3 # Local Resource Manager
7 use POSIX
qw(:sys_wait_h);
11 use PVE
::HA
::Tools
':exit_codes';
12 use PVE
::HA
::Resources
;
14 # Server can have several states:
17 wait_for_agent_lock
=> "waiting for agent lock",
18 active
=> "got agent_lock",
19 maintenance
=> "going into maintenance",
20 lost_agent_lock
=> "lost agent_lock",
23 # we sleep ~10s per 'active' round, so if no services is available for >= 10 min we'd go in wait
24 # state giving up the watchdog and the LRM lock voluntary, ensuring the WD can do no harm
25 my $max_active_idle_rounds = 60;
28 my ($this, $haenv) = @_;
30 my $class = ref($this) || $this;
34 status
=> { state => 'startup' },
38 shutdown_request
=> 0,
40 # mode can be: active, reboot, shutdown, restart
42 cluster_state_update
=> 0,
43 active_idle_rounds
=> 0,
46 $self->set_local_status({ state => 'wait_for_agent_lock' });
51 sub shutdown_request
{
54 return if $self->{shutdown_request
}; # already in shutdown mode
56 my $haenv = $self->{haenv
};
58 my $nodename = $haenv->nodename();
60 my ($shutdown, $reboot) = $haenv->is_node_shutdown();
62 my $dc_ha_cfg = $haenv->get_ha_settings();
63 my $shutdown_policy = $dc_ha_cfg->{shutdown_policy
} // 'conditional';
65 if ($shutdown) { # don't log this on service restart, only on node shutdown
66 $haenv->log('info', "got shutdown request with shutdown policy '$shutdown_policy'");
71 if ($shutdown_policy eq 'conditional') {
72 $freeze_all = $reboot;
73 } elsif ($shutdown_policy eq 'freeze') {
75 } elsif ($shutdown_policy eq 'failover') {
77 } elsif ($shutdown_policy eq 'migrate') {
80 $haenv->log('err', "unknown shutdown policy '$shutdown_policy', fall back to conditional");
81 $freeze_all = $reboot;
85 # we get marked as unaivalable by the manager, then all services will
86 # be migrated away, we'll still have the same "can we exit" clause than
87 # a normal shutdown -> no running service on this node
88 # FIXME: after X minutes, add shutdown command for remaining services,
89 # e.g., if they have no alternative node???
91 # *always* queue stop jobs for all services if the node shuts down,
92 # independent if it's a reboot or a poweroff, else we may corrupt
93 # services or hinder node shutdown
94 my $ss = $self->{service_status
};
96 foreach my $sid (keys %$ss) {
99 next if $sd->{node
} ne $nodename;
100 # Note: use undef uid to mark shutdown/stop jobs
101 $self->queue_resource_command($sid, undef, 'request_stop');
106 my $shutdown_type = $reboot ?
'reboot' : 'shutdown';
108 $haenv->log('info', "$shutdown_type LRM, doing maintenance, removing this node from active list");
109 $self->{mode
} = 'maintenance';
110 } elsif ($freeze_all) {
111 $haenv->log('info', "$shutdown_type LRM, stop and freeze all services");
112 $self->{mode
} = 'restart';
114 $haenv->log('info', "shutdown LRM, stop all services");
115 $self->{mode
} = 'shutdown';
118 $haenv->log('info', "restart LRM, freeze all services");
119 $self->{mode
} = 'restart';
122 $self->{shutdown_request
} = $haenv->get_time();
124 eval { $self->update_lrm_status() or die "not quorate?\n"; };
126 $haenv->log('err', "unable to update lrm status file - $err");
130 sub get_local_status
{
133 return $self->{status
};
136 sub set_local_status
{
137 my ($self, $new) = @_;
139 die "invalid state '$new->{state}'" if !$valid_states->{$new->{state}};
141 my $haenv = $self->{haenv
};
143 my $old = $self->{status
};
145 # important: only update if if really changed
146 return if $old->{state} eq $new->{state};
148 $haenv->log('info', "status change $old->{state} => $new->{state}");
150 $new->{state_change_time
} = $haenv->get_time();
152 $self->{status
} = $new;
155 sub update_lrm_status
{
158 my $haenv = $self->{haenv
};
160 return 0 if !$haenv->quorate();
163 state => $self->{status
}->{state},
164 mode
=> $self->{mode
},
165 results
=> $self->{results
},
166 timestamp
=> $haenv->get_time(),
169 eval { $haenv->write_lrm_status($lrm_status); };
171 $haenv->log('err', "unable to write lrm status file - $err");
178 sub update_service_status
{
181 my $haenv = $self->{haenv
};
183 my $ms = eval { $haenv->read_manager_status(); };
185 $haenv->log('err', "updating service status from manager failed: $err");
188 $self->{service_status
} = $ms->{service_status
} || {};
193 sub get_protected_ha_agent_lock
{
196 my $haenv = $self->{haenv
};
199 my $starttime = $haenv->get_time();
203 if ($haenv->get_ha_agent_lock()) {
204 if ($self->{ha_agent_wd
}) {
205 $haenv->watchdog_update($self->{ha_agent_wd
});
207 my $wfh = $haenv->watchdog_open();
208 $self->{ha_agent_wd
} = $wfh;
213 last if ++$count > 5; # try max 5 time
