]>
Commit | Line | Data |
---|---|---|
5f095798 DM |
1 | package PVE::HA::LRM; |
2 | ||
3 | # Local Resource Manager | |
4 | ||
5 | use strict; | |
6 | use warnings; | |
c4a221bc DM |
7 | use Data::Dumper; |
8 | use POSIX qw(:sys_wait_h); | |
5f095798 DM |
9 | |
10 | use PVE::SafeSyslog; | |
11 | use PVE::Tools; | |
a89ff919 | 12 | use PVE::HA::Tools ':exit_codes'; |
5f095798 DM |
13 | |
14 | # Server can have several states: | |
15 | ||
16 | my $valid_states = { | |
ec911edd | 17 | wait_for_agent_lock => "waiting for agent lock", |
0bba8f60 | 18 | active => "got agent_lock", |
5f095798 DM |
19 | lost_agent_lock => "lost agent_lock", |
20 | }; | |
21 | ||
22 | sub new { | |
23 | my ($this, $haenv) = @_; | |
24 | ||
25 | my $class = ref($this) || $this; | |
26 | ||
27 | my $self = bless { | |
28 | haenv => $haenv, | |
29 | status => { state => 'startup' }, | |
c4a221bc DM |
30 | workers => {}, |
31 | results => {}, | |
ea4443cc | 32 | restart_tries => {}, |
067cdf33 | 33 | shutdown_request => 0, |
9c7d068b DM |
34 | # mode can be: active, reboot, shutdown, restart |
35 | mode => 'active', | |
5f095798 DM |
36 | }, $class; |
37 | ||
b0bf08a9 | 38 | $self->set_local_status({ state => 'wait_for_agent_lock' }); |
9c7d068b | 39 | |
5f095798 DM |
40 | return $self; |
41 | } | |
42 | ||
43 | sub shutdown_request { | |
44 | my ($self) = @_; | |
45 | ||
499f06e3 DM |
46 | my $haenv = $self->{haenv}; |
47 | ||
cde77779 | 48 | my $shutdown = $haenv->is_node_shutdown(); |
499f06e3 DM |
49 | |
50 | if ($shutdown) { | |
51 | $haenv->log('info', "shutdown LRM, stop all services"); | |
52 | $self->{mode} = 'shutdown'; | |
53 | } else { | |
54 | $haenv->log('info', "restart LRM, freeze all services"); | |
55 | $self->{mode} = 'restart'; | |
56 | } | |
9c7d068b | 57 | |
499f06e3 | 58 | $self->{shutdown_request} = 1; |
9c7d068b DM |
59 | |
60 | eval { $self->update_lrm_status(); }; | |
61 | if (my $err = $@) { | |
5bd7aa54 | 62 | $self->log('err', "unable to update lrm status file - $err"); |
9c7d068b | 63 | } |
5f095798 DM |
64 | } |
65 | ||
66 | sub get_local_status { | |
67 | my ($self) = @_; | |
68 | ||
69 | return $self->{status}; | |
70 | } | |
71 | ||
72 | sub set_local_status { | |
73 | my ($self, $new) = @_; | |
74 | ||
75 | die "invalid state '$new->{state}'" if !$valid_states->{$new->{state}}; | |
76 | ||
77 | my $haenv = $self->{haenv}; | |
78 | ||
79 | my $old = $self->{status}; | |
80 | ||
81 | # important: only update if if really changed | |
82 | return if $old->{state} eq $new->{state}; | |
83 | ||
0bba8f60 | 84 | $haenv->log('info', "status change $old->{state} => $new->{state}"); |
5f095798 DM |
85 | |
86 | $new->{state_change_time} = $haenv->get_time(); | |
87 | ||
88 | $self->{status} = $new; | |
89 | } | |
90 | ||
9c7d068b DM |
91 | sub update_lrm_status { |
92 | my ($self) = @_; | |
93 | ||
5bd7aa54 DM |
94 | my $haenv = $self->{haenv}; |
95 | ||
79829202 DM |
96 | return 0 if !$haenv->quorate(); |
97 | ||
9c7d068b DM |
98 | my $lrm_status = { |
99 | mode => $self->{mode}, | |
100 | results => $self->{results}, | |
aa330d1c | 101 | timestamp => $haenv->get_time(), |
9c7d068b DM |
102 | }; |
103 | ||
5bd7aa54 DM |
104 | eval { $haenv->write_lrm_status($lrm_status); }; |
105 | if (my $err = $@) { | |
106 | $haenv->log('err', "unable to write lrm status file - $err"); | |
107 | return 0; | |
108 | } | |
