@ -100,6 +100,8 @@ find_changes($anvil);
check_config($anvil);
fix_things($anvil);
# Shut down.
$anvil->ScanCore->agent_shutdown({agent => $THIS_FILE});
@ -108,6 +110,133 @@ $anvil->ScanCore->agent_shutdown({agent => $THIS_FILE});
# Functions #
#############################################################################################################
# This looks for issues that we can fix, and if any are found, we try to fix them.
sub fix_things
{
my ($anvil) = @_;
# If the cluster is up and both nodes are online, make sure all DRBD resources are connected.
$anvil->Database->get_hosts();
$anvil->DRBD->get_status();
my $host_uuid = $anvil->Get->host_uuid();
my $host_type = $anvil->Get->host_type();
my $short_host_name = $anvil->Get->short_host_name();
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
host_uuid => $host_uuid,
host_type => $host_type,
short_host_name => $short_host_name,
}});
if ($host_type eq "node")
{
my $problem = $anvil->Cluster->parse_cib({debug => 3});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
problem => $problem,
"cib::parsed::local::ready" => $anvil->data->{cib}{parsed}{'local'}{ready},
"cib::parsed::peer::ready" => $anvil->data->{cib}{parsed}{peer}{ready},
}});
if ((not $problem) && ($anvil->data->{cib}{parsed}{peer}{ready}) && ($anvil->data->{cib}{parsed}{'local'}{ready}))
{
# Walk through all resources and make sure the peer and local disks are attached and
# the connection established.
foreach my $resource (sort {$a cmp $b} keys %{$anvil->data->{new}{resource}})
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { resource => $resource }});
# Check connection states.
foreach my $local_name (sort {$a cmp $b} keys%{$anvil->data->{drbd}{status}})
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { local_name => $local_name }});
foreach my $peer (sort {$a cmp $b} keys %{$anvil->data->{drbd}{status}{$local_name}{resource}{$resource}{connection}})
{
my $connection_state = lc($anvil->data->{drbd}{status}{$local_name}{resource}{$resource}{connection}{$peer}{'connection-state'});
my $peer_host_uuid = $anvil->Get->host_uuid_from_name({host_name => $peer, debug => 3});
my $peer_host_type = exists $anvil->data->{hosts}{host_uuid}{$host_uuid} ? $anvil->data->{hosts}{host_uuid}{$host_uuid}{host_type} : "";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:peer' => $peer,
's2:connection_state' => $connection_state,
's3:peer_host_uuid' => $peer_host_uuid,
's4:peer_host_type' => $peer_host_type,
}});
if (($peer_host_type eq "node") && ($connection_state eq "standalone"))
{
# Are either of the nodes still 'allow-two-primary',
# set it to 'no' (no migration is happening anyway,
# if we're StandAlone), and try reconnecting.
my $variables = {
resource => $resource,
peer => $peer,
};
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 2, key => "scan_drbd_message_0037", variables => $variables});
$anvil->Alert->register({alert_level => "notice", message => "scan_drbd_message_0037", variables => $variables, set_by => $THIS_FILE, sort_position => $anvil->data->{'scan-drbd'}{alert_sort}++});
$anvil->DRBD->manage_resource({
debug => 2,
resource => $resource,
task => "up",
});
}
}
}
foreach my $volume (sort {$a <=> $b} keys %{$anvil->data->{new}{resource}{$resource}{volume}})
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { volume => $volume }});
foreach my $peer (sort {$a cmp $b} keys %{$anvil->data->{new}{resource}{$resource}{volume}{$volume}{peer}})
{
my $peer_host_uuid = $anvil->Get->host_uuid_from_name({host_name => $peer});
my $peer_host_type = exists $anvil->data->{hosts}{host_uuid}{$host_uuid} ? $anvil->data->{hosts}{host_uuid}{$host_uuid}{host_type} : "";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:peer' => $peer,
's2:peer_host_uuid' => $peer_host_uuid,
's3:peer_host_type' => $peer_host_type,
}});
if ($peer_host_type eq "node")
{
my $connection_state = $anvil->data->{new}{resource}{$resource}{volume}{$volume}{peer}{$peer}{connection_state};
my $local_role = $anvil->data->{new}{resource}{$resource}{volume}{$volume}{peer}{$peer}{local_role};
my $peer_role = $anvil->data->{new}{resource}{$resource}{volume}{$volume}{peer}{$peer}{peer_role};
my $local_disk_state = $anvil->data->{new}{resource}{$resource}{volume}{$volume}{peer}{$peer}{local_disk_state};
my $peer_disk_state = $anvil->data->{new}{resource}{$resource}{volume}{$volume}{peer}{$peer}{peer_disk_state};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:connection_state' => $connection_state,
's2:local_role' => $local_role,
's3:peer_role' => $peer_role,
's4:local_disk_state' => $local_disk_state,
's5:peer_disk_state' => $peer_disk_state,
}});
# Find the peer's host_uuid and make sure we're
# talking to a node, not a DR host.
if ($local_disk_state eq "down")
{
# Bring it up.
my $variables = {
resource => $resource,
volume => $volume,
peer => $peer,
};
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 2, key => "scan_drbd_message_0036", variables => $variables});
$anvil->Alert->register({alert_level => "notice", message => "scan_drbd_message_0036", variables => $variables, set_by => $THIS_FILE, sort_position => $anvil->data->{'scan-drbd'}{alert_sort}++});
$anvil->DRBD->manage_resource({
debug => 2,
resource => $resource,
task => "up",
});
}
}
}
}
}
}
}
return(0);
}
# In the early days, scan_drbd_peers -> scan_drbd_peer_tcp_port was type numeric. This wasn't compatible with
# drbd-proxy and had to be changed to type text to support csv port lists.
sub check_schema
digimer
2 years ago