]> git.proxmox.com Git - pve-ha-manager.git/blame - PVE/HA/Manager.pm
select_service_node: try to distribute service to all nodes
[pve-ha-manager.git] / PVE / HA / Manager.pm
CommitLineData
c0bbd038
DM
1package PVE::HA::Manager;
2
3use strict;
4use warnings;
c4a221bc 5use Digest::MD5 qw(md5_base64);
c0bbd038
DM
6
7use Data::Dumper;
8
9use PVE::HA::NodeStatus;
10
11sub new {
8f0bb968 12 my ($this, $haenv) = @_;
c0bbd038
DM
13
14 my $class = ref($this) || $this;
15
8f0bb968
DM
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
59fd7207
DM
22 # fixme: use separate class PVE::HA::ServiceStatus
23 my $ss = $ms->{service_status} || {};
24
c0bbd038 25 my $self = bless {
8f0bb968
DM
26 haenv => $haenv,
27 ms => $ms, # master status
28 ns => $ns, # PVE::HA::NodeStatus
59fd7207 29 ss => $ss, # service status
c0bbd038
DM
30 }, $class;
31
32 return $self;
33}
34
d84da043
DM
35sub cleanup {
36 my ($self) = @_;
37
38 # todo: ?
39}
40
8f0bb968 41sub flush_master_status {
c0bbd038
DM
42 my ($self) = @_;
43
59fd7207 44 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
c0bbd038 45
8f0bb968 46 $ms->{node_status} = $ns->{status};
59fd7207
DM
47 $ms->{service_status} = $ss;
48
8f0bb968
DM
49 $haenv->write_manager_status($ms);
50}
c0bbd038 51
f7ccd1b3 52sub select_service_node {
270d4406 53 my ($groups, $online_node_usage, $service_conf, $current_node, $try_next) = @_;
f7ccd1b3 54
abc920b4
DM
55 my $group = { 'nodes' => $service_conf->{node} }; # default group
56
270d4406
DM
57 $group = $groups->{ids}->{$service_conf->{group}} if $service_conf->{group} &&
58 $groups->{ids}->{$service_conf->{group}};
abc920b4
DM
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, $2);
66 }
270d4406 67 next if !defined($online_node_usage->{$node}); # offline
abc920b4
DM
68 $pri_groups->{$pri}->{$node} = 1;
69 $group_members->{$node} = $pri;
70 }
f7ccd1b3 71
270d4406 72
abc920b4
DM
73 # add non-group members to unrestricted groups (priority -1)
74 if (!$group->{restricted}) {
75 my $pri = -1;
270d4406 76 foreach my $node (keys %$online_node_usage) {
abc920b4
DM
77 next if defined($group_members->{$node});
78 $pri_groups->{$pri}->{$node} = 1;
79 $group_members->{$node} = -1;
80 }
81 }
82
270d4406 83
abc920b4
DM
84 my @pri_list = sort {$b <=> $a} keys %$pri_groups;
85 return undef if !scalar(@pri_list);
270d4406 86
abc920b4
DM
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
270d4406 95 my @nodes = sort { $online_node_usage->{$a} <=> $online_node_usage->{$b} } keys %{$pri_groups->{$top_pri}};
abc920b4
DM
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 = $i;
102 last;
103 }
104 }
105
abc920b4
DM
106 if ($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 }
f7ccd1b3
DM
121}
122
c4a221bc
DM
123my $uid_counter = 0;
124
618fbeda
DM
125my $valid_service_states = {
126 stopped => 1,
127 request_stop => 1,
128 started => 1,
129 fence => 1,
130 migrate => 1,
b0fdf86a 131 relocate => 1,
618fbeda
DM
132 error => 1,
133};
134
270d4406
DM
135sub recompute_online_node_usage {
136 my ($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
4e01bc86
DM
166my $change_service_state = sub {
167 my ($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};
e4ffb299 174 my $old_node = $sd->{node};
4e01bc86
DM
175
176 die "no state change" if $old_state eq $new_state; # just to be sure
177
618fbeda
DM
178 die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state};
179
e4ffb299
DM
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 = '';
4e01bc86
DM
186 foreach my $k (keys %params) {
187 my $v = $params{$k};
e4ffb299
DM
188 $text_state .= ", " if $text_state;
189 $text_state .= "$k = $v";
4e01bc86
DM
190 $sd->{$k} = $v;
