use warnings;
use Digest::MD5 qw(md5_base64);
-use Data::Dumper;
use PVE::Tools;
use PVE::HA::Tools ':exit_codes';
use PVE::HA::NodeStatus;
my $class = ref($this) || $this;
- my $ms = $haenv->read_manager_status();
+ my $self = bless { haenv => $haenv }, $class;
- $ms->{master_node} = $haenv->nodename();
+ my $old_ms = $haenv->read_manager_status();
- my $ns = PVE::HA::NodeStatus->new($haenv, $ms->{node_status} || {});
+ # we only copy the state part of the manager which cannot be auto generated
+
+ $self->{ns} = PVE::HA::NodeStatus->new($haenv, $old_ms->{node_status} || {});
# fixme: use separate class PVE::HA::ServiceStatus
- my $ss = $ms->{service_status} || {};
+ $self->{ss} = $old_ms->{service_status} || {};
- my $self = bless {
- haenv => $haenv,
- ms => $ms, # master status
- ns => $ns, # PVE::HA::NodeStatus
- ss => $ss, # service status
- }, $class;
+ $self->{ms} = { master_node => $haenv->nodename() };
return $self;
}
$ms->{node_status} = $ns->{status};
$ms->{service_status} = $ss;
$ms->{timestamp} = $haenv->get_time();
-
+
$haenv->write_manager_status($ms);
-}
+}
-sub select_service_node {
- my ($groups, $online_node_usage, $service_conf, $current_node, $try_next) = @_;
+sub get_service_group {
+ my ($groups, $online_node_usage, $service_conf) = @_;
- my $group = { 'nodes' => { $service_conf->{node} => 1 } }; # default group
+ my $group = {};
+ # add all online nodes to default group to allow try_next when no group set
+ foreach my $node (keys %$online_node_usage) {
+ $group->{nodes}->{$node} = 1;
+ }
- $group = $groups->{ids}->{$service_conf->{group}} if $service_conf->{group} &&
+ # overwrite default if service is bound to a specific group
+ $group = $groups->{ids}->{$service_conf->{group}} if $service_conf->{group} &&
$groups->{ids}->{$service_conf->{group}};
+ return $group;
+}
+
+# groups available nodes with their priority as group index
+sub get_node_priority_groups {
+ my ($group, $online_node_usage) = @_;
+
my $pri_groups = {};
my $group_members = {};
foreach my $entry (keys %{$group->{nodes}}) {
$group_members->{$node} = $pri;
}
-
# add non-group members to unrestricted groups (priority -1)
if (!$group->{restricted}) {
my $pri = -1;
}
}
+ return ($pri_groups, $group_members);
+}
+
+sub select_service_node {
+ my ($groups, $online_node_usage, $service_conf, $current_node, $try_next, $tried_nodes, $maintenance_fallback) = @_;
+
+ my $group = get_service_group($groups, $online_node_usage, $service_conf);
+
+ my ($pri_groups, $group_members) = get_node_priority_groups($group, $online_node_usage);
my @pri_list = sort {$b <=> $a} keys %$pri_groups;
return undef if !scalar(@pri_list);
-
+
+ # stay on current node if possible (avoids random migrations)
if (!$try_next && $group->{nofailback} && defined($group_members->{$current_node})) {
return $current_node;
}
my $top_pri = $pri_list[0];
- my @nodes = sort {
+ # try to avoid nodes where the service failed already if we want to relocate
+ if ($try_next) {
+ foreach my $node (@$tried_nodes) {
+ delete $pri_groups->{$top_pri}->{$node};
+ }
+ }
+
+ my @nodes = sort {
$online_node_usage->{$a} <=> $online_node_usage->{$b} || $a cmp $b
} keys %{$pri_groups->{$top_pri}};
my $found;
+ my $found_maintenace_fallback;
for (my $i = scalar(@nodes) - 1; $i >= 0; $i--) {
my $node = $nodes[$i];
if ($node eq $current_node) {
$found = $i;
- last;
+ }
+ if (defined($maintenance_fallback) && $node eq $maintenance_fallback) {