109 | ||
110 | return 1; | |
9c7d068b DM |
111 | } |
112 | ||
5f095798 DM |
113 | sub get_protected_ha_agent_lock { |
114 | my ($self) = @_; | |
115 | ||
116 | my $haenv = $self->{haenv}; | |
117 | ||
118 | my $count = 0; | |
119 | my $starttime = $haenv->get_time(); | |
120 | ||
121 | for (;;) { | |
122 | ||
123 | if ($haenv->get_ha_agent_lock()) { | |
124 | if ($self->{ha_agent_wd}) { | |
125 | $haenv->watchdog_update($self->{ha_agent_wd}); | |
126 | } else { | |
127 | my $wfh = $haenv->watchdog_open(); | |
128 | $self->{ha_agent_wd} = $wfh; | |
129 | } | |
130 | return 1; | |
131 | } | |
132 | ||
133 | last if ++$count > 5; # try max 5 time | |
134 | ||
135 | my $delay = $haenv->get_time() - $starttime; | |
136 | last if $delay > 5; # for max 5 seconds | |
137 | ||
138 | $haenv->sleep(1); | |
139 | } | |
140 | ||
141 | return 0; | |
142 | } | |
143 | ||
546e2f1f DM |
144 | sub active_service_count { |
145 | my ($self) = @_; | |
146 | ||
147 | my $haenv = $self->{haenv}; | |
148 | ||
149 | my $nodename = $haenv->nodename(); | |
150 | ||
151 | my $ss = $self->{service_status}; | |
152 | ||
153 | my $count = 0; | |
154 | ||
155 | foreach my $sid (keys %$ss) { | |
156 | my $sd = $ss->{$sid}; | |
157 | next if !$sd->{node}; | |
158 | next if $sd->{node} ne $nodename; | |
159 | my $req_state = $sd->{state}; | |
160 | next if !defined($req_state); | |
161 | next if $req_state eq 'stopped'; | |
9c7d068b | 162 | next if $req_state eq 'freeze'; |
546e2f1f DM |
163 | |
164 | $count++; | |
165 | } | |
166 | ||
167 | return $count; | |
168 | } | |
5bd7aa54 DM |
169 | |
170 | my $wrote_lrm_status_at_startup = 0; | |
171 | ||
5f095798 DM |
172 | sub do_one_iteration { |
173 | my ($self) = @_; | |
174 | ||
175 | my $haenv = $self->{haenv}; | |
176 | ||
c5ec095f | 177 | if (!$wrote_lrm_status_at_startup) { |
79829202 | 178 | if ($self->update_lrm_status()) { |
c5ec095f DM |
179 | $wrote_lrm_status_at_startup = 1; |
180 | } else { | |
181 | # do nothing | |
182 | $haenv->sleep(5); | |
183 | return $self->{shutdown_request} ? 0 : 1; | |
184 | } | |
5bd7aa54 DM |
185 | } |
186 | ||
5f095798 DM |
187 | my $status = $self->get_local_status(); |
188 | my $state = $status->{state}; | |
189 | ||
067cdf33 DM |
190 | my $ms = $haenv->read_manager_status(); |
191 | $self->{service_status} = $ms->{service_status} || {}; | |
192 | ||
49777d09 | 193 | my $fence_request = PVE::HA::Tools::count_fenced_services($self->{service_status}, $haenv->nodename()); |
067cdf33 | 194 | |
5f095798 DM |
195 | # do state changes first |
196 | ||
197 | my $ctime = $haenv->get_time(); | |
198 | ||
b0bf08a9 | 199 | if ($state eq 'wait_for_agent_lock') { |
5f095798 | 200 | |
546e2f1f | 201 | my $service_count = $self->active_service_count(); |
5f095798 | 202 | |
067cdf33 | 203 | if (!$fence_request && $service_count && $haenv->quorate()) { |
0bba8f60 DM |
204 | if ($self->get_protected_ha_agent_lock()) { |
205 | $self->set_local_status({ state => 'active' }); | |
5f095798 DM |
206 | } |
207 | } | |
208 | ||
209 | } elsif ($state eq 'lost_agent_lock') { | |
210 | ||
067cdf33 | 211 | if (!$fence_request && $haenv->quorate()) { |
0bba8f60 DM |
212 | if ($self->get_protected_ha_agent_lock()) { |
213 | $self->set_local_status({ state => 'active' }); | |
5f095798 DM |
214 | } |
215 | } | |
216 | ||
0bba8f60 | 217 | } elsif ($state eq 'active') { |
