1 package PVE
::HA
::Manager
;
5 use Digest
::MD5
qw(md5_base64);
9 use PVE
::HA
::NodeStatus
;
14 my ($this, $haenv) = @_;
16 my $class = ref($this) || $this;
18 my $ms = $haenv->read_manager_status();
20 $ms->{master_node
} = $haenv->nodename();
22 my $ns = PVE
::HA
::NodeStatus-
>new($haenv, $ms->{node_status
} || {});
24 # fixme: use separate class PVE::HA::ServiceStatus
25 my $ss = $ms->{service_status
} || {};
29 ms
=> $ms, # master status
30 ns
=> $ns, # PVE::HA::NodeStatus
31 ss
=> $ss, # service status
43 sub flush_master_status
{
46 my ($haenv, $ms, $ns, $ss) = ($self->{haenv
}, $self->{ms
}, $self->{ns
}, $self->{ss
});
48 $ms->{node_status
} = $ns->{status
};
49 $ms->{service_status
} = $ss;
50 $ms->{timestamp
} = $haenv->get_time();
52 $haenv->write_manager_status($ms);
55 sub select_service_node
{
56 my ($groups, $online_node_usage, $service_conf, $current_node, $try_next) = @_;
58 my $group = { 'nodes' => { $service_conf->{node
} => 1 } }; # default group
60 $group = $groups->{ids
}->{$service_conf->{group
}} if $service_conf->{group
} &&
61 $groups->{ids
}->{$service_conf->{group
}};
64 my $group_members = {};
65 foreach my $entry (keys %{$group->{nodes
}}) {
66 my ($node, $pri) = ($entry, 0);
67 if ($entry =~ m/^(\S+):(\d+)$/) {
68 ($node, $pri) = ($1, $2);
70 next if !defined($online_node_usage->{$node}); # offline
71 $pri_groups->{$pri}->{$node} = 1;
72 $group_members->{$node} = $pri;
76 # add non-group members to unrestricted groups (priority -1)
77 if (!$group->{restricted
}) {
79 foreach my $node (keys %$online_node_usage) {
80 next if defined($group_members->{$node});
81 $pri_groups->{$pri}->{$node} = 1;
82 $group_members->{$node} = -1;
87 my @pri_list = sort {$b <=> $a} keys %$pri_groups;
88 return undef if !scalar(@pri_list);
90 if (!$try_next && $group->{nofailback
} && defined($group_members->{$current_node})) {
94 # select node from top priority node list
96 my $top_pri = $pri_list[0];
99 $online_node_usage->{$a} <=> $online_node_usage->{$b} || $a cmp $b
100 } keys %{$pri_groups->{$top_pri}};
103 for (my $i = scalar(@nodes) - 1; $i >= 0; $i--) {
104 my $node = $nodes[$i];
105 if ($node eq $current_node) {
113 if (defined($found) && ($found < (scalar(@nodes) - 1))) {
114 return $nodes[$found + 1];
121 return $nodes[$found] if defined($found);
130 sub compute_new_uuid
{
134 return md5_base64
($state . $$ . time() . $uid_counter);
137 my $valid_service_states = {
148 sub recompute_online_node_usage
{
151 my $online_node_usage = {};
153 my $online_nodes = $self->{ns
}->list_online_nodes();
155 foreach my $node (@$online_nodes) {
156 $online_node_usage->{$node} = 0;
159 foreach my $sid (keys %{$self->{ss
}}) {
160 my $sd = $self->{ss
}->{$sid};
161 my $state = $sd->{state};
162 if (defined($online_node_usage->{$sd->{node
}})) {
163 if (($state eq 'started') || ($state eq 'request_stop') ||
164 ($state eq 'fence') || ($state eq 'freeze') || ($state eq 'error')) {
165 $online_node_usage->{$sd->{node
}}++;
166 } elsif (($state eq 'migrate') || ($state eq 'relocate')) {
167 $online_node_usage->{$sd->{target
}}++;
168 } elsif ($state eq 'stopped') {
171 die "should not be reached";
176 $self->{online_node_usage
} = $online_node_usage;
179 my $change_service_state = sub {
180 my ($self, $sid, $new_state, %params) = @_;