215 my $delay = $haenv->get_time() - $starttime;
216 last if $delay > 5; # for max 5 seconds
224 # only cares if any service has the local node as their node, independent of which req.state it is
225 sub has_configured_service_on_local_node
{
228 my $haenv = $self->{haenv
};
229 my $nodename = $haenv->nodename();
231 my $ss = $self->{service_status
};
232 foreach my $sid (keys %$ss) {
233 my $sd = $ss->{$sid};
234 next if !$sd->{node
} || $sd->{node
} ne $nodename;
241 sub active_service_count
{
244 my $haenv = $self->{haenv
};
245 my $nodename = $haenv->nodename();
247 my $ss = $self->{service_status
};
250 foreach my $sid (keys %$ss) {
251 my $sd = $ss->{$sid};
252 next if !$sd->{node
};
253 next if $sd->{node
} ne $nodename;
254 my $req_state = $sd->{state};
255 next if !defined($req_state);
256 next if $req_state eq 'stopped';
257 # NOTE: 'ignored' ones are already dropped by the manager from service_status
258 next if $req_state eq 'freeze';
259 # erroneous services are not managed by HA, don't count them as active
260 next if $req_state eq 'error';
268 my $wrote_lrm_status_at_startup = 0;
270 sub do_one_iteration
{
273 my $haenv = $self->{haenv
};
275 $haenv->loop_start_hook();
277 $self->{cluster_state_update
} = $haenv->cluster_state_update();
279 my $res = $self->work();
281 $haenv->loop_end_hook();
286 # NOTE: this is disabling the self-fence mechanism, so it must NOT be called with active services
287 # It's normally *only* OK on graceful shutdown (with no services, or all services frozen)
288 my sub give_up_watchdog_protection
{
291 if ($self->{ha_agent_wd
}) {
292 $self->{haenv
}->watchdog_close($self->{ha_agent_wd
});
293 delete $self->{ha_agent_wd
}; # only delete after close!
300 my $haenv = $self->{haenv
};
302 if (!$wrote_lrm_status_at_startup) {
303 if ($self->update_lrm_status()) {
304 $wrote_lrm_status_at_startup = 1;
308 return $self->{shutdown_request
} ?
0 : 1;
312 my $status = $self->get_local_status();
313 my $state = $status->{state};
315 $self->update_service_status();
317 my $fence_request = PVE
::HA
::Tools
::count_fenced_services
($self->{service_status
}, $haenv->nodename());
319 # do state changes first
321 my $ctime = $haenv->get_time();
323 if ($state eq 'wait_for_agent_lock') {
325 my $service_count = $self->active_service_count();
327 if (!$fence_request && $service_count && $haenv->quorate()) {
328 if ($self->get_protected_ha_agent_lock()) {
329 $self->set_local_status({ state => 'active' });
333 } elsif ($state eq 'lost_agent_lock') {
335 if (!$fence_request && $haenv->quorate()) {
336 if ($self->get_protected_ha_agent_lock()) {
337 $self->set_local_status({ state => 'active' });
341 } elsif ($state eq 'active') {
343 if ($fence_request) {
344 $haenv->log('err', "node need to be fenced - releasing agent_lock\n");
345 $self->set_local_status({ state => 'lost_agent_lock'});
346 } elsif (!$self->get_protected_ha_agent_lock()) {
347 $self->set_local_status({ state => 'lost_agent_lock'});
348 } elsif ($self->{mode
} eq 'maintenance') {
349 $self->set_local_status({ state => 'maintenance'});
351 if (!$self->has_configured_service_on_local_node() && !$self->run_workers()) {
352 # no active service configured for this node and all (old) workers are done
353 $self->{active_idle_rounds
}++;
354 if ($self->{active_idle_rounds
} > $max_active_idle_rounds) {
355 $haenv->log('info', "node had no service configured for $max_active_idle_rounds rounds, going idle.\n");
356 # safety: no active service & no running worker for quite some time -> OK
357 $haenv->release_ha_agent_lock();
358 give_up_watchdog_protection
($self);
359 $self->set_local_status({ state => 'wait_for_agent_lock'});
360 $self->{active_idle_rounds
} = 0;
362 } elsif ($self->{active_idle_rounds
}) {
363 $self->{active_idle_rounds
} = 0;
366 } elsif ($state eq 'maintenance') {
368 if ($fence_request) {
369 $haenv->log('err', "node need to be fenced during maintenance mode - releasing agent_lock\n");
370 $self->set_local_status({ state => 'lost_agent_lock'});
371 } elsif (!$self->get_protected_ha_agent_lock()) {
372 $self->set_local_status({ state => 'lost_agent_lock'});
376 $status = $self->get_local_status();
377 $state = $status->{state};
381 if ($state eq 'wait_for_agent_lock') {
383 return 0 if $self->{shutdown_request
};
385 $self->update_lrm_status();
389 } elsif ($state eq 'active') {
391 my $startime = $haenv->get_time();
397 # do work (max_time seconds)
399 # fixme: set alert timer
401 # if we could not get the current service status there's no point
402 # in doing anything, try again next round.
