Chapter 3. Pacemaker Tools
3.1. Simplify administration using a cluster shell
In the dark past, configuring Pacemaker required the administrator to read and write XML. In true UNIX style, there were also a number of different commands that specialized in different aspects of querying and updating the cluster.
All of that has been greatly simplified with the creation of unified command-line shells (and GUIs) that hide all the messy XML scaffolding.
These shells take all the individual aspects required for managing and configuring a cluster, and pack them into one simple-to-use command line tool.
They even allow you to queue up several changes at once and commit them atomically.
Two popular command-line shells are pcs
and crmsh
. This edition of Clusters from Scratch is based on pcs
.
The two shells share many concepts but the scope, layout and syntax does differ, so make sure you read the version of this guide that corresponds to the software installed on your system.
Since pcs
has the ability to manage all aspects of the cluster (both corosync and pacemaker), it requires a specific cluster stack to be in use: corosync 2.0 or later with votequorum plus Pacemaker 1.1.8 or later.
Start by taking some time to familiarize yourself with what pcs
can do.
[root@pcmk-1 ~]# pcs
Usage: pcs [-f file] [-h] [commands]...
Control and configure pacemaker and corosync.
Options:
-h, --help Display usage and exit
-f file Perform actions on file instead of active CIB
--debug Print all network traffic and external commands run
--version Print pcs version information
Commands:
cluster Configure cluster options and nodes
resource Manage cluster resources
stonith Configure fence devices
constraint Set resource constraints
property Set pacemaker properties
acl Set pacemaker access control lists
status View cluster status
config View and manage cluster configuration
As you can see, the different aspects of cluster management are separated into categories: resource, cluster, stonith, property, constraint, and status. To discover the functionality available in each of these categories, one can issue the command pcs category
help
. Below is an example of all the options available under the status category.
[root@pcmk-1 ~]# pcs status help
Usage: pcs status [commands]...
View current cluster and resource status
Commands:
[status] [--full]
View all information about the cluster and resources (--full provides
more details)
resources
View current status of cluster resources
groups
View currently configured groups and their resources
cluster
View current cluster status
corosync
View current membership information as seen by corosync
nodes [corosync|both|config]
View current status of nodes from pacemaker. If 'corosync' is
specified, print nodes currently configured in corosync, if 'both'
is specified, print nodes from both corosync & pacemaker. If 'config'
is specified, print nodes from corosync & pacemaker configuration.
pcsd <node> ...
Show the current status of pcsd on the specified nodes
xml
View xml version of status (output from crm_mon -r -1 -X)
Additionally, if you are interested in the version and supported cluster stack(s) available with your Pacemaker installation, run:
[root@pcmk-1 ~]# pacemakerd --features
Pacemaker 1.1.12 (Build: a14efad)
Supporting v3.0.9: generated-manpages agent-manpages ascii-docs publican-docs ncurses libqb-logging libqb-ipc upstart systemd nagios corosync-native atomic-attrd acls
If the SNMP and/or email options are not listed, then Pacemaker was not built to support them. This may be by the choice of your distribution, or the required libraries may not have been available. Please contact whoever supplied you with the packages for more details.
Chapter 4. Start and Verify Cluster
Now that corosync is configured, it is time to start the cluster. The command below will start corosync and pacemaker on both nodes in the cluster. If you are issuing the start command from a different node than the one you ran the pcs cluster auth
command on earlier, you must authenticate on the current node you are logged into before you will be allowed to start the cluster.
[root@pcmk-1 ~]# pcs cluster start --all
pcmk-1: Starting Cluster...
pcmk-2: Starting Cluster...
An alternative to using the pcs cluster start --all
command is to issue either of the below command sequences on each node in the cluster separately:
# pcs cluster start
Starting Cluster...
or
# systemctl start corosync.service
# systemctl start pacemaker.service
In this example, we are not enabling the corosync and pacemaker services to start at boot. If a cluster node fails or is rebooted, you will need to run pcs cluster start nodename
(or --all
) to start the cluster on it. While you could enable the services to start at boot, requiring a manual start of cluster services gives you the opportunity to do a post-mortem investigation of a node failure before returning it to the cluster.
4.2. Verify Corosync Installation
First, use corosync-cfgtool
to check whether cluster communication is happy:
[root@pcmk-1 ~]# corosync-cfgtool -s
Printing ring status.
Local node ID 1
RING ID 0
id = 192.168.122.101
status = ring 0 active with no faults
We can see here that everything appears normal with our fixed IP address (not a 127.0.0.x loopback address) listed as the id, and no faults for the status.
If you see something different, you might want to start by checking the node’s network, firewall and selinux configurations.
Next, check the membership and quorum APIs:
[root@pcmk-1 ~]# corosync-cmapctl | grep members
runtime.totem.pg.mrp.srp.members.1.config_version (u64) = 0
runtime.totem.pg.mrp.srp.members.1.ip (str) = r(0) ip(192.168.122.101)
runtime.totem.pg.mrp.srp.members.1.join_count (u32) = 1
runtime.totem.pg.mrp.srp.members.1.status (str) = joined
runtime.totem.pg.mrp.srp.members.2.config_version (u64) = 0
runtime.totem.pg.mrp.srp.members.2.ip (str) = r(0) ip(192.168.122.102)
runtime.totem.pg.mrp.srp.members.2.join_count (u32) = 2
runtime.totem.pg.mrp.srp.members.2.status (str) = joined
[root@pcmk-1 ~]# pcs status corosync
Membership information
--------------------------
Nodeid Votes Name
1 1 pcmk-1 (local)
2 1 pcmk-2
You should see both nodes have joined the cluster.
4.3. Verify Pacemaker Installation
Now that we have confirmed that Corosync is functional, we can check the rest of the stack. Pacemaker has already been started, so verify the necessary processes are running:
[root@pcmk-1 ~]# ps axf
PID TTY STAT TIME COMMAND
2 ? S 0:00 [kthreadd]
...lots of processes...
