]> git.proxmox.com Git - pve-ha-manager.git/blob - src/PVE/HA/LRM.pm
avoid out of sync command execution in LRM
[pve-ha-manager.git] / src / PVE / HA / LRM.pm
1 package PVE::HA::LRM;
2
3 # Local Resource Manager
4
5 use strict;
6 use warnings;
7 use Data::Dumper;
8 use POSIX qw(:sys_wait_h);
9
10 use PVE::SafeSyslog;
11 use PVE::Tools;
12 use PVE::HA::Tools ':exit_codes';
13 use PVE::HA::Resources;
14
15 # Server can have several states:
16
17 my $valid_states = {
18 wait_for_agent_lock => "waiting for agent lock",
19 active => "got agent_lock",
20 lost_agent_lock => "lost agent_lock",
21 };
22
23 sub new {
24 my ($this, $haenv) = @_;
25
26 my $class = ref($this) || $this;
27
28 my $self = bless {
29 haenv => $haenv,
30 status => { state => 'startup' },
31 workers => {},
32 results => {},
33 restart_tries => {},
34 shutdown_request => 0,
35 shutdown_errors => 0,
36 # mode can be: active, reboot, shutdown, restart
37 mode => 'active',
38 }, $class;
39
40 $self->set_local_status({ state => 'wait_for_agent_lock' });
41
42 return $self;
43 }
44
45 sub shutdown_request {
46 my ($self) = @_;
47
48 return if $self->{shutdown_request}; # already in shutdown mode
49
50 my $haenv = $self->{haenv};
51
52 my $nodename = $haenv->nodename();
53
54 my $shutdown = $haenv->is_node_shutdown();
55
56 if ($shutdown) {
57 $haenv->log('info', "shutdown LRM, stop all services");
58 $self->{mode} = 'shutdown';
59
60 # queue stop jobs for all services
61
62 my $ss = $self->{service_status};
63
64 foreach my $sid (keys %$ss) {
65 my $sd = $ss->{$sid};
66 next if !$sd->{node};
67 next if $sd->{node} ne $nodename;
68 # Note: use undef uid to mark shutdown/stop jobs
69 $self->queue_resource_command($sid, undef, 'request_stop');
70 }
71
72 } else {
73 $haenv->log('info', "restart LRM, freeze all services");
74 $self->{mode} = 'restart';
75 }
76
77 $self->{shutdown_request} = 1;
78
79 eval { $self->update_lrm_status(); };
80 if (my $err = $@) {
81 $self->log('err', "unable to update lrm status file - $err");
82 }
83 }
84
85 sub get_local_status {
86 my ($self) = @_;
87
88 return $self->{status};
89 }
90
91 sub set_local_status {
92 my ($self, $new) = @_;
93
94 die "invalid state '$new->{state}'" if !$valid_states->{$new->{state}};
95
96 my $haenv = $self->{haenv};
97
98 my $old = $self->{status};
99
100 # important: only update if if really changed
101 return if $old->{state} eq $new->{state};
102
103 $haenv->log('info', "status change $old->{state} => $new->{state}");
104
105 $new->{state_change_time} = $haenv->get_time();
106
107 $self->{status} = $new;
108 }
109
110 sub update_lrm_status {
111 my ($self) = @_;
112
113 my $haenv = $self->{haenv};
114
115 return 0 if !$haenv->quorate();
116
117 my $lrm_status = {
118 state => $self->{status}->{state},
119 mode => $self->{mode},
120 results => $self->{results},
121 timestamp => $haenv->get_time(),
122 };
123
124 eval { $haenv->write_lrm_status($lrm_status); };
125 if (my $err = $@) {
126 $haenv->log('err', "unable to write lrm status file - $err");
127 return 0;
128 }
129
130 return 1;
131 }
132
133 sub get_protected_ha_agent_lock {
134 my ($self) = @_;
135
136 my $haenv = $self->{haenv};
137
138 my $count = 0;
139 my $starttime = $haenv->get_time();
140
141 for (;;) {
142
143 if ($haenv->get_ha_agent_lock()) {
144 if ($self->{ha_agent_wd}) {
145 $haenv->watchdog_update($self->{ha_agent_wd});
146 } else {
147 my $wfh = $haenv->watchdog_open();
148 $self->{ha_agent_wd} = $wfh;
149 }
150 return 1;
151 }
152
153 last if ++$count > 5; # try max 5 time
154
155 my $delay = $haenv->get_time() - $starttime;
156 last if $delay > 5; # for max 5 seconds
157
158 $haenv->sleep(1);
159 }
160
161 return 0;
162 }
163
164 sub active_service_count {