182 my ($haenv, $ss) = ($self->{haenv
}, $self->{ss
});
184 my $sd = $ss->{$sid} || die "no such service '$sid";
186 my $old_state = $sd->{state};
187 my $old_node = $sd->{node
};
189 die "no state change" if $old_state eq $new_state; # just to be sure
191 die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state};
193 foreach my $k (keys %$sd) { delete $sd->{$k}; };
195 $sd->{state} = $new_state;
196 $sd->{node
} = $old_node;
199 foreach my $k (sort keys %params) {
201 $text_state .= ", " if $text_state;
202 $text_state .= "$k = $v";
206 $self->recompute_online_node_usage();
208 $sd->{uid
} = compute_new_uuid
($new_state);
211 $text_state = " ($text_state)" if $text_state;
212 $haenv->log('info', "service '$sid': state changed from '${old_state}' to '${new_state}' $text_state");
215 # read LRM status for all nodes
216 sub read_lrm_status
{
219 my $nodes = $self->{ns
}->list_nodes();
220 my $haenv = $self->{haenv
};
224 foreach my $node (@$nodes) {
225 my $lrm_status = $haenv->read_lrm_status($node);
226 $modes->{$node} = $lrm_status->{mode
} || 'active';
227 foreach my $uid (keys %{$lrm_status->{results
}}) {
228 next if $results->{$uid}; # should not happen
229 $results->{$uid} = $lrm_status->{results
}->{$uid};
234 return ($results, $modes);
237 # read new crm commands and save them into crm master status
238 sub update_crm_commands
{
241 my ($haenv, $ms, $ns, $ss) = ($self->{haenv
}, $self->{ms
}, $self->{ns
}, $self->{ss
});
243 my $cmdlist = $haenv->read_crm_commands();
245 foreach my $cmd (split(/\n/, $cmdlist)) {
248 if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) {
249 my ($task, $sid, $node) = ($1, $2, $3);
250 if (my $sd = $ss->{$sid}) {
251 if (!$ns->node_is_online($node)) {
252 $haenv->log('err', "crm command error - node not online: $cmd");
254 if ($node eq $sd->{node
}) {
255 $haenv->log('info', "ignore crm command - service already on target node: $cmd");
257 $haenv->log('info', "got crm command: $cmd");
258 $ss->{$sid}->{cmd
} = [ $task, $node];
262 $haenv->log('err', "crm command error - no such service: $cmd");
266 $haenv->log('err', "unable to parse crm command: $cmd");
275 my ($haenv, $ms, $ns, $ss) = ($self->{haenv
}, $self->{ms
}, $self->{ns
}, $self->{ss
});
277 $ns->update($haenv->get_node_info());
279 if (!$ns->node_is_online($haenv->nodename())) {
280 $haenv->log('info', "master seems offline");
284 my ($lrm_results, $lrm_modes) = $self->read_lrm_status();
286 my $sc = $haenv->read_service_config();
288 $self->{groups
} = $haenv->read_group_config(); # update
290 # compute new service status
293 foreach my $sid (sort keys %$sc) {
294 next if $ss->{$sid}; # already there
295 $haenv->log('info', "adding new service '$sid' on node '$sc->{$sid}->{node}'");
296 # assume we are running to avoid relocate running service at add
297 $ss->{$sid} = { state => 'started', node
=> $sc->{$sid}->{node
},
298 uid
=> compute_new_uuid
('started') };
301 # remove stale service from manager state
302 foreach my $sid (keys %$ss) {
304 $haenv->log('info', "removing stale service '$sid' (no config)");
308 $self->update_crm_commands();
313 $self->recompute_online_node_usage();
315 foreach my $sid (keys %$ss) {
316 my $sd = $ss->{$sid};
317 my $cd = $sc->{$sid} || { state => 'disabled' };
319 my $lrm_res = $sd->{uid
} ?
