The clusterlabs stack incorporates Corosync and Pacemaker in an Open-Source, High Availability stack for both small and large deployments.
It supports a lot of different HA setups and is very flexible.
This puppet module is suitable for the management of both the software stack (pacemaker and corosync) and the cluster resources (via puppet types and providers).
Note: This module is the successor of puppetlabs-corosync.
To install and configure Corosync
class { 'corosync':
authkey => '/var/lib/puppet/ssl/certs/ca.pem',
bind_address => $::ipaddress,
cluster_name => 'mycluster',
enable_secauth => true,
}To enable Pacemaker
corosync::service { 'pacemaker':
version => '0',
}To configure advanced and (very) verbose logging settings
class { 'corosync':
log_stderr => false,
log_function_name => true,
syslog_priority => 'debug',
debug => true,
}To disable Corosync and Pacemaker services
class { 'corosync':
enable_corosync_service => false,
enable_pacemaker_service => false,
}By default the built-in Puppet CA will be used to perform this authentication, however, generating a dedicated key is a better approach.
-
Generate a new key on a machine with Corosync installed and convert it to Base64.
# Generate the key corosync-keygen -k /tmp/authkey -
Convert the key file to a Base64 string so it can be used in your manifest.
# Convert it to a Base64 string base64 -w 0 /tmp/authkey > /tmp/authkey_base64
-
Declare the corosync module using this string.
class { 'corosync': enable_secauth => true, authkey_source => 'string', authkey => 'MxjvpEztT3Mi+QagUO2cefhLDrP2BSFYKS3g1WXTUj2eCgGDPcSNf3uCKgzJKhoWTgJm2nYDHJv8KiFqMoW3ATuVr/9fLb/lgUVfoz0GnP10S7r77aqaIsERhJcGVQhcteHVlZl6zOo6VQz4ekH7VPmMlKJX0iQPuJTh9o6qhjg=', }
If the authkey is included directly in config, consider storing the value in hiera and encrypting it via hiera-eyaml.
The pacemaker/corosync configuration system (pcs) includes a daemon (pcsd) which can be configured to perform distributed communication across the cluster. This is accomplished by establishing token-based authorization of each cluster node via the pcs auth command.
On systems which support it, management of PCS authorization can be configured and deployed via this module as shown in the following example:
class { 'corosync':
manage_pcsd_service => true,
manage_pcsd_auth => true,
sensitive_hacluster_password => Sensitive('this-is-the-actual-password'),
sensitive_hacluster_hash => Sensitive('a-hash-of-the-passwd-for-the-user-resource'),
}Note that as this must only be executed on one node and by default the 'first' node in the cluster list is used. There may be timing issues if the configuration has not yet been applied on the other nodes as a successful execution requires the password for hacluster to be appropriately set on each system.
To enable Corosync 2 votequorum and define a nodelist of nodes named n1, n2, n3 with auto generated node IDs
class { 'corosync':
set_votequorum => true,
quorum_members => [ 'n1', 'n2', 'n3' ],
}To do the same but with custom node IDs instead
class { 'corosync':
set_votequorum => true,
quorum_members => [ 'n1', 'n2', 'n3' ],
quorum_members_ids => [ 10, 11, 12 ],
}Note: custom IDs may be required when adding or removing nodes to a cluster on a fly. Then each node shall have an unique and persistent ID.
To have multiple rings in the nodelist
class { 'corosync':
set_votequorum => true,
quorum_members => [
['172.31.110.1', '172.31.111.1'],
['172.31.110.2', '172.31.111.2'],
],
}When quorum_members is an array of arrays, each sub array represents one
host IP addresses.
Recent versions of corosync include support for a network based quorum device that is external to the cluster. This provides tiebreaker functionality to clusters with even node counts allowing 2-node or higher clusters which can operate with exactly half of their nodes to function. There are two components to quorum device configuration:
- A node which is not a member of any Corosync cluster will host the corosync-qnet daemon. This node should be outside of the network containing the cluster nodes.
- Each member of the cluster will be authorized to communicate with the quorum node and have the corosync-qdevice service scheduled and operating.
This implementation depends entirely on PCSD authorization and will only execute with that enabled.
-
Configure the qdevice class on the quorum node. Note that the same quorum node can be used for multiple clusters. Additionally, this node cannot be a normal cluster member!
# In this example, the node's name is quorum1.test.org class { 'corosync::qdevice': sensitive_hacluster_hash => Sensitive('hash-of-haclusters-password-on-the-qdevice-node') }
-
Configure and enable qdevice settings on the cluster members via the corosync main class.
class { 'corosync': cluster_name => 'example', manage_pcsd_service => true, manage_pcsd_auth => true, sensitive_hacluster_password => Sensitive('this-is-the-actual-password'), sensitive_hacluster_hash => Sensitive('a-hash-of-the-passwd-for-the-user-resource'), manage_quorum_device => true, quorum_device_host => 'quorum1.test.org', quorum_device_algorithm => 'ffsplit', sensitive_quorum_device_password => Sensitive('Actual password for hacluster on quorum1.test.org'), }
For more information see the following:
The resources that Corosync will manage can be referred to as a primitive. These are things like virtual IPs or services like drbd, nginx, and apache.