165 my ($self) = @_;
166
167 my $haenv = $self->{haenv};
168
169 my $nodename = $haenv->nodename();
170
171 my $ss = $self->{service_status};
172
173 my $count = 0;
174
175 foreach my $sid (keys %$ss) {
176 my $sd = $ss->{$sid};
177 next if !$sd->{node};
178 next if $sd->{node} ne $nodename;
179 my $req_state = $sd->{state};
180 next if !defined($req_state);
181 next if $req_state eq 'stopped';
182 next if $req_state eq 'freeze';
183
184 $count++;
185 }
186
187 return $count;
188 }
189
190 my $wrote_lrm_status_at_startup = 0;
191
192 sub do_one_iteration {
193 my ($self) = @_;
194
195 my $haenv = $self->{haenv};
196
197 if (!$wrote_lrm_status_at_startup) {
198 if ($self->update_lrm_status()) {
199 $wrote_lrm_status_at_startup = 1;
200 } else {
201 # do nothing
202 $haenv->sleep(5);
203 return $self->{shutdown_request} ? 0 : 1;
204 }
205 }
206
207 my $status = $self->get_local_status();
208 my $state = $status->{state};
209
210 my $ms = $haenv->read_manager_status();
211 $self->{service_status} = $ms->{service_status} || {};
212
213 my $fence_request = PVE::HA::Tools::count_fenced_services($self->{service_status}, $haenv->nodename());
214
215 # do state changes first
216
217 my $ctime = $haenv->get_time();
218
219 if ($state eq 'wait_for_agent_lock') {
220
221 my $service_count = $self->active_service_count();
222
223 if (!$fence_request && $service_count && $haenv->quorate()) {
224 if ($self->get_protected_ha_agent_lock()) {
225 $self->set_local_status({ state => 'active' });
226 }
227 }
228
229 } elsif ($state eq 'lost_agent_lock') {
230
231 if (!$fence_request && $haenv->quorate()) {
232 if ($self->get_protected_ha_agent_lock()) {
233 $self->set_local_status({ state => 'active' });
234 }
235 }
236
237 } elsif ($state eq 'active') {
238
239 if ($fence_request) {
240 $haenv->log('err', "node need to be fenced - releasing agent_lock\n");
241 $self->set_local_status({ state => 'lost_agent_lock'});
242 } elsif (!$self->get_protected_ha_agent_lock()) {
243 $self->set_local_status({ state => 'lost_agent_lock'});
244 }
245 }
246
247 $status = $self->get_local_status();
248 $state = $status->{state};
249
250 # do work
251
252 if ($state eq 'wait_for_agent_lock') {
253
254 return 0 if $self->{shutdown_request};
255
256 $self->update_lrm_status();
257
258 $haenv->sleep(5);
259
260 } elsif ($state eq 'active') {
261
262 my $startime = $haenv->get_time();
263
264 my $max_time = 10;
265
266 my $shutdown = 0;
267
268 # do work (max_time seconds)
269 eval {
270 # fixme: set alert timer
271
272 if ($self->{shutdown_request}) {
273
274 if ($self->{mode} eq 'restart') {
275
276 my $service_count = $self->active_service_count();
277
278 if ($service_count == 0) {
279
280 if ($self->run_workers() == 0) {
281 if ($self->{ha_agent_wd}) {
282 $haenv->watchdog_close($self->{ha_agent_wd});
283 delete $self->{ha_agent_wd};
284 }
285
286 $shutdown = 1;
287
288 # restart with no or freezed services, release the lock
289 $haenv->release_ha_agent_lock();
290 }
291 }
292 } else {
293
294 if ($self->run_workers() == 0) {
295 if ($self->{shutdown_errors} == 0) {
296 if ($self->{ha_agent_wd}) {
297 $haenv->watchdog_close($self->{ha_agent_wd});
298 delete $self->{ha_agent_wd};
299 }
300
301 # shutdown with all services stopped thus release the lock
302 $haenv->release_ha_agent_lock();
303 }
304
305 $shutdown = 1;
306 }
307 }
308 } else {
309
310 $self->manage_resources();
311
312 }
313 };
314 if (my $err = $@) {
315 $haenv->log('err', "got unexpected error - $err");
316 }
317
318 $self->update_lrm_status();
319
320 return 0 if $shutdown;
321
322 $haenv->sleep_until($startime + $max_time);
323
324 } elsif ($state eq 'lost_agent_lock') {
325
326 # Note: watchdog is active an will triger soon!