$lrm_results->{$sd->{uid
}} : undef;
321 my $last_state = $sd->{state};
323 if ($last_state eq 'stopped') {
325 $self->next_state_stopped($sid, $cd, $sd, $lrm_res);
327 } elsif ($last_state eq 'started') {
329 $self->next_state_started($sid, $cd, $sd, $lrm_res);
331 } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') {
333 $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res);
335 } elsif ($last_state eq 'fence') {
337 # do nothing here - wait until fenced
339 } elsif ($last_state eq 'request_stop') {
341 $self->next_state_request_stop($sid, $cd, $sd, $lrm_res);
343 } elsif ($last_state eq 'freeze') {
345 my $lrm_mode = $sd->{node
} ?
$lrm_modes->{$sd->{node
}} : undef;
347 &$change_service_state($self, $sid, 'started')
348 if $lrm_mode && $lrm_mode eq 'active';
350 } elsif ($last_state eq 'error') {
352 $self->next_state_error($sid, $cd, $sd, $lrm_res);
356 die "unknown service state '$last_state'";
360 my $lrm_mode = $sd->{node
} ?
$lrm_modes->{$sd->{node
}} : undef;
361 $lrm_mode = 'unknown'if !$lrm_mode;
362 if (($sd->{state} eq 'started' || $sd->{state} eq 'stopped' ||
363 $sd->{state} eq 'request_stop') && ($lrm_mode ne 'active')) {
364 &$change_service_state($self, $sid, 'freeze');
367 $repeat = 1 if $sd->{state} ne $last_state;
371 my $fenced_nodes = {};
372 foreach my $sid (keys %$ss) {
373 my $sd = $ss->{$sid};
374 next if $sd->{state} ne 'fence';
376 if (!defined($fenced_nodes->{$sd->{node
}})) {
377 $fenced_nodes->{$sd->{node
}} = $ns->fence_node($sd->{node
}) || 0;
380 next if !$fenced_nodes->{$sd->{node
}};
382 # node fence was successful - mark service as stopped
383 &$change_service_state($self, $sid, 'stopped');
389 $self->flush_master_status();
392 # functions to compute next service states
393 # $cd: service configuration data (read only)
394 # $sd: service status data (read only)
396 # Note: use change_service_state() to alter state
399 sub next_state_request_stop
{
400 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
402 my $haenv = $self->{haenv
};
403 my $ns = $self->{ns
};
405 # check result from LRM daemon
407 my $exit_code = $lrm_res->{exit_code
};
408 if ($exit_code == 0) {
409 &$change_service_state($self, $sid, 'stopped');
412 &$change_service_state($self, $sid, 'error'); # fixme: what state?
417 if ($ns->node_is_offline_delayed($sd->{node
}, $fence_delay)) {
418 &$change_service_state($self, $sid, 'fence');
423 sub next_state_migrate_relocate
{
424 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
426 my $haenv = $self->{haenv
};
427 my $ns = $self->{ns
};
429 # check result from LRM daemon
431 my $exit_code = $lrm_res->{exit_code
};
432 if ($exit_code == 0) {
433 &$change_service_state($self, $sid, 'started', node
=> $sd->{target
});
436 $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)");
437 &$change_service_state($self, $sid, 'started', node
=> $sd->{node
});
442 if ($ns->node_is_offline_delayed($sd->{node
}, $fence_delay)) {
443 &$change_service_state($self, $sid, 'fence');
449 sub next_state_stopped
{
450 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
452 my $haenv = $self->{haenv
};
453 my $ns = $self->{ns
};
455 if ($sd->{node
} ne $cd->{node
}) {
456 # this can happen if we fence a node with active migrations
457 # hack: modify $sd (normally this should be considered read-only)
458 $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node}");
459 $sd->{node
} = $cd->{node
};
463 my ($cmd, $target) = @{$sd->{cmd
}};
466 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
467 if (!$ns->node_is_online($target)) {
468 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
469 } elsif ($sd->{node
} eq $target) {
470 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
473 $haenv->change_service_location($sid, $sd->{node
}, $target);
474 $cd->{node
} = $sd->{node
} = $target; # fixme: $sd is read-only??!!