+ $found_maintenace_fallback = $i;
}
}
- if ($try_next) {
+ if (defined($found_maintenace_fallback)) {
+ return $nodes[$found_maintenace_fallback];
+ }
+ if ($try_next) {
if (defined($found) && ($found < (scalar(@nodes) - 1))) {
return $nodes[$found + 1];
} else {
return $nodes[0];
}
-
+ } elsif (defined($found)) {
+ return $nodes[$found];
} else {
-
- return $nodes[$found] if defined($found);
-
return $nodes[0];
-
}
}
sub compute_new_uuid {
my ($state) = @_;
-
+
$uid_counter++;
return md5_base64($state . $$ . time() . $uid_counter);
}
my $sd = $self->{ss}->{$sid};
my $state = $sd->{state};
if (defined($online_node_usage->{$sd->{node}})) {
- if (($state eq 'started') || ($state eq 'request_stop') ||
+ if (($state eq 'started') || ($state eq 'request_stop') ||
($state eq 'fence') || ($state eq 'freeze') || ($state eq 'error')) {
$online_node_usage->{$sd->{node}}++;
} elsif (($state eq 'migrate') || ($state eq 'relocate')) {
+ # count it for both, source and target as load is put on both
+ $online_node_usage->{$sd->{node}}++;
$online_node_usage->{$sd->{target}}++;
} elsif ($state eq 'stopped') {
# do nothing
my $old_state = $sd->{state};
my $old_node = $sd->{node};
+ my $old_failed_nodes = $sd->{failed_nodes};
+ my $old_maintenance_node = $sd->{maintenance_node};
die "no state change" if $old_state eq $new_state; # just to be sure
$sd->{state} = $new_state;
$sd->{node} = $old_node;
+ $sd->{failed_nodes} = $old_failed_nodes if defined($old_failed_nodes);
+ $sd->{maintenance_node} = $old_maintenance_node if defined($old_maintenance_node);
my $text_state = '';
foreach my $k (sort keys %params) {
" to '${new_state}'$text_state");
};
+# clean up a possible bad state from a recovered service to allow its start
+my $fence_recovery_cleanup = sub {
+ my ($self, $sid, $fenced_node) = @_;
+
+ my $haenv = $self->{haenv};
+
+ my (undef, $type, $id) = $haenv->parse_sid($sid);
+ my $plugin = PVE::HA::Resources->lookup($type);
+
+ # should not happen
+ die "unknown resource type '$type'" if !$plugin;
+
+ # locks may block recovery, cleanup those which are safe to remove after fencing,
+ # i.e., after the original node was reset and thus all it's state
+ my $removable_locks = ['backup', 'mounted', 'migrate', 'clone', 'rollback', 'snapshot', 'snapshot-delete', 'suspending', 'suspended'];
+ if (my $removed_lock = $plugin->remove_locks($haenv, $id, $removable_locks, $fenced_node)) {
+ $haenv->log('warning', "removed leftover lock '$removed_lock' from recovered " .
+ "service '$sid' to allow its start.");
+ }
+};
+
# after a node was fenced this recovers the service to a new node
my $recover_fenced_service = sub {
my ($self, $sid, $cd) = @_;
$haenv->log('info', "recover service '$sid' from fenced node " .
"'$fenced_node' to node '$recovery_node'");
+ &$fence_recovery_cleanup($self, $sid, $fenced_node);
+
$haenv->steal_service($sid, $sd->{node}, $recovery_node);
+ $self->{online_node_usage}->{$recovery_node}++;
# $sd *is normally read-only*, fencing is the exception
$cd->{node} = $sd->{node} = $recovery_node;
- &$change_service_state($self, $sid, 'started', node => $recovery_node);
+ my $new_state = ($cd->{state} eq 'started') ? 'started' : 'request_stop';
+ &$change_service_state($self, $sid, $new_state, node => $recovery_node);
} else {
- # no node found, let the service in 'fence' state and try again
+ # no possible node found, cannot recover
$haenv->log('err', "recovering service '$sid' from fenced node " .