327
328 # so we hope to get the lock back soon!
329
330 if ($self->{shutdown_request}) {
331
332 my $service_count = $self->active_service_count();
333
334 if ($service_count > 0) {
335 $haenv->log('err', "get shutdown request in state 'lost_agent_lock' - " .
336 "detected $service_count running services");
337
338 } else {
339
340 # all services are stopped, so we can close the watchdog
341
342 if ($self->{ha_agent_wd}) {
343 $haenv->watchdog_close($self->{ha_agent_wd});
344 delete $self->{ha_agent_wd};
345 }
346
347 return 0;
348 }
349 }
350
351 $haenv->sleep(5);
352
353 } else {
354
355 die "got unexpected status '$state'\n";
356
357 }
358
359 return 1;
360 }
361
362 sub run_workers {
363 my ($self) = @_;
364
365 my $haenv = $self->{haenv};
366
367 my $starttime = $haenv->get_time();
368
369 # number of workers to start, if 0 we exec the command directly witouth forking
370 my $max_workers = $haenv->get_max_workers();
371
372 my $sc = $haenv->read_service_config();
373
374 while (($haenv->get_time() - $starttime) < 5) {
375 my $count = $self->check_active_workers();
376
377 foreach my $sid (keys %{$self->{workers}}) {
378 last if $count >= $max_workers && $max_workers > 0;
379
380 my $w = $self->{workers}->{$sid};
381 if (!$w->{pid}) {
382 # only fork if we may else call exec_resource_agent
383 # directly (e.g. for regression tests)
384 if ($max_workers > 0) {
385 my $pid = fork();
386 if (!defined($pid)) {
387 $haenv->log('err', "fork worker failed");
388 $count = 0; last; # abort, try later
389 } elsif ($pid == 0) {
390 $haenv->after_fork(); # cleanup
391
392 # do work
393 my $res = -1;
394 eval {
395 $res = $self->exec_resource_agent($sid, $sc->{$sid}, $w->{state}, $w->{target});
396 };
397 if (my $err = $@) {
398 $haenv->log('err', $err);
399 POSIX::_exit(-1);
400 }
401 POSIX::_exit($res);
402 } else {
403 $count++;
404 $w->{pid} = $pid;
405 }
406 } else {
407 my $res = -1;
408 eval {
409 $res = $self->exec_resource_agent($sid, $sc->{$sid}, $w->{state}, $w->{target});
410 $res = $res << 8 if $res > 0;
411 };
412 if (my $err = $@) {
413 $haenv->log('err', $err);
414 }
415 if (defined($w->{uid})) {
416 $self->resource_command_finished($sid, $w->{uid}, $res);
417 } else {
418 $self->stop_command_finished($sid, $res);
419 }
420 }
421 }
422 }
423
424 last if !$count;
425
426 $haenv->sleep(1);
427 }
428
429 return scalar(keys %{$self->{workers}});
430 }
431
432 sub manage_resources {
433 my ($self) = @_;
434
435 my $haenv = $self->{haenv};
436
437 my $nodename = $haenv->nodename();
438
439 my $ss = $self->{service_status};
440
441 foreach my $sid (keys %$ss) {
442 my $sd = $ss->{$sid};
443 next if !$sd->{node};
444 next if !$sd->{uid};
445 next if $sd->{node} ne $nodename;
446 my $req_state = $sd->{state};
447 next if !defined($req_state);
448 next if $req_state eq 'freeze';
449 $self->queue_resource_command($sid, $sd->{uid}, $req_state, $sd->{target});
450 }
451
452 return $self->run_workers();
453 }
454
455 sub queue_resource_command {
456 my ($self, $sid, $uid, $state, $target) = @_;
457
458 # do not queue the excatly same command twice as this may lead to
459 # an inconsistent HA state when the first command fails but the CRM
460 # does not process its failure right away and the LRM starts a second