To assign a VIP to a network interface to be used by Nginx
cs_primitive { 'nginx_vip':
primitive_class => 'ocf',
primitive_type => 'IPaddr2',
provided_by => 'heartbeat',
parameters => { 'ip' => '172.16.210.100', 'cidr_netmask' => '24' },
operations => { 'monitor' => { 'interval' => '10s' } },
}Make Corosync manage and monitor the state of Nginx using a custom OCF agent
cs_primitive { 'nginx_service':
primitive_class => 'ocf',
primitive_type => 'nginx_fixed',
provided_by => 'pacemaker',
operations => {
'monitor' => { 'interval' => '10s', 'timeout' => '30s' },
'start' => { 'interval' => '0', 'timeout' => '30s', 'on-fail' => 'restart' }
},
require => Cs_primitive['nginx_vip'],
}Make Corosync manage and monitor the state of Apache using a LSB agent
cs_primitive { 'apache_service':
primitive_class => 'lsb',
primitive_type => 'apache2',
provided_by => 'heartbeat',
operations => {
'monitor' => { 'interval' => '10s', 'timeout' => '30s' },
'start' => { 'interval' => '0', 'timeout' => '30s', 'on-fail' => 'restart' }
},
require => Cs_primitive['apache2_vip'],
}Note: If you have multiple operations with the same names, you have to use an array. Example:
cs_primitive { 'pgsql_service':
primitive_class => 'ocf',
primitive_type => 'pgsql',
provided_by => 'heartbeat',
operations => [
{ 'monitor' => { 'interval' => '10s', 'timeout' => '30s' } },
{ 'monitor' => { 'interval' => '5s', 'timeout' => '30s' 'role' => 'Master', } },
{ 'start' => { 'interval' => '0', 'timeout' => '30s', 'on-fail' => 'restart' } }
],
}If you do mot want Puppet to interfere with manually stopped resources
(e.g not change the target-role metaparameter), you can use the
unmanaged_metadata parameter:
cs_primitive { 'pgsql_service':
primitive_class => 'ocf',
primitive_type => 'pgsql',
provided_by => 'heartbeat',
unmanaged_metadata => ['target-role'],
}Special primitives can be configured to support STONITH (Shoot The Other Node In The Head) fencing. This is critical for clusters which include shared resources (shared disk typically) or are vulnerable to cluster splits. The STONITH resource is responsible for providing a mechanism to restart or simply halt a rouge resource, often via power fencing.
The following example performs this configuration via the fence_vmware_soap STONITH agent.
cs_primitive { 'vmfence':
primitive_class => 'stonith',
primitive_type => 'fence_vmware_soap',
operations => {
'monitor' => { 'interval' => '60s'},
},
parameters => {
'ipaddr' => 'vcenter.example.org',
'login' => '[email protected]'
'passwd' => 'some plaintext secret',
'ssl' => '1',
'ssl_insecure' => '1',
'pcmk_host_map' => 'host0.example.org:host0;host1.example.org:host1',
'pcmk_delay_max' => '10s',
},
}Note that currently this implementation only handles STONITH for RHEL/CentOS based clusters which utilize pcs.
Locations determine on which nodes primitive resources run.
cs_location { 'nginx_service_location':
primitive => 'nginx_service',
node_name => 'hostname',
score => 'INFINITY'
}To manage rule on a location. Example to force the location to not run on a container (VM).
cs_location { 'nginx_service_location':
primitive => 'nginx_service',
rules => [
{ 'nginx-service-avoid-container-rule' => {
'score' => '-INFINITY',
'expression' => [
{ 'attribute' => '#kind',
'operation' => 'eq',
'value' => 'container'
},
],
},
},
],
}Example of a virtual ip location that checks ping connectivity for placement.
cs_location { 'vip-ping-connected':
primitive => 'vip',
rules => [
{ 'vip-ping-exclude-rule' => {
'score' => '-INFINITY',
'expression' => [
{ 'attribute' => 'pingd',
'operation' => 'lt',
'value' => '100',
},
],
},
},
{ 'vip-ping-prefer-rule' => {
'score-attribute' => 'pingd',
'expression' => [
{ 'attribute' => 'pingd',
'operation' => 'defined',
}
],
},
},
],
}Example of another possibility to use ping connectivity for placement.