"'$fenced_node' failed, no recovery node found");
+ &$change_service_state($self, $sid, 'error');
}
};
-# read LRM status for all nodes
+# read LRM status for all nodes
sub read_lrm_status {
my ($self) = @_;
}
}
-
+
return ($results, $modes);
}
my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
my $cmdlist = $haenv->read_crm_commands();
-
+
foreach my $cmd (split(/\n/, $cmdlist)) {
chomp $cmd;
if ($cmd =~ m/^(migrate|relocate)\s+(\S+)\s+(\S+)$/) {
- my ($task, $sid, $node) = ($1, $2, $3);
+ my ($task, $sid, $node) = ($1, $2, $3);
if (my $sd = $ss->{$sid}) {
if (!$ns->node_is_online($node)) {
$haenv->log('err', "crm command error - node not online: $cmd");
} else {
if ($node eq $sd->{node}) {
$haenv->log('info', "ignore crm command - service already on target node: $cmd");
- } else {
+ } else {
$haenv->log('info', "got crm command: $cmd");
- $ss->{$sid}->{cmd} = [ $task, $node];
+ $ss->{$sid}->{cmd} = [ $task, $node ];
}
}
} else {
$haenv->log('err', "crm command error - no such service: $cmd");
}
+ } elsif ($cmd =~ m/^stop\s+(\S+)\s+(\S+)$/) {
+ my ($sid, $timeout) = ($1, $2);
+ if (my $sd = $ss->{$sid}) {
+ $haenv->log('info', "got crm command: $cmd");
+ $ss->{$sid}->{cmd} = [ 'stop', $timeout ];
+ } else {
+ $haenv->log('err', "crm command error - no such service: $cmd");
+ }
} else {
$haenv->log('err', "unable to parse crm command: $cmd");
}
my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
- $ns->update($haenv->get_node_info());
+ my ($node_info) = $haenv->get_node_info();
+ my ($lrm_results, $lrm_modes) = $self->read_lrm_status();
+
+ $ns->update($node_info, $lrm_modes);
- if (!$ns->node_is_online($haenv->nodename())) {
+ if (!$ns->node_is_operational($haenv->nodename())) {
$haenv->log('info', "master seems offline");
return;
}
- my ($lrm_results, $lrm_modes) = $self->read_lrm_status();
-
my $sc = $haenv->read_service_config();
$self->{groups} = $haenv->read_group_config(); # update
# add new service
foreach my $sid (sort keys %$sc) {
next if $ss->{$sid}; # already there
- $haenv->log('info', "adding new service '$sid' on node '$sc->{$sid}->{node}'");
+ my $cd = $sc->{$sid};
+ next if $cd->{state} eq 'ignored';
+
+ $haenv->log('info', "adding new service '$sid' on node '$cd->{node}'");
# assume we are running to avoid relocate running service at add
- $ss->{$sid} = { state => 'started', node => $sc->{$sid}->{node},
+ my $state = ($cd->{state} eq 'started') ? 'started' : 'request_stop';
+ $ss->{$sid} = { state => $state, node => $cd->{node},
uid => compute_new_uuid('started') };
}
- # remove stale service from manager state
+ # remove stale or ignored services from manager state
foreach my $sid (keys %$ss) {
- next if $sc->{$sid};
- $haenv->log('info', "removing stale service '$sid' (no config)");
- delete $ss->{$sid};
- }
+ next if $sc->{$sid} && $sc->{$sid}->{state} ne 'ignored';
- # remove stale relocation try entries
- foreach my $sid (keys %{$ms->{relocate_trial}}) {
- delete $ms->{relocate_trial}->{$sid} if !$ss->{$sid};
+ my $reason = defined($sc->{$sid}) ? 'ignored state requested' : 'no config';
+ $haenv->log('info', "removing stale service '$sid' ($reason)");
+
+ # remove all service related state information
+ delete $ss->{$sid};
}
$self->update_crm_commands();
for (;;) {
my $repeat = 0;
-
+
$self->recompute_online_node_usage();
- foreach my $sid (keys %$ss) {
+ foreach my $sid (sort keys %$ss) {
my $sd = $ss->{$sid};
my $cd = $sc->{$sid} || { state => 'disabled' };
my $lrm_mode = $sd->{node} ? $lrm_modes->{$sd->{node}} : undef;
# unfreeze
- &$change_service_state($self, $sid, 'started')
+ my $state = ($cd->{state} eq 'started') ? 'started' : 'request_stop';
+ &$change_service_state($self, $sid, $state)
if $lrm_mode && $lrm_mode eq 'active';
} elsif ($last_state eq 'error') {
# handle fencing
my $fenced_nodes = {};
- foreach my $sid (keys %$ss) {
+ foreach my $sid (sort keys %$ss) {
my $sd = $ss->{$sid};
next if $sd->{state} ne 'fence';
# check result from LRM daemon
if ($lrm_res) {
my $exit_code = $lrm_res->{exit_code};
