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 = {
146 sub recompute_online_node_usage
{
149 my $online_node_usage = {};
151 my $online_nodes = $self->{ns
}->list_online_nodes();
153 foreach my $node (@$online_nodes) {
154 $online_node_usage->{$node} = 0;
157 foreach my $sid (keys %{$self->{ss
}}) {
158 my $sd = $self->{ss
}->{$sid};
159 my $state = $sd->{state};
160 if (defined($online_node_usage->{$sd->{node
}})) {
161 if (($state eq 'started') || ($state eq 'request_stop') ||
162 ($state eq 'fence') || ($state eq 'error')) {
163 $online_node_usage->{$sd->{node
}}++;
164 } elsif (($state eq 'migrate') || ($state eq 'relocate')) {
165 $online_node_usage->{$sd->{target
}}++;
166 } elsif ($state eq 'stopped') {
169 die "should not be reached";
174 $self->{online_node_usage
} = $online_node_usage;
177 my $change_service_state = sub {
178 my ($self, $sid, $new_state, %params) = @_;
180 my ($haenv, $ss) = ($self->{haenv
}, $self->{ss
});
182 my $sd = $ss->{$sid} || die "no such service '$sid";
184 my $old_state = $sd->{state};
185 my $old_node = $sd->{node
};
187 die "no state change" if $old_state eq $new_state; # just to be sure
189 die "invalid CRM service state '$new_state'\n" if !$valid_service_states->{$new_state};
191 foreach my $k (keys %$sd) { delete $sd->{$k}; };
193 $sd->{state} = $new_state;
194 $sd->{node
} = $old_node;
197 foreach my $k (keys %params) {
199 $text_state .= ", " if $text_state;
200 $text_state .= "$k = $v";
204 $self->recompute_online_node_usage();
206 $sd->{uid
} = compute_new_uuid
($new_state);
209 $text_state = " ($text_state)" if $text_state;
210 $haenv->log('info', "service '$sid': state changed from '${old_state}' to '${new_state}' $text_state");
213 # read LRM status for all active nodes
214 sub read_lrm_status
{
217 my $nodes = $self->{ns
}->list_online_nodes();
218 my $haenv = $self->{haenv
};
222 foreach my $node (@$nodes) {
223 my $ls = $haenv->read_lrm_status($node);
224 foreach my $uid (keys %$ls) {
225 next if $res->{$uid}; # should not happen
226 $res->{$uid} = $ls->{$uid};
233 # read new crm commands and save them into crm master status
234 sub update_crm_commands
{
237 my ($haenv, $ms, $ns, $ss) = ($self->{haenv
}, $self->{ms
}, $self->{ns
}, $self->{ss
});
239 my $cmdlist = $haenv->read_crm_commands();
241 foreach my $cmd (split(/\n/, $cmdlist)) {
244 if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) {
245 my ($task, $sid, $node) = ($1, $2, $3);
246 if (my $sd = $ss->{$sid}) {
247 if (!$ns->node_is_online($node)) {
248 $haenv->log('err', "crm command error - node not online: $cmd");
250 if ($node eq $sd->{node
}) {
251 $haenv->log('info', "ignore crm command - service already on target node: $cmd");
253 $haenv->log('info', "got crm command: $cmd");
254 $ss->{$sid}->{cmd
} = [ $task, $node];
258 $haenv->log('err', "crm command error - no such service: $cmd");
262 $haenv->log('err', "unable to parse crm command: $cmd");
271 my ($haenv, $ms, $ns, $ss) = ($self->{haenv
}, $self->{ms
}, $self->{ns
}, $self->{ss
});
273 $ns->update($haenv->get_node_info());
275 if (!$ns->node_is_online($haenv->nodename())) {
276 $haenv->log('info', "master seems offline");
280 my $lrm_status = $self->read_lrm_status();
282 my $sc = $haenv->read_service_config();
284 $self->{groups
} = $haenv->read_group_config(); # update
286 # compute new service status
289 foreach my $sid (sort keys %$sc) {
290 next if $ss->{$sid}; # already there
291 $haenv->log('info', "adding new service '$sid' on node '$sc->{$sid}->{node}'");
292 # assume we are running to avoid relocate running service at add
293 $ss->{$sid} = { state => 'started', node
=> $sc->{$sid}->{node
},
294 uid
=> compute_new_uuid
('started') };
297 # remove stale service from manager state
298 foreach my $sid (keys %$ss) {
300 $haenv->log('info', "removing stale service '$sid' (no config)");
304 $self->update_crm_commands();
309 $self->recompute_online_node_usage();
311 foreach my $sid (keys %$ss) {
312 my $sd = $ss->{$sid};
313 my $cd = $sc->{$sid} || { state => 'disabled' };
315 my $lrm_res = $sd->{uid
} ?
