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;
51 $haenv->write_manager_status($ms);
54 sub select_service_node
{
55 my ($groups, $online_node_usage, $service_conf, $current_node, $try_next) = @_;
57 my $group = { 'nodes' => $service_conf->{node
} }; # default group
59 $group = $groups->{ids
}->{$service_conf->{group
}} if $service_conf->{group
} &&
60 $groups->{ids
}->{$service_conf->{group
}};
63 my $group_members = {};
64 foreach my $entry (PVE
::Tools
::split_list
($group->{nodes
})) {
65 my ($node, $pri) = ($entry, 0);
66 if ($entry =~ m/^(\S+):(\d+)$/) {
67 ($node, $pri) = ($1, $2);
69 next if !defined($online_node_usage->{$node}); # offline
70 $pri_groups->{$pri}->{$node} = 1;
71 $group_members->{$node} = $pri;
75 # add non-group members to unrestricted groups (priority -1)
76 if (!$group->{restricted
}) {
78 foreach my $node (keys %$online_node_usage) {
79 next if defined($group_members->{$node});
80 $pri_groups->{$pri}->{$node} = 1;
81 $group_members->{$node} = -1;
86 my @pri_list = sort {$b <=> $a} keys %$pri_groups;
87 return undef if !scalar(@pri_list);
89 if (!$try_next && $group->{nofailback
} && defined($group_members->{$current_node})) {
93 # select node from top priority node list
95 my $top_pri = $pri_list[0];
98 $online_node_usage->{$a} <=> $online_node_usage->{$b} || $a cmp $b
99 } keys %{$pri_groups->{$top_pri}};
102 for (my $i = scalar(@nodes) - 1; $i >= 0; $i--) {
103 my $node = $nodes[$i];
104 if ($node eq $current_node) {
112 if (defined($found) && ($found < (scalar(@nodes) - 1))) {
113 return $nodes[$found + 1];
120 return $nodes[$found] if defined($found);
129 sub compute_new_uuid
{
133 return md5_base64
($state . $$ . time() . $uid_counter);
136 my $valid_service_states = {
147 sub recompute_online_node_usage
{
150 my $online_node_usage = {};
152 my $online_nodes = $self->{ns
}->list_online_nodes();
154 foreach my $node (@$online_nodes) {
155 $online_node_usage->{$node} = 0;
158 foreach my $sid (keys %{$self->{ss
}}) {
159 my $sd = $self->{ss
}->{$sid};
160 my $state = $sd->{state};
161 if (defined($online_node_usage->{$sd->{node
}})) {
162 if (($state eq 'started') || ($state eq 'request_stop') ||
163 ($state eq 'fence') || ($state eq 'freeze') || ($state eq 'error')) {
164 $online_node_usage->{$sd->{node
}}++;
165 } elsif (($state eq 'migrate') || ($state eq 'relocate')) {
166 $online_node_usage->{$sd->{target
}}++;
167 } elsif ($state eq 'stopped') {
170 die "should not be reached";
175 $self->{online_node_usage
} = $online_node_usage;
178 my $change_service_state = sub {
179 my ($self, $sid, $new_state, %params) = @_;
181 my ($haenv, $ss) = ($self->{haenv
}, $self->{ss
});
183 my $sd = $ss->{$sid} || die "no such service '$sid";
185 my $old_state = $sd->{state};
186 my $old_node = $sd->{node
};
188 die "no state change" if $old_state eq $new_state; # just to be sure
190 die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state};
192 foreach my $k (keys %$sd) { delete $sd->{$k}; };
194 $sd->{state} = $new_state;
195 $sd->{node
} = $old_node;
198 foreach my $k (keys %params) {
200 $text_state .= ", " if $text_state;
201 $text_state .= "$k = $v";
205 $self->recompute_online_node_usage();
207 $sd->{uid
} = compute_new_uuid
($new_state);
210 $text_state = " ($text_state)" if $text_state;
211 $haenv->log('info', "service '$sid': state changed from '${old_state}' to '${new_state}' $text_state");
214 # read LRM status for all nodes
215 sub read_lrm_status
{
218 my $nodes = $self->{ns
}->list_nodes();
219 my $haenv = $self->{haenv
};
223 foreach my $node (@$nodes) {
224 my $lrm_status = $haenv->read_lrm_status($node);
225 $modes->{$node} = $lrm_status->{mode
} || 'active';
226 foreach my $uid (keys %{$lrm_status->{results
}}) {
227 next if $results->{$uid}; # should not happen
228 $results->{$uid} = $lrm_status->{results
}->{$uid};
233 return ($results, $modes);
236 # read new crm commands and save them into crm master status
237 sub update_crm_commands
{
240 my ($haenv, $ms, $ns, $ss) = ($self->{haenv
}, $self->{ms
}, $self->{ns
}, $self->{ss
});