461 # try, without the CRM knowing of it (race condition)
462 # The 'stopped' command is an exception as we do not process its result
463 # in the CRM and we want to execute it always (even with no active CRM)
464 return if $state ne 'stopped' && $uid && defined($self->{results}->{$uid});
465
466 if (my $w = $self->{workers}->{$sid}) {
467 return if $w->{pid}; # already started
468 # else, delete and overwrite queue entry with new command
469 delete $self->{workers}->{$sid};
470 }
471
472 $self->{workers}->{$sid} = {
473 sid => $sid,
474 uid => $uid,
475 state => $state,
476 };
477
478 $self->{workers}->{$sid}->{target} = $target if $target;
479 }
480
481 sub check_active_workers {
482 my ($self) = @_;
483
484 # finish/count workers
485 my $count = 0;
486 foreach my $sid (keys %{$self->{workers}}) {
487 my $w = $self->{workers}->{$sid};
488 if (my $pid = $w->{pid}) {
489 # check status
490 my $waitpid = waitpid($pid, WNOHANG);
491 if (defined($waitpid) && ($waitpid == $pid)) {
492 if (defined($w->{uid})) {
493 $self->resource_command_finished($sid, $w->{uid}, $?);
494 } else {
495 $self->stop_command_finished($sid, $?);
496 }
497 } else {
498 $count++;
499 }
500 }
501 }
502
503 return $count;
504 }
505
506 sub stop_command_finished {
507 my ($self, $sid, $status) = @_;
508
509 my $haenv = $self->{haenv};
510
511 my $w = delete $self->{workers}->{$sid};
512 return if !$w; # should not happen
513
514 my $exit_code = -1;
515
516 if ($status == -1) {
517 $haenv->log('err', "resource agent $sid finished - failed to execute");
518 } elsif (my $sig = ($status & 127)) {
519 $haenv->log('err', "resource agent $sid finished - got signal $sig");
520 } else {
521 $exit_code = ($status >> 8);
522 }
523
524 if ($exit_code != 0) {
525 $self->{shutdown_errors}++;
526 }
527 }
528
529 sub resource_command_finished {
530 my ($self, $sid, $uid, $status) = @_;
531
532 my $haenv = $self->{haenv};
533
534 my $w = delete $self->{workers}->{$sid};
535 return if !$w; # should not happen
536
537 my $exit_code = -1;
538
539 if ($status == -1) {
540 $haenv->log('err', "resource agent $sid finished - failed to execute");
541 } elsif (my $sig = ($status & 127)) {
542 $haenv->log('err', "resource agent $sid finished - got signal $sig");
543 } else {
544 $exit_code = ($status >> 8);
545 }
546
547 $exit_code = $self->handle_service_exitcode($sid, $w->{state}, $exit_code);
548
549 return if $exit_code == ETRY_AGAIN; # tell nobody, simply retry
550
551 $self->{results}->{$uid} = {
552 sid => $w->{sid},
553 state => $w->{state},
554 exit_code => $exit_code,
555 };
556
557 my $ss = $self->{service_status};
558
559 # compute hash of valid/existing uids
560 my $valid_uids = {};
561 foreach my $sid (keys %$ss) {
562 my $sd = $ss->{$sid};
563 next if !$sd->{uid};
564 $valid_uids->{$sd->{uid}} = 1;
565 }
566
567 my $results = {};
568 foreach my $id (keys %{$self->{results}}) {
569 next if !$valid_uids->{$id};
570 $results->{$id} = $self->{results}->{$id};
571 }
572 $self->{results} = $results;
573 }
574
575 # processes the exit code from a finished resource agent, so that the CRM knows
576 # if the LRM wants to retry an action based on the current recovery policies for
577 # the failed service, or the CRM itself must try to recover from the failure.