5f095798 | 218 | |
067cdf33 DM |
219 | if ($fence_request) { |
220 | $haenv->log('err', "node need to be fenced - releasing agent_lock\n"); | |
221 | $self->set_local_status({ state => 'lost_agent_lock'}); | |
222 | } elsif (!$self->get_protected_ha_agent_lock()) { | |
5f095798 DM |
223 | $self->set_local_status({ state => 'lost_agent_lock'}); |
224 | } | |
225 | } | |
226 | ||
227 | $status = $self->get_local_status(); | |
228 | $state = $status->{state}; | |
229 | ||
230 | # do work | |
231 | ||
232 | if ($state eq 'wait_for_agent_lock') { | |
233 | ||
234 | return 0 if $self->{shutdown_request}; | |
79829202 DM |
235 | |
236 | $self->update_lrm_status(); | |
237 | ||
5f095798 DM |
238 | $haenv->sleep(5); |
239 | ||
0bba8f60 | 240 | } elsif ($state eq 'active') { |
5f095798 DM |
241 | |
242 | my $startime = $haenv->get_time(); | |
243 | ||
244 | my $max_time = 10; | |
245 | ||
246 | my $shutdown = 0; | |
247 | ||
248 | # do work (max_time seconds) | |
249 | eval { | |
250 | # fixme: set alert timer | |
251 | ||
252 | if ($self->{shutdown_request}) { | |
253 | ||
499f06e3 | 254 | if ($self->{mode} eq 'restart') { |
5f095798 | 255 | |
499f06e3 | 256 | my $service_count = $self->active_service_count(); |
5f095798 | 257 | |
499f06e3 | 258 | if ($service_count == 0) { |
5f095798 | 259 | |
499f06e3 DM |
260 | if ($self->{ha_agent_wd}) { |
261 | $haenv->watchdog_close($self->{ha_agent_wd}); | |
262 | delete $self->{ha_agent_wd}; | |
263 | } | |
5f095798 | 264 | |
499f06e3 DM |
265 | $shutdown = 1; |
266 | } | |
267 | } else { | |
268 | # fixme: stop all services | |
5f095798 DM |
269 | $shutdown = 1; |
270 | } | |
c4a221bc | 271 | } else { |
c4a221bc DM |
272 | |
273 | $self->manage_resources(); | |
067cdf33 | 274 | |
5f095798 DM |
275 | } |
276 | }; | |
277 | if (my $err = $@) { | |
278 | $haenv->log('err', "got unexpected error - $err"); | |
279 | } | |
280 | ||
79829202 DM |
281 | $self->update_lrm_status(); |
282 | ||
5f095798 DM |
283 | return 0 if $shutdown; |
284 | ||
285 | $haenv->sleep_until($startime + $max_time); | |
286 | ||
287 | } elsif ($state eq 'lost_agent_lock') { | |
288 | ||
289 | # Note: watchdog is active an will triger soon! | |
290 | ||
291 | # so we hope to get the lock back soon! | |
292 | ||
293 | if ($self->{shutdown_request}) { | |
294 | ||
546e2f1f | 295 | my $service_count = $self->active_service_count(); |
5f095798 | 296 | |
546e2f1f | 297 | if ($service_count > 0) { |
5f095798 | 298 | $haenv->log('err', "get shutdown request in state 'lost_agent_lock' - " . |
546e2f1f | 299 | "detected $service_count running services"); |
5f095798 | 300 | |
546e2f1f | 301 | } else { |
5f095798 | 302 | |
546e2f1f | 303 | # all services are stopped, so we can close the watchdog |
5f095798 | 304 | |
546e2f1f DM |
305 | if ($self->{ha_agent_wd}) { |
306 | $haenv->watchdog_close($self->{ha_agent_wd}); | |
307 | delete $self->{ha_agent_wd}; | |
308 | } | |
309 | ||
310 | return 0; | |
5f095798 | 311 | } |
5f095798 DM |
312 | } |
313 | ||
b0bf08a9 DM |
314 | $haenv->sleep(5); |
315 | ||
5f095798 DM |
316 | } else { |
317 | ||
318 | die "got unexpected status '$state'\n"; | |
319 | ||
320 | } | |
321 | ||
322 | return 1; | |
323 | } | |
324 | ||
c4a221bc DM |
325 | sub manage_resources { |
326 | my ($self) = @_; | |
327 | ||
328 | my $haenv = $self->{haenv}; | |
329 | ||
330 | my $nodename = $haenv->nodename(); | |