1362 ? Ssl 0:35 corosync
1379 ? Ss 0:00 /usr/sbin/pacemakerd -f
1380 ? Ss 0:00 \_ /usr/libexec/pacemaker/cib
1381 ? Ss 0:00 \_ /usr/libexec/pacemaker/stonithd
1382 ? Ss 0:00 \_ /usr/libexec/pacemaker/lrmd
1383 ? Ss 0:00 \_ /usr/libexec/pacemaker/attrd
1384 ? Ss 0:00 \_ /usr/libexec/pacemaker/pengine
1385 ? Ss 0:00 \_ /usr/libexec/pacemaker/crmd
If that looks OK, check the pcs status
output:
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
WARNING: no stonith devices and stonith-enabled is not false
Last updated: Tue Dec 16 16:15:29 2014
Last change: Tue Dec 16 15:49:47 2014
Stack: corosync
Current DC: pcmk-2 (2) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
0 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
Finally, ensure there are no startup errors (aside from messages relating to not having STONITH configured, which are OK at this point):
[root@pcmk-1 ~]# journalctl | grep -i error
Other operating systems may report startup errors in other locations, for example /var/log/messages
.
Repeat these checks on the other node. The results should be the same.
Chapter 5. Create an Active/Passive Cluster
5.1. Explore the Existing Configuration
When Pacemaker starts up, it automatically records the number and details of the nodes in the cluster, as well as which stack is being used and the version of Pacemaker being used.
The first few lines of output should look like this:
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
WARNING: no stonith devices and stonith-enabled is not false
Last updated: Tue Dec 16 16:15:29 2014
Last change: Tue Dec 16 15:49:47 2014
Stack: corosync
Current DC: pcmk-2 (2) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
0 Resources configured
Online: [ pcmk-1 pcmk-2 ]
For those who are not of afraid of XML, you can see the raw cluster configuration and status by using the pcs cluster cib
command.
Example 5.1. The last XML you’ll see in this document
[root@pcmk-1 ~]# pcs cluster cib
<cib crm_feature_set="3.0.9" validate-with="pacemaker-2.3" epoch="5" num_updates="8" admin_epoch="0" cib-last-written="Tue Dec 16 15:49:47 2014" have-quorum="1" dc-uuid="2">
<configuration>
<crm_config>
<cluster_property_set id="cib-bootstrap-options">
<nvpair id="cib-bootstrap-options-have-watchdog" name="have-watchdog" value="false"/>
<nvpair id="cib-bootstrap-options-dc-version" name="dc-version" value="1.1.12-a14efad"/>
<nvpair id="cib-bootstrap-options-cluster-infrastructure" name="cluster-infrastructure" value="corosync"/>
<nvpair id="cib-bootstrap-options-cluster-name" name="cluster-name" value="mycluster"/>
</cluster_property_set>
</crm_config>
<nodes>
<node id="1" uname="pcmk-1"/>
<node id="2" uname="pcmk-2"/>
</nodes>
<resources/>
<constraints/>
</configuration>
<status>
<node_state id="2" uname="pcmk-2" in_ccm="true" crmd="online" crm-debug-origin="do_state_transition" join="member" expected="member">
<lrm id="2">
<lrm_resources/>
</lrm>
<transient_attributes id="2">
<instance_attributes id="status-2">
<nvpair id="status-2-shutdown" name="shutdown" value="0"/>
<nvpair id="status-2-probe_complete" name="probe_complete" value="true"/>
</instance_attributes>
</transient_attributes>
</node_state>
<node_state id="1" uname="pcmk-1" in_ccm="true" crmd="online" crm-debug-origin="do_state_transition" join="member" expected="member">
<lrm id="1">
<lrm_resources/>
</lrm>
<transient_attributes id="1">
<instance_attributes id="status-1">
<nvpair id="status-1-shutdown" name="shutdown" value="0"/>
<nvpair id="status-1-probe_complete" name="probe_complete" value="true"/>
</instance_attributes>
</transient_attributes>
</node_state>
</status>
</cib>
Before we make any changes, it’s a good idea to check the validity of the configuration.
[root@pcmk-1 ~]# crm_verify -L -V
error: unpack_resources: Resource start-up disabled since no STONITH resources have been defined
error: unpack_resources: Either configure some or disable STONITH with the stonith-enabled option
error: unpack_resources: NOTE: Clusters with shared data need STONITH to ensure data integrity
Errors found during check: config not valid
As you can see, the tool has found some errors.
In order to guarantee the safety of your data,
the default for STONITH
in Pacemaker is
enabled. However, it also knows when no STONITH configuration has been supplied and reports this as a problem (since the cluster would not be able to make progress if a situation requiring node fencing arose).
We will disable this feature for now and configure it later.
To disable STONITH, set the stonith-enabled cluster option to false:
[root@pcmk-1 ~]# pcs property set stonith-enabled=false
[root@pcmk-1 ~]# crm_verify -L
With the new cluster option set, the configuration is now valid.
The use of stonith-enabled=false
is completely inappropriate for a production cluster. It tells the cluster to simply pretend that failed nodes are safely powered off. Some vendors will refuse to support clusters that have STONITH disabled.
We disable STONITH here only to defer the discussion of its configuration, which can differ widely from one installation to the next. See
Section 8.1, “What is STONITH?” for information on why STONITH is important and details on how to configure it.
Our first resource will be a unique IP address that the cluster can bring up on either node. Regardless of where any cluster service(s) are running, end users need a consistent address to contact them on. Here, I will choose 192.168.122.120 as the floating address, give it the imaginative name ClusterIP and tell the cluster to check whether it is running every 30 seconds.
The chosen address must not already be in use on the network. Do not reuse an IP address one of the nodes already has configured.