- my $req_state = $cd->{state} eq 'enabled' ? 'started' : 'request_stop';
+ my $req_state = $cd->{state} eq 'started' ? 'started' : 'request_stop';
if ($exit_code == SUCCESS) {
&$change_service_state($self, $sid, $req_state, node => $sd->{target});
return;
}
}
-
sub next_state_stopped {
my ($self, $sid, $cd, $sd, $lrm_res) = @_;
# this can happen if we fence a node with active migrations
# hack: modify $sd (normally this should be considered read-only)
$haenv->log('info', "fixup service '$sid' location ($sd->{node} => $cd->{node})");
- $sd->{node} = $cd->{node};
+ $sd->{node} = $cd->{node};
}
if ($sd->{cmd}) {
- my ($cmd, $target) = @{$sd->{cmd}};
- delete $sd->{cmd};
+ my $cmd = shift @{$sd->{cmd}};
if ($cmd eq 'migrate' || $cmd eq 'relocate') {
+ my $target = shift @{$sd->{cmd}};
if (!$ns->node_is_online($target)) {
$haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
} elsif ($sd->{node} eq $target) {
target => $target);
return;
}
+ } elsif ($cmd eq 'stop') {
+ $haenv->log('info', "ignore service '$sid' $cmd request - service already stopped");
} else {
- $haenv->log('err', "unknown command '$cmd' for service '$sid'");
+ $haenv->log('err', "unknown command '$cmd' for service '$sid'");
}
+ delete $sd->{cmd};
}
if ($cd->{state} eq 'disabled') {
return;
}
- if ($cd->{state} eq 'enabled') {
+ if ($ns->node_is_offline_delayed($sd->{node})) {
+ &$change_service_state($self, $sid, 'fence');
+ return;
+ }
+
+ if ($cd->{state} eq 'stopped') {
+ # almost the same as 'disabled' state but the service will also get recovered
+ return;
+ }
+
+ if ($cd->{state} eq 'started') {
# simply mark it started, if it's on the wrong node
# next_state_started will fix that for us
&$change_service_state($self, $sid, 'started', node => $sd->{node});
$haenv->log('err', "service '$sid' - unknown state '$cd->{state}' in service configuration");
}
+sub record_service_failed_on_node {
+ my ($self, $sid, $node) = @_;
+
+ if (!defined($self->{ss}->{$sid}->{failed_nodes})) {
+ $self->{ss}->{$sid}->{failed_nodes} = [];
+ }
+
+ push @{$self->{ss}->{$sid}->{failed_nodes}}, $node;
+}
+
sub next_state_started {
my ($self, $sid, $cd, $sd, $lrm_res) = @_;
if ($ns->node_is_offline_delayed($sd->{node})) {
&$change_service_state($self, $sid, 'fence');
}
- return;
+ if ($ns->get_node_state($sd->{node}) ne 'maintenance') {
+ return;
+ } else {
+ # save current node as fallback for when it comes out of
+ # maintenance
+ $sd->{maintenance_node} = $sd->{node};
+ }
}
-
- if ($cd->{state} eq 'disabled') {
+
+ if ($cd->{state} eq 'disabled' || $cd->{state} eq 'stopped') {
&$change_service_state($self, $sid, 'request_stop');
return;
}
- if ($cd->{state} eq 'enabled') {
+ if ($cd->{state} eq 'started') {
if ($sd->{cmd}) {
- my ($cmd, $target) = @{$sd->{cmd}};
- delete $sd->{cmd};
+ my $cmd = shift @{$sd->{cmd}};
if ($cmd eq 'migrate' || $cmd eq 'relocate') {
+ my $target = shift @{$sd->{cmd}};
if (!$ns->node_is_online($target)) {
$haenv->log('err', "ignore service '$sid' $cmd request - node '$target' not online");
} elsif ($sd->{node} eq $target) {
$haenv->log('info', "$cmd service '$sid' to node '$target'");
&$change_service_state($self, $sid, $cmd, node => $sd->{node}, target => $target);
}
+ } elsif ($cmd eq 'stop') {
+ my $timeout = shift @{$sd->{cmd}};
+ if ($timeout == 0) {
+ $haenv->log('info', "request immediate service hard-stop for service '$sid'");
+ } else {
+ $haenv->log('info', "request graceful stop with timeout '$timeout' for service '$sid'");
+ }
+ &$change_service_state($self, $sid, 'request_stop', timeout => $timeout);
+ $haenv->update_service_config($sid, {'state' => 'stopped'});
} else {
- $haenv->log('err', "unknown command '$cmd' for service '$sid'");
+ $haenv->log('err', "unknown command '$cmd' for service '$sid'");
}
+
+ delete $sd->{cmd};
+
} else {
my $try_next = 0;
+
if ($lrm_res) {
+
my $ec = $lrm_res->{exit_code};
if ($ec == SUCCESS) {
- $master_status->{relocate_trial}->{$sid} = 0;
+ if (defined($sd->{failed_nodes})) {
+ $haenv->log('info', "relocation policy successful for '$sid' on node '$sd->{node}'," .