331 | ||
c4a221bc DM |
332 | my $ss = $self->{service_status}; |
333 | ||
334 | foreach my $sid (keys %$ss) { | |
335 | my $sd = $ss->{$sid}; | |
336 | next if !$sd->{node}; | |
337 | next if !$sd->{uid}; | |
338 | next if $sd->{node} ne $nodename; | |
339 | my $req_state = $sd->{state}; | |
340 | next if !defined($req_state); | |
9c7d068b | 341 | next if $req_state eq 'freeze'; |
c4a221bc | 342 | eval { |
e88469ba | 343 | $self->queue_resource_command($sid, $sd->{uid}, $req_state, $sd->{target}); |
c4a221bc DM |
344 | }; |
345 | if (my $err = $@) { | |
f31b7e94 | 346 | $haenv->log('err', "unable to run resource agent for '$sid' - $err"); # fixme |
c4a221bc DM |
347 | } |
348 | } | |
349 | ||
f31b7e94 | 350 | my $starttime = $haenv->get_time(); |
c4a221bc DM |
351 | |
352 | # start workers | |
353 | my $max_workers = 4; | |
354 | ||
6dbf93a0 | 355 | my $sc = $haenv->read_service_config(); |
f31b7e94 DM |
356 | |
357 | while (($haenv->get_time() - $starttime) < 5) { | |
c4a221bc DM |
358 | my $count = $self->check_active_workers(); |
359 | ||
360 | foreach my $sid (keys %{$self->{workers}}) { | |
361 | last if $count >= $max_workers; | |
362 | my $w = $self->{workers}->{$sid}; | |
6dbf93a0 DM |
363 | my $cd = $sc->{$sid}; |
364 | if (!$cd) { | |
f31b7e94 | 365 | $haenv->log('err', "missing resource configuration for '$sid'"); |
6dbf93a0 DM |
366 | next; |
367 | } | |
c4a221bc | 368 | if (!$w->{pid}) { |
f31b7e94 DM |
369 | if ($haenv->can_fork()) { |
370 | my $pid = fork(); | |
371 | if (!defined($pid)) { | |
372 | $haenv->log('err', "fork worker failed"); | |
373 | $count = 0; last; # abort, try later | |
374 | } elsif ($pid == 0) { | |
375 | # do work | |
376 | my $res = -1; | |
377 | eval { | |
378 | $res = $haenv->exec_resource_agent($sid, $cd, $w->{state}, $w->{target}); | |
379 | }; | |
380 | if (my $err = $@) { | |
381 | $haenv->log('err', $err); | |
382 | POSIX::_exit(-1); | |
383 | } | |
384 | POSIX::_exit($res); | |
385 | } else { | |
386 | $count++; | |
387 | $w->{pid} = $pid; | |
388 | } | |
389 | } else { | |
c4a221bc DM |
390 | my $res = -1; |
391 | eval { | |
6dbf93a0 | 392 | $res = $haenv->exec_resource_agent($sid, $cd, $w->{state}, $w->{target}); |
c4a221bc DM |
393 | }; |
394 | if (my $err = $@) { | |
f31b7e94 DM |
395 | $haenv->log('err', $err); |
396 | } | |
397 | $self->resource_command_finished($sid, $w->{uid}, $res); | |
c4a221bc DM |
398 | } |
399 | } | |
400 | } | |
401 | ||
402 | last if !$count; | |
403 | ||
f31b7e94 | 404 | $haenv->sleep(1); |
c4a221bc DM |
405 | } |
406 | } | |
407 | ||
408 | # fixme: use a queue an limit number of parallel workers? | |
409 | sub queue_resource_command { | |
e88469ba | 410 | my ($self, $sid, $uid, $state, $target) = @_; |
c4a221bc DM |
411 | |
412 | if (my $w = $self->{workers}->{$sid}) { | |
413 | return if $w->{pid}; # already started | |
414 | # else, delete and overwrite queue entry with new command | |
415 | delete $self->{workers}->{$sid}; | |
416 | } | |
417 | ||
418 | $self->{workers}->{$sid} = { | |
419 | sid => $sid, | |
420 | uid => $uid, | |
421 | state => $state, | |
422 | }; | |
e88469ba DM |
423 | |
424 | $self->{workers}->{$sid}->{target} = $target if $target; | |
c4a221bc DM |
425 | } |
426 | ||
427 | sub check_active_workers { | |
428 | my ($self) = @_; | |
429 | ||
430 | # finish/count workers | |
431 | my $count = 0; | |
