@ -8,6 +8,10 @@
# WARNING: This is a pretty purpose-specific resource agent. No effort was made to test this on an rgmanager
# cluster or on any configuration outside how the Anvil! m3 uses it. If you plan to adapt it to
# another purpose, let us know and we'll try to help.
#
# NOTE: This method, for the sake of speed and reliability, does not connect to the Anvil! database. If you
# do work on this RA, be sure that a check is made for database connections before SQL calls are made
# in module methods.
#
# Based on: https://github.com/ClusterLabs/resource-agents/blob/master/doc/dev-guides/ra-dev-guide.asc
#
@ -143,13 +147,23 @@ $anvil->data->{environment}{OCF_RESKEY_CRM_meta_stop_drbd_resources} = 0;
# We're used by anvil-boot-server and anvil-stop-server. They don't set environment variables, but instead
# use switches. Pick those up, if passed.
$anvil->data->{switches}{migrate_to} = ""; # Sets 'meta_migrate_target'
$anvil->data->{switches}{migrate_from} = ""; # Sets 'meta_migrate_source' When set without 'migrate_to', does a status check after migration
$anvil->data->{switches}{server} = ""; # Sets 'name'.
$anvil->data->{switches}{start} = "";
$anvil->data->{switches}{stop} = "";
$anvil->data->{switches}{monitor} = "";
$anvil->data->{switches}{migrate_to} = ""; # Sets 'meta_migrate_target'
$anvil->data->{switches}{'migrate-to'} = "";
$anvil->data->{switches}{migrate_from} = ""; # Sets 'meta_migrate_source' When set without 'migrate_to', does a status check after migration
$anvil->data->{switches}{'migrate-from'} = "";
$anvil->data->{switches}{server} = ""; # Sets 'name'.
$anvil->data->{switches}{start} = "";
$anvil->data->{switches}{stop} = "";
$anvil->data->{switches}{monitor} = "";
$anvil->Get->switches();
if (($anvil->data->{switches}{'migrate-to'}) && not ($anvil->data->{switches}{migrate_to}))
{
$anvil->data->{switches}{migrate_to} = $anvil->data->{switches}{'migrate-to'};
}
if (($anvil->data->{switches}{'migrate-from'}) && not ($anvil->data->{switches}{migrate_from}))
{
$anvil->data->{switches}{migrate_from} = $anvil->data->{switches}{'migrate-from'};
}
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
"switches::migrate_to" => $anvil->data->{switches}{migrate_to},
"switches::migrate_from" => $anvil->data->{switches}{migrate_from},
@ -159,16 +173,6 @@ $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list
"switches::monitor" => $anvil->data->{switches}{monitor},
}});