cs_location { 'vip-ping-connected':
primitive => 'vip',
rules => [
{ 'vip-ping-connected-rule' => {
'score' => '-INFINITY',
'boolean-op' => 'or',
'expression' => [
{ 'attribute' => 'pingd',
'operation' => 'not_defined',
},
{ 'attribute' => 'pingd',
'operation' => 'lte',
'value' => '100',
},
],
},
},
],
}Colocations keep primitives together. Meaning if a vip moves to web02 from web01 because web01 just hit the dirt it will drag the nginx service with it.
cs_colocation { 'vip_with_service':
primitives => [ 'nginx_vip', 'nginx_service' ],
}pcs only Advanced colocations are also possible with colocation sets by using arrays instead of strings in the primitives array. Additionally, a hash can be added to the inner array with the specific options for that resource set.
cs_colocation { 'mysql_and_ptheartbeat':
primitives => [
['mysql', {'role' => 'master'}],
[ 'ptheartbeat' ],
],
}cs_colocation { 'mysql_apache_munin_and_ptheartbeat':
primitives => [
['mysql', 'apache', {'role' => 'master'}],
[ 'munin', 'ptheartbeat' ],
],
}Colocation defines that a set of primitives must live together on the same node but order definitions will define the order of which each primitive is started. If Nginx is configured to listen only on our vip we definitely want the vip to be migrated to a new node before nginx comes up or the migration will fail.
cs_order { 'vip_before_service':
first => 'nginx_vip',
second => 'nginx_service',
require => Cs_colocation['vip_with_service'],
}Cloned resources should be active on multiple hosts at the same time. You can clone any existing resource provided the resource agent supports it.
cs_clone { 'nginx_service-clone' :
ensure => present,
primitive => 'nginx_service',
clone_max => 3,
require => Cs_primitive['nginx_service'],
}You can also clone groups:
cs_clone { 'nginx_service-clone' :
ensure => present,
group => 'nginx_group',
clone_max => 3,
require => Cs_primitive['nginx_service'],
}A few global settings can be changed with the "cs_property" section.
Disable STONITH if required.
cs_property { 'stonith-enabled' :
value => 'false',
}Change quorum policy
cs_property { 'no-quorum-policy' :
value => 'ignore',
}You can use the replace parameter to create but not update some values:
cs_property { 'maintenance-mode':
value => 'true',
replace => false,
}A few global settings can be changed with the "cs_rsc_defaults" section.
Don't move resources.
cs_rsc_defaults { 'resource-stickiness' :
value => 'INFINITY',
}In unicast mode, you can have multiple rings by specifying unicast_address and bind_address as arrays:
class { 'corosync':
enable_secauth => true,
authkey => '/var/lib/puppet/ssl/certs/ca.pem',
bind_address => ['10.0.0.1', '10.0.1.1'],
unicast_addresses => [
[ '10.0.0.1',
'10.0.1.1'
], [
'10.0.0.2',
'10.0.1.2'
],
],
}The unicast_addresses is an array of arrays. One sub array matches one host IP addresses. In this example host2 has IP addresses 10.0.0.2 and 10.0.1.2.
Shadow CIB allows you to apply all the changes at the same time. For that, you
need to use the cib parameter and the cs_commit and cs_shadow types.
Shadow CIB is the recommended way to manage large CIB with puppet, as it will apply all your changes at once, starting the cluster when everything is in place: primitives, constraints, properties.
If you set the cib parameter to one cs_* resource we recommend you to set
that cib parameter to all the cs_* resources.
cs_shadow {
'puppet':
}
cs_primitive { 'pgsql_service':
primitive_class => 'ocf',
primitive_type => 'pgsql',
provided_by => 'heartbeat',
cib => 'puppet'
}
cs_commit {
'puppet':
}In Puppet < 4.0, you also need the resources to notify their cs_commit:
Cs_primitive['pgsql_service'] ~> Cs_commit['puppet']Tested and built on Debian 6 using backports so version 1.4.2 of Corosync is validated to function.
We suggest you at least go read the Clusters from Scratch document from Cluster Labs. It will help you out a lot when understanding how all the pieces fall together a point you in the right direction when Corosync/Pacemaker fails unexpectedly.
We do maintain a roadmap regarding next releases of this module.
| OS | release | Puppet 3.8.7 | Puppet 4 (PC1) | Puppet 5.X |
|---|---|---|---|---|
| CentOS/RHEL | 7 | Not supported | Supported | Supported |
| Debian | 9 | Not supported | Supported | Supported |
| Ubuntu | 16.04 | Not supported | Supported | Supported |
Special thanks to Puppet, Inc for initial development and Vox Pupuli to provide a platform that allows us to continue the development of this module.
See the contributing guide for details. Additionally, some general guidelines on PR structure can be found here.
Copyright © 2012-2014 Puppet Inc
Copyright © 2012-2018 Multiple contributors
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.