432 | foreach my $sid (keys %{$self->{workers}}) { | |
433 | my $w = $self->{workers}->{$sid}; | |
434 | if (my $pid = $w->{pid}) { | |
435 | # check status | |
436 | my $waitpid = waitpid($pid, WNOHANG); | |
437 | if (defined($waitpid) && ($waitpid == $pid)) { | |
438 | $self->resource_command_finished($sid, $w->{uid}, $?); | |
439 | } else { | |
440 | $count++; | |
441 | } | |
442 | } | |
443 | } | |
444 | ||
445 | return $count; | |
446 | } | |
447 | ||
448 | sub resource_command_finished { | |
449 | my ($self, $sid, $uid, $status) = @_; | |
450 | ||
451 | my $haenv = $self->{haenv}; | |
452 | ||
453 | my $w = delete $self->{workers}->{$sid}; | |
454 | return if !$w; # should not happen | |
455 | ||
456 | my $exit_code = -1; | |
457 | ||
458 | if ($status == -1) { | |
0f70400d | 459 | $haenv->log('err', "resource agent $sid finished - failed to execute"); |
c4a221bc | 460 | } elsif (my $sig = ($status & 127)) { |
0f70400d | 461 | $haenv->log('err', "resource agent $sid finished - got signal $sig"); |
c4a221bc DM |
462 | } else { |
463 | $exit_code = ($status >> 8); | |
c4a221bc DM |
464 | } |
465 | ||
ea4443cc TL |
466 | $exit_code = $self->handle_service_exitcode($sid, $w->{state}, $exit_code); |
467 | ||
c4a221bc DM |
468 | $self->{results}->{$uid} = { |
469 | sid => $w->{sid}, | |
470 | state => $w->{state}, | |
471 | exit_code => $exit_code, | |
472 | }; | |
473 | ||
474 | my $ss = $self->{service_status}; | |
475 | ||
476 | # compute hash of valid/existing uids | |
477 | my $valid_uids = {}; | |
478 | foreach my $sid (keys %$ss) { | |
479 | my $sd = $ss->{$sid}; | |
480 | next if !$sd->{uid}; | |
481 | $valid_uids->{$sd->{uid}} = 1; | |
482 | } | |
483 | ||
484 | my $results = {}; | |
485 | foreach my $id (keys %{$self->{results}}) { | |
486 | next if !$valid_uids->{$id}; | |
487 | $results->{$id} = $self->{results}->{$id}; | |
488 | } | |
489 | $self->{results} = $results; | |
c4a221bc DM |
490 | } |
491 | ||
ea4443cc TL |
492 | # processes the exit code from a finished resource agent, so that the CRM knows |
493 | # if the LRM wants to retry an action based on the current recovery policies for | |
494 | # the failed service, or the CRM itself must try to recover from the failure. | |
495 | sub handle_service_exitcode { | |
496 | my ($self, $sid, $cmd, $exit_code) = @_; | |
497 | ||
498 | my $haenv = $self->{haenv}; | |
499 | my $tries = $self->{restart_tries}; | |
500 | ||
501 | my $sc = $haenv->read_service_config(); | |
502 | my $cd = $sc->{$sid}; | |
503 | ||
504 | if ($cmd eq 'started') { | |
505 | ||
a89ff919 | 506 | if ($exit_code == SUCCESS) { |
ea4443cc TL |
507 | |
508 | $tries->{$sid} = 0; | |
509 | ||
510 | return $exit_code; | |
511 | ||
a89ff919 | 512 | } elsif ($exit_code == ERROR) { |
ea4443cc TL |
513 | |
514 | $tries->{$sid} = 0 if !defined($tries->{$sid}); | |
515 | ||
516 | $tries->{$sid}++; | |
517 | if ($tries->{$sid} >= $cd->{max_restart}) { | |
518 | $haenv->log('err', "unable to start service $sid on local node". | |
519 | " after $tries->{$sid} retries"); | |
520 | $tries->{$sid} = 0; | |
a89ff919 | 521 | return ERROR; |
ea4443cc TL |
522 | } |
523 | ||
a89ff919 TL |
524 | # tell CRM that we retry the start |
525 | return ETRY_AGAIN; | |
ea4443cc TL |
526 | } |
527 | } | |
528 | ||
529 | return $exit_code; | |
530 | ||
531 | } | |
532 | ||
5f095798 | 533 | 1; |