# If we can connect to a database, we'll set/clear the 'migrating' flag during migrations. For timing reasons
# we don't let the RA do resyncs.
$anvil->Database->connect({sensitive => 1});
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 3, secure => 0, key => "log_0132"});
if (not $anvil->data->{sys}{database}{connections})
{
# No databases,
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, secure => 0, key => "warning_0073"});
}
if ($anvil->data->{switches}{stop_drbd_resources})
{
$anvil->data->{environment}{OCF_RESKEY_CRM_meta_migrate_source} = 1;
@ -177,190 +181,43 @@ if ($anvil->data->{switches}{stop_drbd_resources})
# Something for the logs
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 0, level => 2, key => "log_0298"});
=cut
Start:
environment::OCF_RESKEY_CRM_meta_migrate_source: []
environment::OCF_RESKEY_CRM_meta_migrate_target: []
environment::OCF_RESKEY_CRM_meta_name: [start]
environment::OCF_RESKEY_CRM_meta_on_fail: [block]
environment::OCF_RESKEY_CRM_meta_on_node: [an-a02n01]
environment::OCF_RESKEY_CRM_meta_on_node_uuid: [1]
environment::OCF_RESKEY_CRM_meta_record_pending: []
environment::OCF_RESKEY_CRM_meta_stop_drbd_resources: [0]
environment::OCF_RESKEY_CRM_meta_timeout: [300000]
environment::OCF_RESKEY_name: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_INSTANCE: [srv02-c8s-fujitsu]
environment::OCF_ROOT: [/usr/lib/ocf]
Monitor after start:
environment::OCF_RESKEY_CRM_meta_interval: [60000]
environment::OCF_RESKEY_CRM_meta_migrate_source: []
environment::OCF_RESKEY_CRM_meta_migrate_target: []
environment::OCF_RESKEY_CRM_meta_name: [monitor]
environment::OCF_RESKEY_CRM_meta_on_node: [an-a02n01]
environment::OCF_RESKEY_CRM_meta_on_node_uuid: [1]
environment::OCF_RESKEY_CRM_meta_record_pending: []
environment::OCF_RESKEY_CRM_meta_stop_drbd_resources: [0]
environment::OCF_RESKEY_CRM_meta_timeout: [20000]
environment::OCF_RESKEY_name: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_INSTANCE: [srv02-c8s-fujitsu]
environment::OCF_ROOT: [/usr/lib/ocf]
Monitor one minute later:
environment::OCF_RESKEY_CRM_meta_interval: [60000]
environment::OCF_RESKEY_CRM_meta_migrate_source: []
environment::OCF_RESKEY_CRM_meta_migrate_target: []
environment::OCF_RESKEY_CRM_meta_name: [monitor]
environment::OCF_RESKEY_CRM_meta_on_node: [an-a02n01]
environment::OCF_RESKEY_CRM_meta_on_node_uuid: [1]
environment::OCF_RESKEY_CRM_meta_record_pending: []
environment::OCF_RESKEY_CRM_meta_stop_drbd_resources: [0]
environment::OCF_RESKEY_CRM_meta_timeout: [20000]
environment::OCF_RESKEY_name: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_INSTANCE: [srv02-c8s-fujitsu]
environment::OCF_ROOT: [/usr/lib/ocf]
=cut Manual calls;
Migrate from an-a02n01 to an-a02n02
# Start a server;
/usr/lib/ocf/resource.d/alteeve/server -vv --log-secure --server <server_name> --start
environment::OCF_RESKEY_CRM_meta_interval: [60000]
environment::OCF_RESKEY_CRM_meta_migrate_source: []
environment::OCF_RESKEY_CRM_meta_migrate_target: []
environment::OCF_RESKEY_CRM_meta_name: [monitor]
environment::OCF_RESKEY_CRM_meta_on_node: [an-a02n01]
environment::OCF_RESKEY_CRM_meta_on_node_uuid: [1]
environment::OCF_RESKEY_CRM_meta_record_pending: []
environment::OCF_RESKEY_CRM_meta_stop_drbd_resources: [0]
environment::OCF_RESKEY_CRM_meta_timeout: [20000]
environment::OCF_RESKEY_name: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_INSTANCE: [srv02-c8s-fujitsu]
environment::OCF_ROOT: [/usr/lib/ocf]