[root@pcmk-1 ~]# pcs resource create ClusterIP ocf:heartbeat:IPaddr2 \
ip=192.168.122.120 cidr_netmask=32 op monitor interval=30s
Another important piece of information here is ocf:heartbeat:IPaddr2. This tells Pacemaker three things about the resource you want to add:
-
The first field (ocf in this case) is the standard to which the resource script conforms and where to find it.
-
The second field (heartbeat in this case) is standard-specific; for OCF resources, it tells the cluster which OCF namespace the resource script is in.
-
The third field (IPaddr2 in this case) is the name of the resource script.
To obtain a list of the available resource standards (the ocf part of ocf:heartbeat:IPaddr2), run:
[root@pcmk-1 ~]# pcs resource standards
ocf
lsb
service
systemd
stonith
To obtain a list of the available OCF resource providers (the heartbeat part of ocf:heartbeat:IPaddr2), run:
[root@pcmk-1 ~]# pcs resource providers
heartbeat
openstack
pacemaker
Finally, if you want to see all the resource agents available for a specific OCF provider (the IPaddr2 part of ocf:heartbeat:IPaddr2), run:
[root@pcmk-1 ~]# pcs resource agents ocf:heartbeat
CTDB
Delay
Dummy
Filesystem
IPaddr
IPaddr2
.
. (skipping lots of resources to save space)
.
rsyncd
slapd
symlink
tomcat
Now, verify that the IP resource has been added, and display the cluster’s status to see that it is now active:
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Tue Dec 16 17:44:40 2014
Last change: Tue Dec 16 17:44:26 2014
Stack: corosync
Current DC: pcmk-1 (1) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
1 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-1
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
Since our ultimate goal is high availability, we should test failover of our new resource before moving on.
First, find the node on which the IP address is running.
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Tue Dec 16 17:44:40 2014
Last change: Tue Dec 16 17:44:26 2014
Stack: corosync
Current DC: pcmk-1 (1) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
1 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-1
You can see that the status of the ClusterIP resource is Started on a particular node (in this example, pcmk-1). Shut down Pacemaker and Corosync on that machine to trigger a failover.
[root@pcmk-1 ~]# pcs cluster stop pcmk-1
Stopping Cluster...
A cluster command such as pcs cluster stop nodename
can be run from any node in the cluster, not just the affected node.
Verify that pacemaker and corosync are no longer running:
[root@pcmk-1 ~]# pcs status
Error: cluster is not currently running on this node
Go to the other node, and check the cluster status.
[root@pcmk-2 ~]# pcs status
Cluster name: mycluster
Last updated: Wed Dec 17 10:30:56 2014
Last change: Tue Dec 16 17:44:26 2014
Stack: corosync
Current DC: pcmk-2 (2) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
1 Resources configured
Online: [ pcmk-2 ]
OFFLINE: [ pcmk-1 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-2
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
Notice that pcmk-1 is OFFLINE for cluster purposes (its PCSD is still Online, allowing it to receive pcs
commands, but it is not participating in the cluster).
Also notice that ClusterIP is now running on pcmk-2 — failover happened automatically, and no errors are reported.
If a cluster splits into two (or more) groups of nodes that can no longer communicate with each other (aka. partitions), quorum is used to prevent resources from starting on more nodes than desired, which would risk data corruption.
A cluster has quorum when more than half of all known nodes are online in the same partition, or for the mathematically inclined, whenever the following equation is true:
total_nodes < 2 * active_nodes
For example, if a 5-node cluster split into 3- and 2-node paritions, the 3-node partition would have quorum and could continue serving resources. If a 6-node cluster split into two 3-node partitions, neither partition would have quorum; pacemaker’s default behavior in such cases is to stop all resources, in order to prevent data corruption.
Two-node clusters are a special case. By the above definition, a two-node cluster would only have quorum when both nodes are running. This would make the creation of a two-node cluster pointless,
but corosync has the ability to treat two-node clusters as if only one node is required for quorum.
The pcs cluster setup
command will automatically configure two_node: 1 in corosync.conf
, so a two-node cluster will "just work".
If you are using a different cluster shell, you will have to configure corosync.conf
appropriately yourself. If you are using older versions of corosync, you will have to ignore quorum at the pacemaker level, using pcs property set no-quorum-policy=ignore
(or the equivalent command if you are using a different cluster shell).
Now, simulate node recovery by restarting the cluster stack on pcmk-1, and check the cluster’s status. (It may take a little while before the cluster gets going on the node, but it eventually will look like the below.)
[root@pcmk-1 ~]# pcs cluster start pcmk-1
pcmk-1: Starting Cluster...
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Wed Dec 17 10:50:11 2014
Last change: Tue Dec 16 17:44:26 2014
Stack: corosync
Current DC: pcmk-2 (2) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
1 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-2
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
With older versions of pacemaker, the cluster might move the IP back to its original location (pcmk-1). Usually, this is no longer the case.
5.4. Prevent Resources from Moving after Recovery
In most circumstances, it is highly desirable to prevent healthy resources from being moved around the cluster. Moving resources almost always requires a period of downtime. For complex services such as databases, this period can be quite long.
To address this, Pacemaker has the concept of resource
stickiness, which controls how strongly a service prefers to stay running where it is. You may like to think of it as the "cost" of any downtime. By default, Pacemaker assumes there is zero cost associated with moving resources and will do so to achieve "optimal"
resource placement. We can specify a different stickiness for every resource, but it is often sufficient to change the default.
[root@pcmk-1 ~]# pcs resource defaults resource-stickiness=100
[root@pcmk-1 ~]# pcs resource defaults
resource-stickiness: 100
Older versions of pcs
required that rsc
be added after resource
in the above commands.
Chapter 6. Add Apache as a Cluster Service
Now that we have a basic but functional active/passive two-node cluster, we’re ready to add some real services. We’re going to start with Apache because it is a feature of many clusters and relatively simple to configure.