403 return if !$self->update_service_status();
405 if ($self->{shutdown_request
}) {
407 if ($self->{mode
} eq 'restart') {
409 my $service_count = $self->active_service_count();
411 if ($service_count == 0) {
412 if ($self->run_workers() == 0) {
413 # safety: no active services or workers -> OK
414 give_up_watchdog_protection
($self);
417 # restart with no or freezed services, release the lock
418 $haenv->release_ha_agent_lock();
423 if ($self->run_workers() == 0) {
424 if ($self->{shutdown_errors
} == 0) {
425 # safety: no active services and LRM shutdown -> OK
426 give_up_watchdog_protection
($self);
428 # shutdown with all services stopped thus release the lock
429 $haenv->release_ha_agent_lock();
436 if (!$self->{cluster_state_update
}) {
437 # update failed but we could still renew our lock (cfs restart?),
438 # safely skip manage and expect to update just fine next round
439 $haenv->log('notice', "temporary inconsistent cluster state " .
440 "(cfs restart?), skip round");
444 $self->manage_resources();
449 $haenv->log('err', "got unexpected error - $err");
452 $self->update_lrm_status();
454 return 0 if $shutdown;
456 $haenv->sleep_until($startime + $max_time);
458 } elsif ($state eq 'lost_agent_lock') {
460 # NOTE: watchdog is active an will trigger soon!
461 # so we hope to get the lock back soon!
462 if ($self->{shutdown_request
}) {
464 my $service_count = $self->active_service_count();
466 if ($service_count > 0) {
467 $haenv->log('err', "get shutdown request in state 'lost_agent_lock' - " .
468 "detected $service_count running services");
470 if ($self->{mode
} eq 'restart') {
471 my $state_mt = $self->{status
}->{state_change_time
};
473 # watchdog should have already triggered, so either it's set
474 # set to noboot or it failed. As we are in restart mode, and
475 # have infinity stoptimeout -> exit now - we don't touch services
476 # or change state, so this is save, relatively speaking
477 if (($haenv->get_time() - $state_mt) > 90) {
478 $haenv->log('err', "lost agent lock and restart request for over 90 seconds - giving up!");
483 # safety: all services are stopped, so we can close the watchdog
484 give_up_watchdog_protection
($self);
492 } elsif ($state eq 'maintenance') {
494 my $startime = $haenv->get_time();
495 return if !$self->update_service_status();
497 # wait until all active services moved away
498 my $service_count = $self->active_service_count();
502 if ($self->{shutdown_request
}) {
503 if ($service_count == 0 && $self->run_workers() == 0) {
504 # safety: going into maintenance and all active services got moved -> OK
505 give_up_watchdog_protection
($self);
509 # restart with no or freezed services, release the lock
510 $haenv->release_ha_agent_lock();
514 $self->manage_resources() if !$exit_lrm;
516 $self->update_lrm_status();
518 return 0 if $exit_lrm;
520 $haenv->sleep_until($startime + 5);
524 die "got unexpected status '$state'\n";
534 my $haenv = $self->{haenv
};
536 my $starttime = $haenv->get_time();
538 # number of workers to start, if 0 we exec the command directly witouth forking
539 my $max_workers = $haenv->get_max_workers();
541 my $sc = $haenv->read_service_config();
543 while (($haenv->get_time() - $starttime) < 5) {
544 my $count = $self->check_active_workers();
546 foreach my $sid (sort keys %{$self->{workers
}}) {
547 last if $count >= $max_workers && $max_workers > 0;
549 my $w = $self->{workers
}->{$sid};
551 # only fork if we may else call exec_resource_agent
552 # directly (e.g. for regression tests)
553 if ($max_workers > 0) {
555 if (!defined($pid)) {
556 $haenv->log('err', "fork worker failed");
557 $count = 0; last; # abort, try later
558 } elsif ($pid == 0) {
559 $haenv->after_fork(); # cleanup
564 $res = $self->exec_resource_agent($sid, $sc->{$sid}, $w->{state}, $w->{params
});
567 $haenv->log('err', $err);
578 $res = $self->exec_resource_agent($sid, $sc->{$sid}, $w->{state}, $w->{params
});
579 $res = $res << 8 if $res > 0;
582 $haenv->log('err', $err);
584 if (defined($w->{uid
})) {
585 $self->resource_command_finished($sid, $w->{uid