475 $haenv->log('info', "$cmd service '$sid' to node '$target' (stopped)");
478 $haenv->log('err', "$cmd service '$sid' to node '$target' failed - $err");
482 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
486 if ($cd->{state} eq 'disabled') {
491 if ($cd->{state} eq 'enabled') {
492 if (my $node = select_service_node
($self->{groups
}, $self->{online_node_usage
}, $cd, $sd->{node
})) {
493 if ($node && ($sd->{node
} ne $node)) {
495 $haenv->change_service_location($sid, $sd->{node
}, $node);
496 $cd->{node
} = $sd->{node
} = $node; # fixme: $sd is read-only??!!
499 $haenv->log('err', "move service '$sid' to node '$node' failed - $err");
501 &$change_service_state($self, $sid, 'started', node
=> $node);
504 &$change_service_state($self, $sid, 'started', node
=> $node);
513 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
516 sub next_state_started
{
517 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
519 my $haenv = $self->{haenv
};
520 my $master_status = $self->{ms
};
521 my $ns = $self->{ns
};
523 if (!$ns->node_is_online($sd->{node
})) {
524 if ($ns->node_is_offline_delayed($sd->{node
}, $fence_delay)) {
525 &$change_service_state($self, $sid, 'fence');
530 if ($cd->{state} eq 'disabled') {
531 &$change_service_state($self, $sid, 'request_stop');
535 if ($cd->{state} eq 'enabled') {
538 my ($cmd, $target) = @{$sd->{cmd
}};
541 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
542 if (!$ns->node_is_online($target)) {
543 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
544 } elsif ($sd->{node
} eq $target) {
545 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
547 $haenv->log('info', "$cmd service '$sid' to node '$target' (running)");
548 &$change_service_state($self, $sid, $cmd, node
=> $sd->{node
}, target
=> $target);
551 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
557 if ($lrm_res->{exit_code
} == 1) {
559 my $try = $master_status->{relocate_trial
}->{$sid} || 0;
561 if ($try < $cd->{max_relocate
}) {
564 $try_next = 1; # tell select_service_node to relocate
566 $haenv->log('warning', "starting service $sid on node".
567 " '$sd->{node}' failed, relocating service.");
568 $master_status->{relocate_trial
}->{$sid} = $try;
572 $haenv->log('err', "recovery policy for service".
573 " $sid failed, entering error state!");
574 &$change_service_state($self, $sid, 'error');
578 } elsif ($lrm_res->{exit_code
} == 0) {
579 $master_status->{relocate_trial
}->{$sid} = 0;
583 my $node = select_service_node
($self->{groups
}, $self->{online_node_usage
},
584 $cd, $sd->{node
}, $try_next);
586 if ($node && ($sd->{node
} ne $node)) {
587 $haenv->log('info', "migrate service '$sid' to node '$node' (running)");
588 &$change_service_state($self, $sid, 'migrate', node
=> $sd->{node
}, target
=> $node);
597 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
600 sub next_state_error
{
601 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
603 my $ns = $self->{ns
};
605 if ($cd->{state} eq 'disabled') {
606 &$change_service_state($self, $sid, 'stopped');
610 if ($ns->node_is_offline_delayed($sd->{node
}, $fence_delay)) {
611 &$change_service_state($self, $sid, 'fence');