Before continuing, we need to make sure Apache is installed on both hosts. We also need the wget tool in order for the cluster to be able to check the status of the Apache server.
# yum install -y httpd wget
# firewall-cmd --permanent --add-service=http
# firewall-cmd --reload
Do not enable the httpd service. Services that are intended to be managed via the cluster software should never be managed by the OS.
It is often useful, however, to manually start the service, verify that it works, then stop it again, before adding it to the cluster. This allows you to resolve any non-cluster-related problems before continuing. Since this is a simple example, we’ll skip that step here.
6.2. Create Website Documents
We need to create a page for Apache to serve. On CentOS 7.1, the default Apache document root is /var/www/html, so we’ll create an index file there. For the moment, we will simplify things by serving a static site and manually synchronizing the data between the two nodes, so run this command on both nodes:
# cat <<-END >/var/www/html/index.html
<html>
<body>My Test Site - $(hostname)</body>
</html>
END
6.3. Enable the Apache status URL
In order to monitor the health of your Apache instance, and recover it if it fails, the resource agent used by Pacemaker assumes the server-status URL is available. On both nodes, enable the URL with:
# cat <<-END >/etc/httpd/conf.d/status.conf
<Location /server-status>
SetHandler server-status
Order deny,allow
Deny from all
Allow from 127.0.0.1
</Location>
END
If you are using a different operating system, server-status may already be enabled or may be configurable in a different location.
6.4. Configure the Cluster
At this point, Apache is ready to go, and all that needs to be done is to add it to the cluster. Let’s call the resource WebSite. We need to use an OCF resource script called apache in the heartbeat namespace.
The script’s only required parameter is the path to the main Apache configuration file, and we’ll tell the cluster to check once a minute that Apache is still running.
[root@pcmk-1 ~]# pcs resource create WebSite ocf:heartbeat:apache \
configfile=/etc/httpd/conf/httpd.conf \
statusurl="http://localhost/server-status" \
op monitor interval=1min
By default, the operation timeout for all resources' start, stop, and monitor operations is 20 seconds. In many cases, this timeout period is less than a particular resource’s advised timeout period. For the purposes of this tutorial, we will adjust the global operation timeout default to 240 seconds.
[root@pcmk-1 ~]# pcs resource op defaults timeout=240s
[root@pcmk-1 ~]# pcs resource op defaults
timeout: 240s
In a production cluster, it is usually better to adjust each resource’s start, stop, and monitor timeouts to values that are appropriate to the behavior observed in your environment, rather than adjust the global default.
After a short delay, we should see the cluster start Apache.
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Wed Dec 17 12:40:41 2014
Last change: Wed Dec 17 12:40:05 2014
Stack: corosync
Current DC: pcmk-2 (2) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
2 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-2
WebSite (ocf::heartbeat:apache): Started pcmk-1
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
Wait a moment, the WebSite resource isn’t running on the same host as our IP address!
If, in the pcs status
output, you see the WebSite resource has failed to start, then you’ve likely not enabled the status URL correctly. You can check whether this is the problem by running:
wget -O - http://127.0.0.1/server-status
If you see Connection refused in the output, then this is likely the problem. Ensure that Allow from 127.0.0.1 is present for the <Location /server-status> block.
6.5. Ensure Resources Run on the Same Host
To reduce the load on any one machine, Pacemaker will generally try to spread the configured resources across the cluster nodes. However, we can tell the cluster that two resources are related and need to run on the same host (or not at all). Here, we instruct the cluster that WebSite can only run on the host that ClusterIP is active on.
To achieve this, we use a colocation constraint that indicates it is mandatory for WebSite to run on the same node as ClusterIP. The "mandatory" part of the colocation constraint is indicated by using a score of INFINITY. The INFINITY score also means that if ClusterIP is not active anywhere, WebSite will not be permitted to run.
If ClusterIP is not active anywhere, WebSite will not be permitted to run anywhere.
Colocation constraints are "directional", in that they imply certain things about the order in which the two resources will have a location chosen. In this case, we’re saying that WebSite needs to be placed on the same machine as ClusterIP, which implies that the cluster must know the location of ClusterIP before choosing a location for WebSite.
[root@pcmk-1 ~]# pcs constraint colocation add WebSite with ClusterIP INFINITY
[root@pcmk-1 ~]# pcs constraint
Location Constraints:
Ordering Constraints:
Colocation Constraints:
WebSite with ClusterIP (score:INFINITY)
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Wed Dec 17 13:57:58 2014
Last change: Wed Dec 17 13:57:22 2014
Stack: corosync
Current DC: pcmk-2 (2) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
2 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-2
WebSite (ocf::heartbeat:apache): Started pcmk-2
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
6.6. Ensure Resources Start and Stop in Order
Like many services, Apache can be configured to bind to specific IP addresses on a host or to the wildcard IP address. If Apache binds to the wildcard, it doesn’t matter whether an IP address is added before or after Apache starts; Apache will respond on that IP just the same. However, if Apache binds only to certain IP address(es), the order matters: If the address is added after Apache starts, Apache won’t respond on that address.
To be sure our WebSite responds regardless of Apache’s address configuration, we need to make sure ClusterIP not only runs on the same node, but starts before WebSite. A colocation constraint only ensures the resources run together, not the order in which they are started and stopped.
We do this by adding an ordering constraint. By default, all order constraints are mandatory, which means that the recovery of ClusterIP will also trigger the recovery of WebSite.
[root@pcmk-1 ~]# pcs constraint order ClusterIP then WebSite
Adding ClusterIP WebSite (kind: Mandatory) (Options: first-action=start then-action=start)
[root@pcmk-1 ~]# pcs constraint
Location Constraints:
Ordering Constraints:
start ClusterIP then start WebSite (kind:Mandatory)
Colocation Constraints:
WebSite with ClusterIP (score:INFINITY)
6.7. Prefer One Node Over Another
Pacemaker does not rely on any sort of hardware symmetry between nodes, so it may well be that one machine is more powerful than the other. In such cases, it makes sense to host the resources on the more powerful node if it is available. To do this, we create a location constraint.