191 }
270d4406
DM
192
193 $self->recompute_online_node_usage();
194
c4a221bc
DM
195 $uid_counter++;
196 $sd->{uid} = md5_base64($new_state . $$ . time() . $uid_counter);
4e01bc86 197
e4ffb299
DM
198 $text_state = " ($text_state)" if $text_state;
199 $haenv->log('info', "service '$sid': state changed from '${old_state}' to '${new_state}' $text_state\n");
4e01bc86
DM
200};
201
332170bd 202# read LRM status for all active nodes
c4a221bc 203sub read_lrm_status {
332170bd 204 my ($self) = @_;
c4a221bc 205
332170bd 206 my $nodes = $self->{ns}->list_online_nodes();
c4a221bc
DM
207 my $haenv = $self->{haenv};
208
209 my $res = {};
210
332170bd 211 foreach my $node (@$nodes) {
c4a221bc
DM
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 $res;
220}
221
aa98a844
DM
222# read new crm commands and save them into crm master status
223sub update_crm_commands {
224 my ($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 $cmd;
232
b0fdf86a
DM
233 if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) {
234 my ($task, $sid, $node) = ($1, $2, $3);
aa98a844
DM
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");
b0fdf86a 243 $ss->{$sid}->{cmd} = [ $task, $node];
aa98a844
DM
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: $cmd");
252 }
253 }
254
255}
256
8f0bb968
DM
257sub manage {
258 my ($self) = @_;
c0bbd038 259
59fd7207 260 my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
c0bbd038 261
332170bd 262 $ns->update($haenv->get_node_info());
c79442f2
DM
263
264 if (!$ns->node_is_online($haenv->nodename())) {
265 $haenv->log('info', "master seems offline\n");
266 return;
267 }
268
332170bd 269 my $lrm_status = $self->read_lrm_status();
c4a221bc 270
f7ccd1b3
DM
271 my $sc = $haenv->read_service_config();
272
abc920b4
DM
273 $self->{groups} = $haenv->read_group_config(); # update
274
f7ccd1b3
DM
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
8456bde2 282 $ss->{$sid} = { state => 'started', node => $sc->{$sid}->{node}};
f7ccd1b3
DM
283 }
284
aa98a844
DM
285 $self->update_crm_commands();
286
c79442f2
DM
287 for (;;) {
288 my $repeat = 0;
270d4406
DM
289
290 $self->recompute_online_node_usage();
f7ccd1b3 291
c79442f2
DM
292 foreach my $sid (keys %$ss) {
293 my $sd = $ss->{$sid};
294 my $cd = $sc->{$sid} || { state => 'disabled' };
f7ccd1b3 295
a875fbe8
DM
296 my $lrm_res = $sd->{uid} ? $lrm_status->{$sd->{uid}} : undef;
297
c79442f2
DM
298 my $last_state = $sd->{state};
299
300 if ($last_state eq 'stopped') {
301
abc920b4 302 $self->next_state_stopped($sid, $cd, $sd, $lrm_res);
f7ccd1b3 303
c79442f2 304 } elsif ($last_state eq 'started') {
f7ccd1b3 305
abc920b4 306 $self->next_state_started($sid, $cd, $sd, $lrm_res);
f7ccd1b3 307
b0fdf86a 308 } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') {
f7ccd1b3 309
e88469ba
DM
310 # check result from LRM daemon
311 if ($lrm_res) {
312 my $exit_code = $lrm_res->{exit_code};
313 if ($exit_code == 0) {
314 &$change_service_state($self, $sid, 'started', node => $sd->{target});
315 } else {
316 $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)");
317 &$change_service_state($self, $sid, 'started', node => $sd->{node});
318 }
319 }
f7ccd1b3 320
c79442f2 321 } elsif ($last_state eq 'fence') {
f7ccd1b3 322
21e37ed4 323 # do nothing here - wait until fenced
f7ccd1b3 324
c79442f2 325 } elsif ($last_state eq 'request_stop') {
f7ccd1b3 326
a875fbe8
DM
327 # check result from LRM daemon
328 if ($lrm_res) {
329 my $exit_code = $lrm_res->{exit_code};
618fbeda
DM
330 if ($exit_code == 0) {
331 &$change_service_state($self, $sid, 'stopped');
332 } else {
333 &$change_service_state($self, $sid, 'error'); # fixme: what state?