<2 seconds later>
environment::OCF_RESKEY_CRM_meta_migrate_source: [an-a02n01]
environment::OCF_RESKEY_CRM_meta_migrate_target: [an-a02n02]
environment::OCF_RESKEY_CRM_meta_name: [migrate_to]
environment::OCF_RESKEY_CRM_meta_on_node: [an-a02n01]
environment::OCF_RESKEY_CRM_meta_on_node_uuid: [1]
environment::OCF_RESKEY_CRM_meta_record_pending: [true]
environment::OCF_RESKEY_CRM_meta_stop_drbd_resources: [0]
environment::OCF_RESKEY_CRM_meta_timeout: [86400000]
environment::OCF_RESKEY_name: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_INSTANCE: [srv02-c8s-fujitsu]
environment::OCF_ROOT: [/usr/lib/ocf]
# Stop a server
/usr/lib/ocf/resource.d/alteeve/server -vv --log-secure --server <server_name> --stop
Post migration on an-a02n01
# Monitor a server
/usr/lib/ocf/resource.d/alteeve/server -vv --log-secure --server <server_name> --monitor
environment::OCF_RESKEY_CRM_meta_migrate_source: []
environment::OCF_RESKEY_CRM_meta_migrate_target: []
environment::OCF_RESKEY_CRM_meta_name: [stop]
environment::OCF_RESKEY_CRM_meta_on_fail: [block]
environment::OCF_RESKEY_CRM_meta_on_node: [an-a02n01]
environment::OCF_RESKEY_CRM_meta_on_node_uuid: [1]
environment::OCF_RESKEY_CRM_meta_record_pending: []
environment::OCF_RESKEY_CRM_meta_stop_drbd_resources: [0]
environment::OCF_RESKEY_CRM_meta_timeout: [86400000]
environment::OCF_RESKEY_name: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_INSTANCE: [srv02-c8s-fujitsu]
environment::PCMK_debug: [0]
Post migration on an-a02n02
environment::OCF_RESKEY_CRM_meta_migrate_source: [an-a02n01]
environment::OCF_RESKEY_CRM_meta_migrate_target: [an-a02n02]
environment::OCF_RESKEY_CRM_meta_name: [migrate_from]
environment::OCF_RESKEY_CRM_meta_on_node: [an-a02n02]
environment::OCF_RESKEY_CRM_meta_on_node_uuid: [2]
environment::OCF_RESKEY_CRM_meta_record_pending: []
environment::OCF_RESKEY_CRM_meta_stop_drbd_resources: [0]
environment::OCF_RESKEY_CRM_meta_timeout: [600000]
environment::OCF_RESKEY_name: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_INSTANCE: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_PROVIDER: [alteeve]
environment::OCF_RESOURCE_TYPE: [server]
environment::OCF_ROOT: [/usr/lib/ocf]
environment::PCMK_debug: [0]
Checking server state after: [srv02-c8s-fujitsu] was migrated to this host.
environment::OCF_EXIT_REASON_PREFIX: [ocf-exit-reason:]
environment::OCF_RA_VERSION_MAJOR: [1]
environment::OCF_RA_VERSION_MINOR: [0]
environment::OCF_RESKEY_CRM_meta_interval: [60000]
environment::OCF_RESKEY_CRM_meta_migrate_source: []
environment::OCF_RESKEY_CRM_meta_migrate_target: []
environment::OCF_RESKEY_CRM_meta_name: [monitor]
environment::OCF_RESKEY_CRM_meta_on_node: [an-a02n02]
environment::OCF_RESKEY_CRM_meta_on_node_uuid: [2]
environment::OCF_RESKEY_CRM_meta_record_pending: []
environment::OCF_RESKEY_CRM_meta_stop_drbd_resources: [0]
environment::OCF_RESKEY_CRM_meta_timeout: [20000]
environment::OCF_RESKEY_crm_feature_set: [3.7.1]
environment::OCF_RESKEY_name: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_INSTANCE: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_PROVIDER: [alteeve]
environment::OCF_RESOURCE_TYPE: [server]
environment::OCF_ROOT: [/usr/lib/ocf]
environment::PCMK_debug: [0]
Monitor on an-a02n02 after a minute
environment::OCF_EXIT_REASON_PREFIX: [ocf-exit-reason:]
environment::OCF_RA_VERSION_MAJOR: [1]
environment::OCF_RA_VERSION_MINOR: [0]
environment::OCF_RESKEY_CRM_meta_interval: [60000]