In the location constraint below, we are saying the WebSite resource prefers the node pcmk-1 with a score of 50. Here, the score indicates how badly we’d like the resource to run at this location.
[root@pcmk-1 ~]# pcs constraint location WebSite prefers pcmk-1=50
[root@pcmk-1 ~]# pcs constraint
Location Constraints:
Resource: WebSite
Enabled on: pcmk-1 (score:50)
Ordering Constraints:
start ClusterIP then start WebSite (kind:Mandatory)
Colocation Constraints:
WebSite with ClusterIP (score:INFINITY)
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Wed Dec 17 14:11:49 2014
Last change: Wed Dec 17 14:11:20 2014
Stack: corosync
Current DC: pcmk-2 (2) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
2 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-2
WebSite (ocf::heartbeat:apache): Started pcmk-2
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
Wait a minute, the resources are still on pcmk-2!
Even though WebSite now prefers to run on pcmk-1, that preference is (intentionally) less than the resource stickiness (how much we preferred not to have unnecessary downtime).
To see the current placement scores, you can use a tool called crm_simulate.
[root@pcmk-1 ~]# crm_simulate -sL
Current cluster status:
Online: [ pcmk-1 pcmk-2 ]
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-2
WebSite (ocf::heartbeat:apache): Started pcmk-2
Allocation scores:
native_color: ClusterIP allocation score on pcmk-1: 50
native_color: ClusterIP allocation score on pcmk-2: 200
native_color: WebSite allocation score on pcmk-1: -INFINITY
native_color: WebSite allocation score on pcmk-2: 100
Transition Summary:
6.8. Move Resources Manually
There are always times when an administrator needs to override the cluster and force resources to move to a specific location. In this example, we will force the WebSite to move to pcmk-1 by updating our previous location constraint with a score of INFINITY.
[root@pcmk-1 ~]# pcs constraint location WebSite prefers pcmk-1=INFINITY
[root@pcmk-1 ~]# pcs constraint
Location Constraints:
Resource: WebSite
Enabled on: pcmk-1 (score:INFINITY)
Ordering Constraints:
start ClusterIP then start WebSite (kind:Mandatory)
Colocation Constraints:
WebSite with ClusterIP (score:INFINITY)
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Wed Dec 17 14:19:34 2014
Last change: Wed Dec 17 14:18:37 2014
Stack: corosync
Current DC: pcmk-2 (2) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
2 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-1
WebSite (ocf::heartbeat:apache): Started pcmk-1
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
Once we’ve finished whatever activity required us to move the resources to pcmk-1 (in our case nothing), we can then allow the cluster to resume normal operation by removing the new constraint. Since we previously configured a default stickiness, the resources will remain on pcmk-1.
First, use the --full
option to get the constraint’s ID:
[root@pcmk-1 ~]# pcs constraint --full
Location Constraints:
Resource: WebSite
Enabled on: pcmk-1 (score:INFINITY) (id:location-WebSite-pcmk-1-INFINITY)
Ordering Constraints:
start ClusterIP then start WebSite (kind:Mandatory) (id:order-ClusterIP-WebSite-mandatory)
Colocation Constraints:
WebSite with ClusterIP (score:INFINITY) (id:colocation-WebSite-ClusterIP-INFINITY)
Then remove the desired contraint using its ID:
[root@pcmk-1 ~]# pcs constraint remove location-WebSite-pcmk-1-INFINITY
[root@pcmk-1 ~]# pcs constraint
Location Constraints:
Ordering Constraints:
start ClusterIP then start WebSite (kind:Mandatory)
Colocation Constraints:
WebSite with ClusterIP (score:INFINITY)
Note that the location constraint is now gone. If we check the cluster status, we can also see that (as expected) the resources are still active on pcmk-1.
# pcs status
Cluster name: mycluster
Last updated: Wed Dec 17 14:25:21 2014
Last change: Wed Dec 17 14:24:29 2014
Stack: corosync
Current DC: pcmk-2 (2) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
2 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-1
WebSite (ocf::heartbeat:apache): Started pcmk-1
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
Chapter 7. Replicate Storage Using DRBD
Even if you’re serving up static websites, having to manually synchronize the contents of that website to all the machines in the cluster is not ideal. For dynamic websites, such as a wiki, it’s not even an option. Not everyone care afford network-attached storage, but somehow the data needs to be kept in sync.
Enter DRBD, which can be thought of as network-based RAID-1.
7.1. Install the DRBD Packages
DRBD itself is included in the upstream kernel,
but we do need some utilities to use it effectively.
CentOS does not ship these utilities, so we need to enable a third-party repository to get them. Supported packages for many OSes are available from DRBD’s maker
LINBIT, but here we’ll use the free
ELRepo repository.
On both nodes, import the ELRepo package signing key, and enable the repository:
# rpm --import https://www.elrepo.org/RPM-GPG-KEY-elrepo.org
# rpm -Uvh http://www.elrepo.org/elrepo-release-7.0-2.el7.elrepo.noarch.rpm
Now, we can install the DRBD kernel module and utilities:
# yum install -y kmod-drbd84 drbd84-utils
The version of drbd84-utils shipped with CentOS 7.1 has a bug in the Pacemaker integration script. Until a fix is packaged, download the affected script directly from the upstream, on both nodes:
# curl -o /usr/lib/ocf/resource.d/linbit/drbd 'http://git.linbit.com/gitweb.cgi?p=drbd-utils.git;a=blob_plain;f=scripts/drbd.ocf;h=cf6b966341377a993d1bf5f585a5b9fe72eaa5f2;hb=c11ba026bbbbc647b8112543df142f2185cb4b4b'
This is a temporary fix that will be overwritten if the package is upgraded.
