrm_status();
270
271 my $sc = $haenv->read_service_config();
272
273 $self->{groups} = $haenv->read_group_config(); # update
274
275 # compute new service status
276
277 # add new service
278 foreach my $sid (keys %$sc) {
279 next if $ss->{$sid}; # already there
280 $haenv->log('info', "Adding new service '$sid'\n");
281 # assume we are running to avoid relocate running service at add
282 $ss->{$sid} = { state => 'started', node => $sc->{$sid}->{node}};
283 }
284
285 $self->update_crm_commands();
286
287 for (;;) {
288 my $repeat = 0;
289
290 $self->recompute_online_node_usage();
291
292 foreach my $sid (keys %$ss) {
293 my $sd = $ss->{$sid};
294 my $cd = $sc->{$sid} || { state => 'disabled' };
295
296 my $lrm_res = $sd->{uid} ? $lrm_status->{$sd->{uid}} : undef;
297
298 my $last_state = $sd->{state};
299
300 if ($last_state eq 'stopped') {
301
302 $self->next_state_stopped($sid, $cd, $sd, $lrm_res);
303
304 } elsif ($last_state eq 'started') {
305
306 $self->next_state_started($sid, $cd, $sd, $lrm_res);
307
308 } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') {
309
310 $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res);
311
312 } elsif ($last_state eq 'fence') {
313
314 # do nothing here - wait until fenced
315
316 } elsif ($last_state eq 'request_stop') {
317
318 $self->next_state_request_stop($sid, $cd, $sd, $lrm_res);
319
320 } elsif ($last_state eq 'error') {
321
322 # fixme:
323
324 } else {
325
326 die "unknown service state '$last_state'";
327 }
328
329 $repeat = 1 if $sd->{state} ne $last_state;
330 }
331
332 # handle fencing
333 my $fenced_nodes = {};
334 foreach my $sid (keys %$ss) {
335 my $sd = $ss->{$sid};
336 next if $sd->{state} ne 'fence';
337
338 if (!defined($fenced_nodes->{$sd->{node}})) {
339 $fenced_nodes->{$sd->{node}} = $ns->fence_node($sd->{node}) || 0;
340 }
341
342 next if !$fenced_nodes->{$sd->{node}};
343
344 # node fence was sucessful - mark service as stopped
345 &$change_service_state($self, $sid, 'stopped');
346 }
347
348 last if !$repeat;
349 }
350
351 # remove stale services
352 # fixme:
353
354 $self->flush_master_status();
355 }
356
357 # functions to compute next service states
358 # $cd: service configuration data (read only)
359 # $sd: service status data (read only)
360 #
361 # Note: use change_service_state() to alter state
362 #
363
364 sub next_state_request_stop {
365 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
366
367 my $haenv = $self->{haenv};
368 my $ns = $self->{ns};
369
370 # check result from LRM daemon
371 if ($lrm_res) {
372 my $exit_code = $lrm_res->{exit_code};
373 if ($exit_code == 0) {
374 &$change_service_state($self, $sid, 'stopped');
375 return;
376 } else {
377 &$change_service_state($self, $sid, 'error'); # fixme: what state?
378 return;
379 }
380 }
381
382 if (!$ns->node_is_online($sd->{node})) {
383 &$change_service_state($self, $sid, 'fence');
384 return;
385 }
386 }
387
388 sub next_state_migrate_relocate {
389 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
390
391 my $haenv = $self->{haenv};
392 my $ns = $self->{ns};
393
394 # check result from LRM daemon
395 if ($lrm_res) {
396 my $exit_code = $lrm_res->{exit_code};
397 if ($exit_code == 0) {
398 &$change_service_state($self, $sid, 'started', node => $sd->{target});
399 return;
400 } else {
401 $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)");
402 &$change_service_state($self, $sid, 'started', node => $sd->{node});
403 return;
404 }
405 }
406
407 if (!$ns->node_is_online($sd->{node})) {
408 &$change_service_state($self, $sid, 'fence');
409 return;
410 }
411 }
412
413
414 sub next_state_stopped {
415 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
416
417 my $haenv = $self->{haenv};
418 my $ns = $self->{ns};
419
420 if ($sd->{node} ne $cd->{node}) {
421 # this can happen if we fence a node with active migrations
422 # hack: modify $sd (normally this should be considered read-only)
423 $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node}");
424 $sd->{node} = $cd->{node};
425 }
426
427 if ($sd->{cmd}) {
428 my ($cmd, $target) = @{$sd->{cmd}};
429 delete $sd->{cmd};
430
431 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
432 if (!$ns->node_is_online($target)) {
433 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
434 } elsif ($sd->{node} eq $target) {
435 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
436 } else {
437 $haenv->change_service_location($sid, $target);
438 $cd->{node} = $sd->{node} = $target; # fixme: $sd is read-only??!!
439 $haenv->log('info', "$cmd service '$sid' to node '$target' (stopped)");
440 }
441 } else {
442 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
443 }
444 }
445
446 if ($cd->{state} eq 'disabled') {
447 # do nothing
448 return;
449 }
450
451 if ($cd->{state} eq 'enabled') {
452 if (my $node = select_service_node($self->{groups}, $self->{online_node_usage}, $cd, $sd->{node})) {
453 if ($node && ($sd->{node} ne $node)) {
454 $haenv->change_service_location($sid, $node);
455 }
456 &$change_service_state($self, $sid, 'started', node => $node);
457 } else {
458 # fixme: warn
459 }
460
461 return;
462 }
463
464 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
465 }
466
467 sub next_state_started {
468 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
469
470 my $haenv = $self->{haenv};
471 my $ns = $self->{ns};
472
473 if (!$ns->node_is_online($sd->{node})) {
474
475 &$change_service_state($self, $sid, 'fence');
476 return;
477 }
478
479 if ($cd->{state} eq 'disabled') {
480 &$change_service_state($self, $sid, 'request_stop');
481 return;
482 }
483
484 if ($cd->{state} eq 'enabled') {
485
486 if ($sd->{cmd}) {
487 my ($cmd, $target) = @{$sd->{cmd}};
488 delete $sd->{cmd};
489
490 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
491 if (!$ns->node_is_online($target)) {
492 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
493 } elsif ($sd->{node} eq $target) {
494 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
495 } else {
496 $haenv->log('info', "$cmd service '$sid' to node '$target' (running)");
497 &$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target);
498 }
499 } else {
500 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
501 }
502 } else {
503
504 my $try_next = 0;
505 if ($lrm_res && ($lrm_res->{exit_code} != 0)) { # fixme: other exit codes?
506 $try_next = 1;
507 }
508
509 my $node = select_service_node($self->{groups}, $self->{online_node_usage},
510 $cd, $sd->{node}, $try_next);
511
512 if ($node && ($sd->{node} ne $node)) {
513 $haenv->log('info', "migrate service '$sid' to node '$node' (running)");
514 &$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node);
515 } else {
516 # do nothing
517 }
518 }
519
520 return;
521 }
522
523 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
524 }
525
526 1;