578 sub handle_service_exitcode {
579 my ($self, $sid, $cmd, $exit_code) = @_;
580
581 my $haenv = $self->{haenv};
582 my $tries = $self->{restart_tries};
583
584 my $sc = $haenv->read_service_config();
585
586 my $max_restart = 0;
587
588 if (my $cd = $sc->{$sid}) {
589 $max_restart = $cd->{max_restart};
590 }
591
592 if ($cmd eq 'started') {
593
594 if ($exit_code == SUCCESS) {
595
596 $tries->{$sid} = 0;
597
598 return $exit_code;
599
600 } elsif ($exit_code == ERROR) {
601
602 $tries->{$sid} = 0 if !defined($tries->{$sid});
603
604 if ($tries->{$sid} >= $max_restart) {
605 $haenv->log('err', "unable to start service $sid on local node".
606 " after $tries->{$sid} retries");
607 $tries->{$sid} = 0;
608 return ERROR;
609 }
610
611 $tries->{$sid}++;
612
613 $haenv->log('warning', "restart policy: retry number $tries->{$sid}" .
614 " for service '$sid'");
615 # tell CRM that we retry the start
616 return ETRY_AGAIN;
617 }
618 }
619
620 return $exit_code;
621
622 }
623
624 sub exec_resource_agent {
625 my ($self, $sid, $service_config, $cmd, @params) = @_;
626
627 # setup execution environment
628
629 $ENV{'PATH'} = '/sbin:/bin:/usr/sbin:/usr/bin';
630
631 my $haenv = $self->{haenv};
632
633 my $nodename = $haenv->nodename();
634
635 my (undef, $service_type, $service_name) = PVE::HA::Tools::parse_sid($sid);
636
637 my $plugin = PVE::HA::Resources->lookup($service_type);
638 if (!$plugin) {
639 $haenv->log('err', "service type '$service_type' not implemented");
640 return EUNKNOWN_SERVICE_TYPE;
641 }
642
643 if (!$service_config) {
644 $haenv->log('err', "missing resource configuration for '$sid'");
645 return EUNKNOWN_SERVICE;
646 }
647
648 if ($service_config->{node} ne $nodename) {
649 $haenv->log('err', "service '$sid' not on this node");
650 return EWRONG_NODE;
651 }
652
653 my $id = $service_name;
654
655 my $running = $plugin->check_running($haenv, $id);
656
657 if ($cmd eq 'started') {
658
659 return SUCCESS if $running;
660
661 $haenv->log("info", "starting service $sid");
662
663 $plugin->start($haenv, $id);
664
665 $running = $plugin->check_running($haenv, $id);
666
667 if ($running) {
668 $haenv->log("info", "service status $sid started");
669 return SUCCESS;
670 } else {
671 $haenv->log("warning", "unable to start service $sid");
672 return ERROR;
673 }
674
675 } elsif ($cmd eq 'request_stop' || $cmd eq 'stopped') {
676
677 return SUCCESS if !$running;
678
679 $haenv->log("info", "stopping service $sid");
680
681 $plugin->shutdown($haenv, $id);
682
683 $running = $plugin->check_running($haenv, $id);
684
685 if (!$running) {
686 $haenv->log("info", "service status $sid stopped");
687 return SUCCESS;
688 } else {
689 $haenv->log("info", "unable to stop stop service $sid (still running)");
690 return ERROR;
691 }
692
693 } elsif ($cmd eq 'migrate' || $cmd eq 'relocate') {
694
695 my $target = $params[0];
696 if (!defined($target)) {
697 die "$cmd '$sid' failed - missing target\n" if !defined($target);
698 return EINVALID_PARAMETER;
699 }
700
701 if ($service_config->{node} eq $target) {
702 # already there
703 return SUCCESS;
704 }
705
706 my $online = ($cmd eq 'migrate') ? 1 : 0;
707
708 my $res = $plugin->migrate($haenv, $id, $target, $online);
709
710 # something went wrong if service is still on this node
711 if (!$res) {
712 $haenv->log("err", "service $sid not moved (migration error)");
713 return ERROR;
714 }
715
716 return SUCCESS;
717
718 } elsif ($cmd eq 'error') {
719
720 if ($running) {
721 $haenv->log("err", "service $sid is in an error state while running");
722 } else {
723 $haenv->log("warning", "service $sid is not running and in an error state");
724 }
725 return SUCCESS; # error always succeeds
726
727 }
728
729 $haenv->log("err", "implement me (cmd '$cmd')");
730 return EUNKNOWN_COMMAND;
731 }
732
733
734 1;