DRBD will not be able to run under the default SELinux security policies. If you are familiar with SELinux, you can modify the policies in a more fine-grained manner, but here we will simply exempt DRBD processes from SELinux control:
# semanage permissive -a drbd_t
We will configure DRBD to use port 7789, so allow that port from each host to the other:
[root@pcmk-1 ~]# firewall-cmd --permanent --add-rich-rule='rule family="ipv4" source address="192.168.122.102" port port="7789" protocol="tcp" accept'
success
[root@pcmk-1 ~]# firewall-cmd --reload
success
[root@pcmk-2 ~]# firewall-cmd --permanent --add-rich-rule='rule family="ipv4" source address="192.168.122.101" port port="7789" protocol="tcp" accept'
success
[root@pcmk-2 ~]# firewall-cmd --reload
success
In this example, we have only two nodes, and all network traffic is on the same LAN. In production, it is recommended to use a dedicated, isolated network for cluster-related traffic, so the firewall configuration would likely be different; one approach would be to add the dedicated network interfaces to the trusted zone.
7.2. Allocate a Disk Volume for DRBD
DRBD will need its own block device on each node. This can be a physical disk partition or logical volume, of whatever size you need for your data. For this document, we will use a 1GiB logical volume, which is more than sufficient for a single HTML file and (later) GFS2 metadata.
[root@pcmk-1 ~]# vgdisplay | grep -e Name -e Free
VG Name centos_pcmk-1
Free PE / Size 382 / 1.49 GiB
[root@pcmk-1 ~]# lvcreate --name drbd-demo --size 1G centos_pcmk-1
Logical volume "drbd-demo" created
[root@pcmk-1 ~]# lvs
LV VG Attr LSize Pool Origin Data% Meta% Move Log Cpy%Sync Convert
drbd-demo centos_pcmk-1 -wi-a----- 1.00g
root centos_pcmk-1 -wi-ao---- 5.00g
swap centos_pcmk-1 -wi-ao---- 1.00g
Repeat for the second node, making sure to use the same size:
[root@pcmk-1 ~]# ssh pcmk-2 -- lvcreate --name drbd-demo --size 1G centos_pcmk-2
Logical volume "drbd-demo" created
There is no series of commands for building a DRBD configuration, so simply run this on both nodes to use this sample configuration:
# cat <<END >/etc/drbd.d/wwwdata.res
resource wwwdata {
protocol C;
meta-disk internal;
device /dev/drbd1;
syncer {
verify-alg sha1;
}
net {
allow-two-primaries;
}
on pcmk-1 {
disk /dev/centos_pcmk-1/drbd-demo;
address 192.168.122.101:7789;
}
on pcmk-2 {
disk /dev/centos_pcmk-2/drbd-demo;
address 192.168.122.102:7789;
}
}
END
Edit the file to use the hostnames, IP addresses and logical volume paths of your nodes if they differ from the ones used in this guide.
The allow-two-primaries option would not normally be used in an active/passive cluster. We are adding it here for the convenience of changing to an active/active cluster later.
With the configuration in place, we can now get DRBD running.
These commands create the local metadata for the DRBD resource, ensure the DRBD kernel module is loaded, and bring up the DRBD resource. Run them on one node:
[root@pcmk-1 ~]# drbdadm create-md wwwdata
initializing activity log
NOT initializing bitmap
Writing meta data...
New drbd meta data block successfully created.
[root@pcmk-1 ~]# modprobe drbd
[root@pcmk-1 ~]# drbdadm up wwwdata
We can confirm DRBD’s status on this node:
[root@pcmk-1 ~]# cat /proc/drbd
version: 8.4.6 (api:1/proto:86-101)
GIT-hash: 833d830e0152d1e457fa7856e71e11248ccf3f70 build by phil@Build64R7, 2015-04-10 05:13:52
1: cs:WFConnection ro:Secondary/Unknown ds:Inconsistent/DUnknown C r----s
ns:0 nr:0 dw:0 dr:0 al:0 bm:0 lo:0 pe:0 ua:0 ap:0 ep:1 wo:f oos:1048508
Because we have not yet initialized the data, this node’s data is marked as Inconsistent. Because we have not yet initialized the second node, the local state is WFConnection (waiting for connection), and the partner node’s status is marked as Unknown.
Now, repeat the above commands on the second node. This time, when we check the status, it shows:
[root@pcmk-2 ~]# cat /proc/drbd
version: 8.4.6 (api:1/proto:86-101)
GIT-hash: 833d830e0152d1e457fa7856e71e11248ccf3f70 build by phil@Build64R7, 2015-04-10 05:13:52
1: cs:Connected ro:Secondary/Secondary ds:Inconsistent/Inconsistent C r-----
ns:0 nr:0 dw:0 dr:0 al:0 bm:0 lo:0 pe:0 ua:0 ap:0 ep:1 wo:f oos:1048508
You can see the state has changed to Connected, meaning the two DRBD nodes are communicating properly, and both nodes are in Secondary role with Inconsistent data.
To make the data consistent, we need to tell DRBD which node should be considered to have the correct data. In this case, since we are creating a new resource, both have garbage, so we’ll just pick pcmk-1 and run this command on it:
[root@pcmk-1 ~]# drbdadm primary --force wwwdata
If you are using an older version of DRBD, the required syntax may be different. See the documentation for your version for how to perform these commands.