}, $res);
587 $self->stop_command_finished($sid, $res);
598 return scalar(keys %{$self->{workers
}});
601 sub manage_resources
{
604 my $haenv = $self->{haenv
};
606 my $nodename = $haenv->nodename();
608 my $ss = $self->{service_status
};
610 foreach my $sid (keys %{$self->{restart_tries
}}) {
611 delete $self->{restart_tries
}->{$sid} if !$ss->{$sid};
614 foreach my $sid (keys %$ss) {
615 my $sd = $ss->{$sid};
616 next if !$sd->{node
};
618 next if $sd->{node
} ne $nodename;
619 my $req_state = $sd->{state};
620 next if !defined($req_state);
621 # can only happen for restricted groups where the failed node itself needs to be the
622 # reocvery target. Always let the master first do so, it will then marked as 'stopped' and
623 # we can just continue normally. But we must NOT do anything with it while still in recovery
624 next if $req_state eq 'recovery';
625 next if $req_state eq 'freeze';
627 $self->queue_resource_command($sid, $sd->{uid
}, $req_state, {
628 'target' => $sd->{target
},
629 'timeout' => $sd->{timeout
},
633 return $self->run_workers();
636 sub queue_resource_command
{
637 my ($self, $sid, $uid, $state, $params) = @_;
639 # do not queue the excatly same command twice as this may lead to
640 # an inconsistent HA state when the first command fails but the CRM
641 # does not process its failure right away and the LRM starts a second
642 # try, without the CRM knowing of it (race condition)
643 # The 'stopped' command is an exception as we do not process its result
644 # in the CRM and we want to execute it always (even with no active CRM)
645 return if $state ne 'stopped' && $uid && defined($self->{results
}->{$uid});
647 if (my $w = $self->{workers
}->{$sid}) {
648 return if $w->{pid
}; # already started
649 # else, delete and overwrite queue entry with new command
650 delete $self->{workers
}->{$sid};
653 $self->{workers
}->{$sid} = {
659 $self->{workers
}->{$sid}->{params
} = $params if $params;
662 sub check_active_workers
{
665 # finish/count workers
667 foreach my $sid (keys %{$self->{workers
}}) {
668 my $w = $self->{workers
}->{$sid};
669 if (my $pid = $w->{pid
}) {
671 my $waitpid = waitpid($pid, WNOHANG
);
672 if (defined($waitpid) && ($waitpid == $pid)) {
673 if (defined($w->{uid
})) {
674 $self->resource_command_finished($sid, $w->{uid
}, $?);
676 $self->stop_command_finished($sid, $?);
687 sub stop_command_finished
{
688 my ($self, $sid, $status) = @_;
690 my $haenv = $self->{haenv
};
692 my $w = delete $self->{workers
}->{$sid};
693 return if !$w; # should not happen
698 $haenv->log('err', "resource agent $sid finished - failed to execute");
699 } elsif (my $sig = ($status & 127)) {
700 $haenv->log('err', "resource agent $sid finished - got signal $sig");
702 $exit_code = ($status >> 8);
705 if ($exit_code != 0) {
706 $self->{shutdown_errors
}++;
710 sub resource_command_finished
{
711 my ($self, $sid, $uid, $status) = @_;
713 my $haenv = $self->{haenv
};
715 my $w = delete $self->{workers
}->{$sid};
716 return if !$w; # should not happen
721 $haenv->log('err', "resource agent $sid finished - failed to execute");
722 } elsif (my $sig = ($status & 127)) {
723 $haenv->log('err', "resource agent $sid finished - got signal $sig");
725 $exit_code = ($status >> 8);
728 $exit_code = $self->handle_service_exitcode($sid, $w->{state}, $exit_code);
730 return if $exit_code == ETRY_AGAIN
; # tell nobody, simply retry
732 $self->{results
}->{$uid} = {
734 state => $w->{state},
735 exit_code
=> $exit_code,
738 my $ss = $self->{service_status
};
740 # compute hash of valid/existing uids
742 foreach my $sid (keys %$ss) {
743 my $sd = $ss->{$sid};
745 $valid_uids->{$sd->{uid
}} = 1;
749 foreach my $id (keys %{$self->{results
}}) {
750 next if !$valid_uids->{$id};
751 $results->{$id} = $self->{results
}->{$id};
753 $self->{results
} = $results;
756 # processes the exit code from a finished resource agent, so that the CRM knows
757 # if the LRM wants to retry an action based on the current recovery policies for
758 # the failed service, or the CRM itself must try to recover from the failure.