334 }
618fbeda
DM
335 }
336
e88469ba
DM
337 } elsif ($last_state eq 'error') {
338
339 # fixme:
340
a875fbe8
DM
341 } else {
342
343 die "unknown service state '$last_state'";
618fbeda 344 }
21e37ed4 345
c79442f2 346 $repeat = 1 if $sd->{state} ne $last_state;
f7ccd1b3
DM
347 }
348
21e37ed4
DM
349 # handle fencing
350 my $fenced_nodes = {};
351 foreach my $sid (keys %$ss) {
352 my $sd = $ss->{$sid};
353 next if $sd->{state} ne 'fence';
354
355 if (!defined($fenced_nodes->{$sd->{node}})) {
356 $fenced_nodes->{$sd->{node}} = $ns->fence_node($sd->{node}) || 0;
357 }
358
359 next if !$fenced_nodes->{$sd->{node}};
360
361 # node fence was sucessful - mark service as stopped
362 &$change_service_state($self, $sid, 'stopped');
363 }
364
c79442f2 365 last if !$repeat;
f7ccd1b3 366 }
f7ccd1b3
DM
367
368 # remove stale services
369 # fixme:
370
8f0bb968 371 $self->flush_master_status();
c0bbd038
DM
372}
373
a875fbe8
DM
374# functions to compute next service states
375# $cd: service configuration data (read only)
376# $sd: service status data (read only)
377#
378# Note: use change_service_state() to alter state
379#
380
381sub next_state_stopped {
abc920b4 382 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
a875fbe8
DM
383
384 my $haenv = $self->{haenv};
e88469ba 385 my $ns = $self->{ns};
a875fbe8 386
ff6f1c5c
DM
387 if ($sd->{node} ne $cd->{node}) {
388 # this can happen if we fence a node with active migrations
389 # hack: modify $sd (normally this should be considered read-only)
390 $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node}");
391 $sd->{node} = $cd->{node};
392 }
393
94b7ebe2
DM
394 if ($sd->{cmd}) {
395 my ($cmd, $target) = @{$sd->{cmd}};
396 delete $sd->{cmd};
397
b0fdf86a 398 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
94b7ebe2 399 if (!$ns->node_is_online($target)) {
b0fdf86a 400 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
e88469ba 401 } elsif ($sd->{node} eq $target) {
b0fdf86a 402 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
94b7ebe2
DM
403 } else {
404 $haenv->change_service_location($sid, $target);
405 $cd->{node} = $sd->{node} = $target; # fixme: $sd is read-only??!!
b0fdf86a 406 $haenv->log('info', "$cmd service '$sid' to node '$target' (stopped)");
94b7ebe2
DM
407 }
408 } else {
409 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
410 }
411 }
412
a875fbe8
DM
413 if ($cd->{state} eq 'disabled') {
414 # do nothing
e88469ba
DM
415 return;
416 }
417
418 if ($cd->{state} eq 'enabled') {
270d4406 419 if (my $node = select_service_node($self->{groups}, $self->{online_node_usage}, $cd, $sd->{node})) {
a875fbe8
DM
420 if ($node && ($sd->{node} ne $node)) {
421 $haenv->change_service_location($sid, $node);
422 }
423 &$change_service_state($self, $sid, 'started', node => $node);
424 } else {
425 # fixme: warn
426 }
e88469ba
DM
427
428 return;
a875fbe8 429 }
e88469ba
DM
430
431 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
a875fbe8
DM
432}
433
434sub next_state_started {
abc920b4 435 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
a875fbe8
DM
436
437 my $haenv = $self->{haenv};
438 my $ns = $self->{ns};
439
440 if (!$ns->node_is_online($sd->{node})) {
441
442 &$change_service_state($self, $sid, 'fence');
e88469ba
DM
443 return;
444 }
a875fbe8 445
e88469ba
DM
446 if ($cd->{state} eq 'disabled') {
447 &$change_service_state($self, $sid, 'request_stop');
448 return;
449 }
450
451 if ($cd->{state} eq 'enabled') {
e88469ba
DM
452
453 if ($sd->{cmd}) {
454 my ($cmd, $target) = @{$sd->{cmd}};
455 delete $sd->{cmd};
456
b0fdf86a 457 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
e88469ba 458 if (!$ns->node_is_online($target)) {
b0fdf86a 459 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
e88469ba 460 } elsif ($sd->{node} eq $target) {
b0fdf86a 461 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
e88469ba 462 } else {
b0fdf86a
DM
463 $haenv->log('info', "$cmd service '$sid' to node '$target' (running)");
464 &$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target);
e88469ba 465 }
a875fbe8 466 } else {
e88469ba 467 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
a875fbe8
DM
468 }
469 } else {
b0fdf86a 470
abc920b4
DM
471 my $try_next = 0;
472 if ($lrm_res && ($lrm_res->{exit_code} != 0)) { # fixme: other exit codes?
473 $try_next = 1;
474 }
475
270d4406
DM
476 my $node = select_service_node($self->{groups}, $self->{online_node_usage},
477 $cd, $sd->{node}, $try_next);
abc920b4 478
b0fdf86a
DM
479 if ($node && ($sd->{node} ne $node)) {
480 $haenv->log('info', "migrate service '$sid' to node '$node' (running)");
481 &$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node);
482 } else {
483 # do nothing
484 }
a875fbe8 485 }
e88469ba
DM
486
487 return;
488 }
489
490 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
a875fbe8 491}
c0bbd038
DM
492
4931;