If we check the status immediately, we’ll see something like this:
[root@pcmk-1 ~]# cat /proc/drbd
version: 8.4.6 (api:1/proto:86-101)
GIT-hash: 833d830e0152d1e457fa7856e71e11248ccf3f70 build by phil@Build64R7, 2015-04-10 05:13:52
1: cs:SyncSource ro:Primary/Secondary ds:UpToDate/Inconsistent C r-----
ns:2872 nr:0 dw:0 dr:3784 al:0 bm:0 lo:0 pe:0 ua:0 ap:0 ep:1 wo:f oos:1045636
[>....................] sync'ed: 0.4% (1045636/1048508)K
finish: 0:10:53 speed: 1,436 (1,436) K/sec
We can see that this node has the Primary role, the partner node has the Secondary role, this node’s data is now considered UpToDate, the partner node’s data is still Inconsistent, and a progress bar shows how far along the partner node is in synchronizing the data.
After a while, the sync should finish, and you’ll see something like:
[root@pcmk-1 ~]# cat /proc/drbd
version: 8.4.6 (api:1/proto:86-101)
GIT-hash: 833d830e0152d1e457fa7856e71e11248ccf3f70 build by phil@Build64R7, 2015-04-10 05:13:52
1: cs:Connected ro:Primary/Secondary ds:UpToDate/UpToDate C r-----
ns:1048508 nr:0 dw:0 dr:1049420 al:0 bm:0 lo:0 pe:0 ua:0 ap:0 ep:1 wo:f oos:0
Both sets of data are now UpToDate, and we can proceed to creating and populating a filesystem for our WebSite resource’s documents.
7.5. Populate the DRBD Disk
On the node with the primary role (pcmk-1 in this example), create a filesystem on the DRBD device:
[root@pcmk-1 ~]# mkfs.xfs /dev/drbd1
meta-data=/dev/drbd1 isize=256 agcount=4, agsize=65532 blks
= sectsz=512 attr=2, projid32bit=1
= crc=0 finobt=0
data = bsize=4096 blocks=262127, imaxpct=25
= sunit=0 swidth=0 blks
naming =version 2 bsize=4096 ascii-ci=0 ftype=0
log =internal log bsize=4096 blocks=853, version=2
= sectsz=512 sunit=0 blks, lazy-count=1
realtime =none extsz=4096 blocks=0, rtextents=0
In this example, we create an xfs filesystem with no special options. In a production environment, you should choose a filesystem type and options that are suitable for your application.
Mount the newly created filesystem, populate it with our web document, give it the same SELinux policy as the web document root, then unmount it (the cluster will handle mounting and unmounting it later):
[root@pcmk-1 ~]# mount /dev/drbd1 /mnt
[root@pcmk-1 ~]# cat <<-END >/mnt/index.html
<html>
<body>My Test Site - DRBD</body>
</html>
END
[root@pcmk-1 ~]# chcon -R --reference=/var/www/html /mnt
[root@pcmk-1 ~]# umount /dev/drbd1
7.6. Configure the Cluster for the DRBD device
One handy feature pcs
has is the ability to queue up several changes into a file and commit those changes atomically. To do this, start by populating the file with the current raw XML config from the CIB.
[root@pcmk-1 ~]# pcs cluster cib drbd_cfg
Using the pcs -f
option, make changes to the configuration saved in the drbd_cfg
file. These changes will not be seen by the cluster until the drbd_cfg
file is pushed into the live cluster’s CIB later.
Here, we create a cluster resource for the DRBD device, and an additional clone resource to allow the resource to run on both nodes at the same time.
[root@pcmk-1 ~]# pcs -f drbd_cfg resource create WebData ocf:linbit:drbd \
drbd_resource=wwwdata op monitor interval=60s
[root@pcmk-1 ~]# pcs -f drbd_cfg resource master WebDataClone WebData \
master-max=1 master-node-max=1 clone-max=2 clone-node-max=1 \
notify=true
[root@pcmk-1 ~]# pcs -f drbd_cfg resource show
ClusterIP (ocf::heartbeat:IPaddr2): Started
WebSite (ocf::heartbeat:apache): Started
Master/Slave Set: WebDataClone [WebData]
Stopped: [ pcmk-1 pcmk-2 ]
After you are satisfied with all the changes, you can commit them all at once by pushing the drbd_cfg file into the live CIB.
[root@pcmk-1 ~]# pcs cluster cib-push drbd_cfg
CIB updated
Early versions of pcs
required push cib
in place of cib-push
above.
Let’s see what the cluster did with the new configuration:
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Fri Aug 14 09:29:41 2015
Last change: Fri Aug 14 09:29:25 2015
Stack: corosync
Current DC: pcmk-1 (1) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
4 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-1
WebSite (ocf::heartbeat:apache): Started pcmk-1
Master/Slave Set: WebDataClone [WebData]
Masters: [ pcmk-1 ]
Slaves: [ pcmk-2 ]
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
We can see that WebDataClone (our DRBD device) is running as master (DRBD’s primary role) on pcmk-1 and slave (DRBD’s secondary role) on pcmk-2.
The resource agent should load the DRBD module when needed if it’s not already loaded. If that does not happen, configure your operating system to load the module at boot time. For CentOS 7.1, you would run this on both nodes:
# echo drbd >/etc/modules-load.d/drbd.conf
7.7. Configure the Cluster for the Filesystem
Now that we have a working DRBD device, we need to mount its filesystem.
In addition to defining the filesystem, we also need to tell the cluster where it can be located (only on the DRBD Primary) and when it is allowed to start (after the Primary was promoted).
We are going to take a shortcut when creating the resource this time. Instead of explicitly saying we want the ocf:heartbeat:Filesystem script, we are only going to ask for Filesystem. We can do this because we know there is only one resource script named Filesystem available to pacemaker, and that pcs is smart enough to fill in the ocf:heartbeat: portion for us correctly in the configuration. If there were multiple Filesystem scripts from different OCF providers, we would need to specify the exact one we wanted.
Once again, we will queue our changes to a file and then push the new configuration to the cluster as the final step.