environment::OCF_RESKEY_CRM_meta_migrate_source: []
environment::OCF_RESKEY_CRM_meta_migrate_target: []
environment::OCF_RESKEY_CRM_meta_name: [monitor]
environment::OCF_RESKEY_CRM_meta_on_node: [an-a02n02]
environment::OCF_RESKEY_CRM_meta_on_node_uuid: [2]
environment::OCF_RESKEY_CRM_meta_record_pending: []
environment::OCF_RESKEY_CRM_meta_stop_drbd_resources: [0]
environment::OCF_RESKEY_CRM_meta_timeout: [20000]
environment::OCF_RESKEY_crm_feature_set: [3.7.1]
environment::OCF_RESKEY_name: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_INSTANCE: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_PROVIDER: [alteeve]
environment::OCF_RESOURCE_TYPE: [server]
environment::OCF_ROOT: [/usr/lib/ocf]
environment::PCMK_debug: [0]
Stop server:
environment::OCF_EXIT_REASON_PREFIX: [ocf-exit-reason:]
environment::OCF_RA_VERSION_MAJOR: [1]
environment::OCF_RA_VERSION_MINOR: [0]
environment::OCF_RESKEY_CRM_meta_migrate_source: []
environment::OCF_RESKEY_CRM_meta_migrate_target: []
environment::OCF_RESKEY_CRM_meta_name: [stop]
environment::OCF_RESKEY_CRM_meta_on_fail: [block]
environment::OCF_RESKEY_CRM_meta_on_node: [an-a02n02]
environment::OCF_RESKEY_CRM_meta_on_node_uuid: [2]
environment::OCF_RESKEY_CRM_meta_record_pending: []
environment::OCF_RESKEY_CRM_meta_stop_drbd_resources: [0]
environment::OCF_RESKEY_CRM_meta_timeout: [86400000]
environment::OCF_RESKEY_crm_feature_set: [3.7.1]
environment::OCF_RESKEY_name: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_INSTANCE: [srv02-c8s-fujitsu]
environment::OCF_RESOURCE_PROVIDER: [alteeve]
environment::OCF_RESOURCE_TYPE: [server]
environment::OCF_ROOT: [/usr/lib/ocf]
environment::PCMK_debug: [0]
# Migrate (run on current host)
/usr/lib/ocf/resource.d/alteeve/server -vv --log-secure --server <server_name> --migrate-to <target_node_name> --migrate-from <current_node_name>
=cut
foreach my $key (sort {$a cmp $b} keys %{$anvil->data->{environment}})
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2 , list => {
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 1, list => {
"environment::${key}" => $anvil->data->{environment}{$key},
}});
}
foreach my $key (sort {$a cmp $b} keys %{$anvil->data->{switches}})
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 1, list => {
"switches::${key}" => $anvil->data->{switches}{$key},
}});
}
# Set environment variables from switches, if otherwise not set.
if (($anvil->data->{switches}{server}) && (not $anvil->data->{environment}{OCF_RESKEY_name}))
{
$anvil->data->{environment}{OCF_RESKEY_name} = $anvil->data->{switches}{server};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
"environment::OCF_RESKEY_name" => $anvil->data->{environment}{OCF_RESKEY_name},
}});
}
if (($anvil->data->{switches}{migrate_to}) && (not $anvil->data->{environment}{OCF_RESKEY_CRM_meta_migrate_target}))
{
$anvil->data->{environment}{OCF_RESKEY_CRM_meta_migrate_target} = $anvil->data->{switches}{migrate_to};
@ -376,13 +233,6 @@ if (($anvil->data->{switches}{migrate_from}) && (not $anvil->data->{environment}
"environment::OCF_RESKEY_CRM_meta_migrate_source" => $anvil->data->{environment}{OCF_RESKEY_CRM_meta_migrate_source},
}});
}
if (($anvil->data->{switches}{server}) && (not $anvil->data->{environment}{OCF_RESKEY_name}))
{
$anvil->data->{environment}{OCF_RESKEY_name} = $anvil->data->{switches}{server};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
"environment::OCF_RESKEY_name" => $anvil->data->{environment}{OCF_RESKEY_name},
}});