242 my $cmdlist = $haenv->read_crm_commands();
244 foreach my $cmd (split(/\n/, $cmdlist)) {
247 if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) {
248 my ($task, $sid, $node) = ($1, $2, $3);
249 if (my $sd = $ss->{$sid}) {
250 if (!$ns->node_is_online($node)) {
251 $haenv->log('err', "crm command error - node not online: $cmd");
253 if ($node eq $sd->{node
}) {
254 $haenv->log('info', "ignore crm command - service already on target node: $cmd");
256 $haenv->log('info', "got crm command: $cmd");
257 $ss->{$sid}->{cmd
} = [ $task, $node];
261 $haenv->log('err', "crm command error - no such service: $cmd");
265 $haenv->log('err', "unable to parse crm command: $cmd");
274 my ($haenv, $ms, $ns, $ss) = ($self->{haenv
}, $self->{ms
}, $self->{ns
}, $self->{ss
});
276 $ns->update($haenv->get_node_info());
278 if (!$ns->node_is_online($haenv->nodename())) {
279 $haenv->log('info', "master seems offline");
283 my ($lrm_results, $lrm_modes) = $self->read_lrm_status();
285 my $sc = $haenv->read_service_config();
287 $self->{groups
} = $haenv->read_group_config(); # update
289 # compute new service status
292 foreach my $sid (sort keys %$sc) {
293 next if $ss->{$sid}; # already there
294 $haenv->log('info', "adding new service '$sid' on node '$sc->{$sid}->{node}'");
295 # assume we are running to avoid relocate running service at add
296 $ss->{$sid} = { state => 'started', node
=> $sc->{$sid}->{node
},
297 uid
=> compute_new_uuid
('started') };
300 # remove stale service from manager state
301 foreach my $sid (keys %$ss) {
303 $haenv->log('info', "removing stale service '$sid' (no config)");
307 $self->update_crm_commands();
312 $self->recompute_online_node_usage();
314 foreach my $sid (keys %$ss) {
315 my $sd = $ss->{$sid};
316 my $cd = $sc->{$sid} || { state => 'disabled' };
318 my $lrm_res = $sd->{uid
} ?
$lrm_results->{$sd->{uid
}} : undef;
320 my $last_state = $sd->{state};
322 if ($last_state eq 'stopped') {
324 $self->next_state_stopped($sid, $cd, $sd, $lrm_res);
326 } elsif ($last_state eq 'started') {
328 $self->next_state_started($sid, $cd, $sd, $lrm_res);
330 } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') {
332 $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res);
334 } elsif ($last_state eq 'fence') {
336 # do nothing here - wait until fenced
338 } elsif ($last_state eq 'request_stop') {
340 $self->next_state_request_stop($sid, $cd, $sd, $lrm_res);
342 } elsif ($last_state eq 'freeze') {
344 my $lrm_mode = $sd->{node
} ?
$lrm_modes->{$sd->{node
}} : undef;
346 &$change_service_state($self, $sid, 'started')
347 if $lrm_mode && $lrm_mode eq 'active';
349 } elsif ($last_state eq 'error') {
355 die "unknown service state '$last_state'";
359 my $lrm_mode = $sd->{node
} ?
$lrm_modes->{$sd->{node
}} : undef;
360 $lrm_mode = 'unknown'if !$lrm_mode;
361 if (($sd->{state} eq 'started' || $sd->{state} eq 'stopped' ||
362 $sd->{state} eq 'request_stop') && ($lrm_mode ne 'active')) {
363 &$change_service_state($self, $sid, 'freeze');
366 $repeat = 1 if $sd->{state} ne $last_state;
370 my $fenced_nodes = {};
371 foreach my $sid (keys %$ss) {
372 my $sd = $ss->{$sid};
373 next if $sd->{state} ne 'fence';
375 if (!defined($fenced_nodes->{$sd->{node
}})) {
376 $fenced_nodes->{$sd->{node
}} = $ns->fence_node($sd->{node
}) || 0;
379 next if !$fenced_nodes->{$sd->{node
}};
381 # node fence was sucessful - mark service as stopped
382 &$change_service_state($self, $sid, 'stopped');
388 $self->flush_master_status();
391 # functions to compute next service states
392 # $cd: service configuration data (read only)
393 # $sd: service status data (read only)
395 # Note: use change_service_state() to alter state
398 sub next_state_request_stop
{
399 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
401 my $haenv = $self->{haenv
};
402 my $ns = $self->{ns
};
404 # check result from LRM daemon
406 my $exit_code = $lrm_res->{exit_code
};
407 if ($exit_code == 0) {
408 &$change_service_state($self, $sid, 'stopped');
411 &$change_service_state($self, $sid, 'error'); # fixme: what state?