759 sub handle_service_exitcode
{
760 my ($self, $sid, $cmd, $exit_code) = @_;
762 my $haenv = $self->{haenv
};
763 my $tries = $self->{restart_tries
};
765 my $sc = $haenv->read_service_config();
769 if (my $cd = $sc->{$sid}) {
770 $max_restart = $cd->{max_restart
};
773 if ($cmd eq 'started') {
775 if ($exit_code == SUCCESS
) {
781 } elsif ($exit_code == ERROR
) {
783 $tries->{$sid} = 0 if !defined($tries->{$sid});
785 if ($tries->{$sid} >= $max_restart) {
786 $haenv->log('err', "unable to start service $sid on local node".
787 " after $tries->{$sid} retries");
794 $haenv->log('warning', "restart policy: retry number $tries->{$sid}" .
795 " for service '$sid'");
796 # tell CRM that we retry the start
805 sub exec_resource_agent
{
806 my ($self, $sid, $service_config, $cmd, $params) = @_;
808 # setup execution environment
810 $ENV{'PATH'} = '/sbin:/bin:/usr/sbin:/usr/bin';
812 my $haenv = $self->{haenv
};
814 my $nodename = $haenv->nodename();
816 my (undef, $service_type, $service_name) = $haenv->parse_sid($sid);
818 my $plugin = PVE
::HA
::Resources-
>lookup($service_type);
820 $haenv->log('err', "service type '$service_type' not implemented");
821 return EUNKNOWN_SERVICE_TYPE
;
824 if (!$service_config) {
825 $haenv->log('err', "missing resource configuration for '$sid'");
826 return EUNKNOWN_SERVICE
;
829 # process error state early
830 if ($cmd eq 'error') {
831 $haenv->log('err', "service $sid is in an error state and needs manual " .
832 "intervention. Look up 'ERROR RECOVERY' in the documentation.");
834 return SUCCESS
; # error always succeeds
837 if ($service_config->{node
} ne $nodename) {
838 $haenv->log('err', "service '$sid' not on this node");
842 my $id = $service_name;
844 my $running = $plugin->check_running($haenv, $id);
846 if ($cmd eq 'started') {
848 return SUCCESS
if $running;
850 $haenv->log("info", "starting service $sid");
852 $plugin->start($haenv, $id);
854 $running = $plugin->check_running($haenv, $id);
857 $haenv->log("info", "service status $sid started");
860 $haenv->log("warning", "unable to start service $sid");
864 } elsif ($cmd eq 'request_stop' || $cmd eq 'stopped') {
866 return SUCCESS
if !$running;
868 if (defined($params->{timeout
})) {
869 $haenv->log("info", "stopping service $sid (timeout=$params->{timeout})");
871 $haenv->log("info", "stopping service $sid");
874 $plugin->shutdown($haenv, $id, $params->{timeout
});
876 $running = $plugin->check_running($haenv, $id);
879 $haenv->log("info", "service status $sid stopped");
882 $haenv->log("info", "unable to stop stop service $sid (still running)");
886 } elsif ($cmd eq 'migrate' || $cmd eq 'relocate') {
888 my $target = $params->{target
};
889 if (!defined($target)) {
890 die "$cmd '$sid' failed - missing target\n" if !defined($target);
891 return EINVALID_PARAMETER
;
894 if ($service_config->{node
} eq $target) {
899 my $online = ($cmd eq 'migrate') ?
1 : 0;
901 my $res = $plugin->migrate($haenv, $id, $target, $online);
903 # something went wrong if service is still on this node
905 $haenv->log("err", "service $sid not moved (migration error)");
913 $haenv->log("err", "implement me (cmd '$cmd')");
914 return EUNKNOWN_COMMAND
;