}
# This is for debugging.
if (not $anvil->data->{switches}{monitor})
@ -486,7 +336,8 @@ sub check_daemons
{
my ($anvil, $task) = @_;
my $problem = $anvil->Cluster->parse_cib();
my $problem = $anvil->Cluster->parse_cib({debug => 3});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 3, list => { problem => $problem }});
if ($problem)
{
# Pacemaker isn't running, or some other problem. Someone must have called this script
@ -498,7 +349,7 @@ sub check_daemons
# Is the peer running? We'll use this to know whether to try and start daemons on the peer.
my $peer_name = $anvil->data->{cib}{parsed}{peer}{name};
my $peer_ready = $anvil->data->{cib}{parsed}{peer}{ready};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2 , list => {
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 3 , list => {
peer_name => $peer_name,
peer_ready => $peer_ready,
}});
@ -513,7 +364,7 @@ sub check_daemons
my $running_peer = 0;
my ($output, $return_code) = $anvil->System->call({shell_call => $anvil->data->{path}{exe}{systemctl}." status ".$daemon});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2 , list => {
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 3 , list => {
output => $output,
return_code => $return_code,
}});
@ -522,7 +373,7 @@ sub check_daemons
# It is stopped, start it..
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0482", variables => { daemon => $daemon }});
my ($output, $return_code) = $anvil->System->call({shell_call => $anvil->data->{path}{exe}{systemctl}." start ".$daemon});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2 , list => {
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 3 , list => {
output => $output,
return_code => $return_code,
}});
@ -532,7 +383,7 @@ sub check_daemons
until ($running)
{
my ($output, $return_code) = $anvil->System->call({shell_call => $anvil->data->{path}{exe}{systemctl}." status ".$daemon});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2 , list => {
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 3 , list => {
output => $output,
return_code => $return_code,
}});
@ -588,7 +439,7 @@ sub check_daemons
target => $peer_name,
shell_call => $anvil->data->{path}{exe}{systemctl}." start ".$daemon,
});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2 , list => {
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 3 , list => {
output => $output,
error => $error,
return_code => $return_code,
@ -602,7 +453,7 @@ sub check_daemons
target => $peer_name,
shell_call => $anvil->data->{path}{exe}{systemctl}." status ".$daemon,
});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2 , list => {
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 3 , list => {
output => $output,
error => $error,
return_code => $return_code,
@ -847,6 +698,14 @@ sub start_server
my $server = $anvil->data->{environment}{OCF_RESKEY_name};
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0303", variables => { server => $server }});
if ((not $anvil->data->{environment}{OCF_RESKEY_CRM_meta_on_node}) && ($anvil->data->{switches}{target}))
{
$anvil->data->{environment}{OCF_RESKEY_CRM_meta_on_node} = $anvil->data->{switches}{target};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
"environment::OCF_RESKEY_CRM_meta_on_node" => $anvil->data->{environment}{OCF_RESKEY_CRM_meta_on_node},
}});
}
# Make sure things are sane.
validate_all($anvil);
@ -1362,23 +1221,11 @@ pmsuspended - The domain has been suspended by guest power management, e.g. ente
$anvil->nice_exit({exit_code => 1});
}
### TODO: Write the migration duration to /tmp/anvil.migration.<server>.data and have 'anvil-migrate-server' read that in to update the DB.
# Migrate the server
sub migrate_server
{
my ($anvil) = @_;
### This requires a database
# If we can connect to a database, we'll set/clear the 'migrating' flag during migrations. For timing
# reasons we don't let the RA do resyncs.
# $anvil->Database->connect({sensitive => 1});
# $anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 2, secure => 0, key => "log_0132"});
# if (not $anvil->data->{sys}{database}{connections})
# {
# # No databases, exit.
# $anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, secure => 0, key => "error_0003"});
# return(1);
# }
### NOTE: For now, we're not going to block if the target is not UpToDate. There are times when a
### user might want to do this (ie: sync will be done soon and the need to evacuate the node
digimer-bot
4 years ago
committed by
GitHub