$lrm_status->{$sd->{uid
}} : undef;
317 my $last_state = $sd->{state};
319 if ($last_state eq 'stopped') {
321 $self->next_state_stopped($sid, $cd, $sd, $lrm_res);
323 } elsif ($last_state eq 'started') {
325 $self->next_state_started($sid, $cd, $sd, $lrm_res);
327 } elsif ($last_state eq 'migrate' || $last_state eq 'relocate') {
329 $self->next_state_migrate_relocate($sid, $cd, $sd, $lrm_res);
331 } elsif ($last_state eq 'fence') {
333 # do nothing here - wait until fenced
335 } elsif ($last_state eq 'request_stop') {
337 $self->next_state_request_stop($sid, $cd, $sd, $lrm_res);
339 } elsif ($last_state eq 'error') {
345 die "unknown service state '$last_state'";
348 $repeat = 1 if $sd->{state} ne $last_state;
352 my $fenced_nodes = {};
353 foreach my $sid (keys %$ss) {
354 my $sd = $ss->{$sid};
355 next if $sd->{state} ne 'fence';
357 if (!defined($fenced_nodes->{$sd->{node
}})) {
358 $fenced_nodes->{$sd->{node
}} = $ns->fence_node($sd->{node
}) || 0;
361 next if !$fenced_nodes->{$sd->{node
}};
363 # node fence was sucessful - mark service as stopped
364 &$change_service_state($self, $sid, 'stopped');
370 $self->flush_master_status();
373 # functions to compute next service states
374 # $cd: service configuration data (read only)
375 # $sd: service status data (read only)
377 # Note: use change_service_state() to alter state
380 sub next_state_request_stop
{
381 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
383 my $haenv = $self->{haenv
};
384 my $ns = $self->{ns
};
386 # check result from LRM daemon
388 my $exit_code = $lrm_res->{exit_code
};
389 if ($exit_code == 0) {
390 &$change_service_state($self, $sid, 'stopped');
393 &$change_service_state($self, $sid, 'error'); # fixme: what state?
398 if ($ns->node_is_offline_delayed($sd->{node
}, $fence_delay)) {
399 &$change_service_state($self, $sid, 'fence');
404 sub next_state_migrate_relocate
{
405 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
407 my $haenv = $self->{haenv
};
408 my $ns = $self->{ns
};
410 # check result from LRM daemon
412 my $exit_code = $lrm_res->{exit_code
};
413 if ($exit_code == 0) {
414 &$change_service_state($self, $sid, 'started', node
=> $sd->{target
});
417 $haenv->log('err', "service '$sid' - migration failed (exit code $exit_code)");
418 &$change_service_state($self, $sid, 'started', node
=> $sd->{node
});
423 if ($ns->node_is_offline_delayed($sd->{node
}, $fence_delay)) {
424 &$change_service_state($self, $sid, 'fence');
430 sub next_state_stopped
{
431 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
433 my $haenv = $self->{haenv
};
434 my $ns = $self->{ns
};
436 if ($sd->{node
} ne $cd->{node
}) {
437 # this can happen if we fence a node with active migrations
438 # hack: modify $sd (normally this should be considered read-only)
439 $haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node}");
440 $sd->{node
} = $cd->{node
};
444 my ($cmd, $target) = @{$sd->{cmd
}};
447 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
448 if (!$ns->node_is_online($target)) {
449 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
450 } elsif ($sd->{node
} eq $target) {
451 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
453 $haenv->change_service_location($sid, $target);
454 $cd->{node
} = $sd->{node
} = $target; # fixme: $sd is read-only??!!
455 $haenv->log('info', "$cmd service '$sid' to node '$target' (stopped)");
458 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
462 if ($cd->{state} eq 'disabled') {
467 if ($cd->{state} eq 'enabled') {
468 if (my $node = select_service_node
($self->{groups
}, $self->{online_node_usage
}, $cd, $sd->{node
})) {
469 if ($node && ($sd->{node
} ne $node)) {
470 $haenv->change_service_location($sid, $node);
472 &$change_service_state($self, $sid, 'started', node
=> $node);
480 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
483 sub next_state_started
{
484 my ($self, $sid, $cd, $sd, $lrm_res) = @_;
486 my $haenv = $self->{haenv
};
487 my $ns = $self->{ns
};
489 if (!$ns->node_is_online($sd->{node
})) {
490 if ($ns->node_is_offline_delayed($sd->{node
}, $fence_delay)) {
491 &$change_service_state($self, $sid, 'fence');
496 if ($cd->{state} eq 'disabled') {
497 &$change_service_state($self, $sid, 'request_stop');
501 if ($cd->{state} eq 'enabled') {
504 my ($cmd, $target) = @{$sd->{cmd
}};
507 if ($cmd eq 'migrate' || $cmd eq 'relocate') {
508 if (!$ns->node_is_online($target)) {
509 $haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
510 } elsif ($sd->{node
} eq $target) {
511 $haenv->log('info', "ignore service '$sid' $cmd request - service already on node '$target'");
513 $haenv->log('info', "$cmd service '$sid' to node '$target' (running)");
514 &$change_service_state($self, $sid, $cmd, node
=> $sd->{node
}, target
=> $target);
517 $haenv->log('err', "unknown command '$cmd' for service '$sid'");
522 if ($lrm_res && ($lrm_res->{exit_code
} != 0)) { # fixme: other exit codes?
526 my $node = select_service_node
($self->{groups
}, $self->{online_node_usage
},
527 $cd, $sd->{node
}, $try_next);
529 if ($node && ($sd->{node
} ne $node)) {
530 $haenv->log('info', "migrate service '$sid' to node '$node' (running)");
531 &$change_service_state($self, $sid, 'migrate', node
=> $sd->{node
}, target
=> $node);
540 $haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");