+ " failed nodes: " . join(', ', @{$sd->{failed_nodes}}) );
+ }
+
+ delete $sd->{failed_nodes};
+
+ # store flag to indicate successful start - only valid while state == 'started'
+ $sd->{running} = 1;
} elsif ($ec == ERROR) {
- # apply our relocate policy if we got ERROR from the LRM
- my $try = $master_status->{relocate_trial}->{$sid} || 0;
+ delete $sd->{running};
- if ($try < $cd->{max_relocate}) {
+ # apply our relocate policy if we got ERROR from the LRM
+ $self->record_service_failed_on_node($sid, $sd->{node});
+
+ if (scalar(@{$sd->{failed_nodes}}) <= $cd->{max_relocate}) {
- $try++;
# tell select_service_node to relocate if possible
$try_next = 1;
$haenv->log('warning', "starting service $sid on node".
" '$sd->{node}' failed, relocating service.");
- $master_status->{relocate_trial}->{$sid} = $try;
} else {
- $haenv->log('err', "recovery policy for service".
- " $sid failed, entering error state!");
+ $haenv->log('err', "recovery policy for service $sid " .
+ "failed, entering error state. Failed nodes: ".
+ join(', ', @{$sd->{failed_nodes}}));
&$change_service_state($self, $sid, 'error');
return;
}
} else {
+ $self->record_service_failed_on_node($sid, $sd->{node});
+
$haenv->log('err', "service '$sid' got unrecoverable error" .
" (exit code $ec))");
# we have no save way out (yet) for other errors
}
}
- my $node = select_service_node($self->{groups}, $self->{online_node_usage},
- $cd, $sd->{node}, $try_next);
+ my $node = select_service_node(
+ $self->{groups},
+ $self->{online_node_usage},
+ $cd,
+ $sd->{node},
+ $try_next,
+ $sd->{failed_nodes},
+ $sd->{maintenance_node},
+ );
if ($node && ($sd->{node} ne $node)) {
+ $self->{online_node_usage}->{$node}++;
+
+ if (defined(my $fallback = $sd->{maintenance_node})) {
+ if ($node eq $fallback) {
+ $haenv->log('info', "moving service '$sid' back to '$fallback', node came back from maintenance.");
+ delete $sd->{maintenance_node};
+ } elsif ($sd->{node} ne $fallback) {
+ $haenv->log('info', "dropping maintenance fallback node '$fallback' for '$sid'");
+ delete $sd->{maintenance_node};
+ }
+ }
+
if ($cd->{type} eq 'vm') {
$haenv->log('info', "migrate service '$sid' to node '$node' (running)");
&$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node);
&$change_service_state($self, $sid, 'relocate', node => $sd->{node}, target => $node);
}
} else {
+ if ($try_next && !defined($node)) {
+ $haenv->log('warning', "Start Error Recovery: Tried all available " .
+ " nodes for service '$sid', retry start on current node. " .
+ "Tried nodes: " . join(', ', @{$sd->{failed_nodes}}));
+ }
# ensure service get started again if it went unexpected down
- $sd->{uid} = compute_new_uuid($sd->{state});
+ # but ensure also no LRM result gets lost
+ $sd->{uid} = compute_new_uuid($sd->{state}) if defined($lrm_res);
}
}
my ($self, $sid, $cd, $sd, $lrm_res) = @_;
my $ns = $self->{ns};
+ my $ms = $self->{ms};
if ($cd->{state} eq 'disabled') {
- &$change_service_state($self, $sid, 'stopped');
- return;
- }
+ # clean up on error recovery
+ delete $sd->{failed_nodes};
- if ($ns->node_is_offline_delayed($sd->{node})) {
- &$change_service_state($self, $sid, 'fence');
+ &$change_service_state($self, $sid, 'stopped');
return;
}
projects / pve-ha-manager.git / blobdiff