416 if ($ns->node_is_offline_delayed($sd->{node
}, $fence_delay)) {
417 &$change_service_state($self, $sid, 'fence');
422 sub next_state_migrate_relocate
{
423 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
425 my $haenv = $self->{haenv
};
426 my $ns = $self->{ns
};
428 # check result from LRM daemon
430 my $exit_code = $lrm_res->{exit_code
};
431 if ($exit_code == 0) {
432 &$change_service_state($self, $sid, 'started', node
=> $sd->{target
});
435 $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)");
436 &$change_service_state($self, $sid, 'started', node
=> $sd->{node
});
441 if ($ns->node_is_offline_delayed($sd->{node
}, $fence_delay)) {
442 &$change_service_state($self, $sid, 'fence');
448 sub next_state_stopped
{
449 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
451 my $haenv = $self->{haenv
};
452 my $ns = $self->{ns
};
454 if ($sd->{node
} ne $cd->{node
}) {
455 # this can happen if we fence a node with active migrations
456 # hack: modify $sd (normally this should be considered read-only)
457 $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node}");
458 $sd->{node
} = $cd->{node
};
462 my ($cmd, $target) = @{$sd->{cmd
}};
465 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
466 if (!$ns->node_is_online($target)) {
467 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
468 } elsif ($sd->{node
} eq $target) {
469 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
472 $haenv->change_service_location($sid, $sd->{node
}, $target);
473 $cd->{node
} = $sd->{node
} = $target; # fixme: $sd is read-only??!!
474 $haenv->log('info', "$cmd service '$sid' to node '$target' (stopped)");
477 $haenv->log('err', "$cmd service '$sid' to node '$target' failed - $err");
481 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
485 if ($cd->{state} eq 'disabled') {
490 if ($cd->{state} eq 'enabled') {
491 if (my $node = select_service_node
($self->{groups
}, $self->{online_node_usage
}, $cd, $sd->{node
})) {
492 if ($node && ($sd->{node
} ne $node)) {
494 $haenv->change_service_location($sid, $sd->{node
}, $node);
495 $cd->{node
} = $sd->{node
} = $node; # fixme: $sd is read-only??!!
498 $haenv->log('err', "move service '$sid' to node '$node' failed - $err");
500 &$change_service_state($self, $sid, 'started', node
=> $node);
503 &$change_service_state($self, $sid, 'started', node
=> $node);
512 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
515 sub next_state_started
{
516 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
518 my $haenv = $self->{haenv
};
519 my $ns = $self->{ns
};
521 if (!$ns->node_is_online($sd->{node
})) {
522 if ($ns->node_is_offline_delayed($sd->{node
}, $fence_delay)) {
523 &$change_service_state($self, $sid, 'fence');
528 if ($cd->{state} eq 'disabled') {
529 &$change_service_state($self, $sid, 'request_stop');
533 if ($cd->{state} eq 'enabled') {
536 my ($cmd, $target) = @{$sd->{cmd
}};
539 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
540 if (!$ns->node_is_online($target)) {
541 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
542 } elsif ($sd->{node
} eq $target) {
543 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
545 $haenv->log('info', "$cmd service '$sid' to node '$target' (running)");
546 &$change_service_state($self, $sid, $cmd, node
=> $sd->{node
}, target
=> $target);
549 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
554 if ($lrm_res && ($lrm_res->{exit_code
} != 0)) { # fixme: other exit codes?
558 my $node = select_service_node
($self->{groups
}, $self->{online_node_usage
},
559 $cd, $sd->{node
}, $try_next);
561 if ($node && ($sd->{node
} ne $node)) {
562 $haenv->log('info', "migrate service '$sid' to node '$node' (running)");
563 &$change_service_state($self, $sid, 'migrate', node
=> $sd->{node
}, target
=> $node);
572 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");