[root@pcmk-1 ~]# pcs cluster cib fs_cfg
[root@pcmk-1 ~]# pcs -f fs_cfg resource create WebFS Filesystem \
device="/dev/drbd1" directory="/var/www/html" fstype="xfs"
[root@pcmk-1 ~]# pcs -f fs_cfg constraint colocation add WebFS with WebDataClone INFINITY with-rsc-role=Master
[root@pcmk-1 ~]# pcs -f fs_cfg constraint order promote WebDataClone then start WebFS
Adding WebDataClone WebFS (kind: Mandatory) (Options: first-action=promote then-action=start)
We also need to tell the cluster that Apache needs to run on the same machine as the filesystem and that it must be active before Apache can start.
[root@pcmk-1 ~]# pcs -f fs_cfg constraint colocation add WebSite with WebFS INFINITY
[root@pcmk-1 ~]# pcs -f fs_cfg constraint order WebFS then WebSite
Adding WebFS WebSite (kind: Mandatory) (Options: first-action=start then-action=start)
Review the updated configuration.
[root@pcmk-1 ~]# pcs -f fs_cfg constraint
Location Constraints:
Ordering Constraints:
start ClusterIP then start WebSite (kind:Mandatory)
promote WebDataClone then start WebFS (kind:Mandatory)
start WebFS then start WebSite (kind:Mandatory)
Colocation Constraints:
WebSite with ClusterIP (score:INFINITY)
WebFS with WebDataClone (score:INFINITY) (with-rsc-role:Master)
WebSite with WebFS (score:INFINITY)
[root@pcmk-1 ~]# pcs -f fs_cfg resource show
ClusterIP (ocf::heartbeat:IPaddr2): Started
WebSite (ocf::heartbeat:apache): Started
Master/Slave Set: WebDataClone [WebData]
Masters: [ pcmk-1 ]
Slaves: [ pcmk-2 ]
WebFS (ocf::heartbeat:Filesystem): Stopped
After reviewing the new configuration, upload it and watch the cluster put it into effect.
[root@pcmk-1 ~]# pcs cluster cib-push fs_cfg
[root@pcmk-1 ~]# pcs status
Last updated: Fri Aug 14 09:34:11 2015
Last change: Fri Aug 14 09:34:09 2015
Stack: corosync
Current DC: pcmk-1 (1) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
5 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-1
WebSite (ocf::heartbeat:apache): Started pcmk-1
Master/Slave Set: WebDataClone [WebData]
Masters: [ pcmk-1 ]
Slaves: [ pcmk-2 ]
WebFS (ocf::heartbeat:Filesystem): Started pcmk-1
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
7.8. Test Cluster Failover
Previously, we used pcs cluster stop pcmk-1
to stop all cluster services on pcmk-1, failing over the cluster resources, but there is another way to safely simulate node failure.
We can put the node into standby mode. Nodes in this state continue to run corosync and pacemaker but are not allowed to run resources. Any resources found active there will be moved elsewhere. This feature can be particularly useful when performing system administration tasks such as updating packages used by cluster resources.
Put the active node into standby mode, and observe the cluster move all the resources to the other node. The node’s status will change to indicate that it can no longer host resources.
[root@pcmk-1 ~]# pcs cluster standby pcmk-1
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Fri Aug 14 09:36:49 2015
Last change: Fri Aug 14 09:36:43 2015
Stack: corosync
Current DC: pcmk-1 (1) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
5 Resources configured
Node pcmk-1 (1): standby
Online: [ pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-2
WebSite (ocf::heartbeat:apache): Started pcmk-2
Master/Slave Set: WebDataClone [WebData]
Masters: [ pcmk-2 ]
Stopped: [ pcmk-1 ]
WebFS (ocf::heartbeat:Filesystem): Started pcmk-2
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
Once we’ve done everything we needed to on pcmk-1 (in this case nothing, we just wanted to see the resources move), we can allow the node to be a full cluster member again.
[root@pcmk-1 ~]# pcs cluster unstandby pcmk-1
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Fri Aug 14 09:38:02 2015
Last change: Fri Aug 14 09:37:56 2015
Stack: corosync
Current DC: pcmk-1 (1) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
5 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-2
WebSite (ocf::heartbeat:apache): Started pcmk-2
Master/Slave Set: WebDataClone [WebData]
Masters: [ pcmk-2 ]
Slaves: [ pcmk-1 ]
WebFS (ocf::heartbeat:Filesystem): Started pcmk-2
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
Notice that pcmk-1 is back to the Online state, and that the cluster resources stay where they are due to our resource stickiness settings configured earlier.
Chapter 8. Configure STONITH
STONITH (Shoot The Other Node In The Head aka. fencing) protects your data from being corrupted by rogue nodes or unintended concurrent access.
Just because a node is unresponsive doesn’t mean it has stopped accessing your data. The only way to be 100% sure that your data is safe, is to use STONITH to ensure that the node is truly offline before allowing the data to be accessed from another node.
STONITH also has a role to play in the event that a clustered service cannot be stopped. In this case, the cluster uses STONITH to force the whole node offline, thereby making it safe to start the service elsewhere.
8.2. Choose a STONITH Device
It is crucial that your STONITH device can allow the cluster to differentiate between a node failure and a network failure.
A common mistake people make when choosing a STONITH device is to use a remote power switch (such as many on-board IPMI controllers) that shares power with the node it controls. If the power fails in such a case, the cluster cannot be sure whether the node is really offline, or active and suffering from a network fault, so the cluster will stop all resources to avoid a possible split-brain situation.
Likewise, any device that relies on the machine being active (such as SSH-based "devices" sometimes used during testing) is inappropriate.
8.3. Configure the Cluster for STONITH
-
Install the STONITH agent(s). To see what packages are available, run yum search fence-
. Be sure to install the package(s) on all cluster nodes.
-
Configure the STONITH device itself to be able to fence your nodes and accept fencing requests. This includes any necessary configuration on the device and on the nodes, and any firewall or SELinux changes needed. Test the communication between the device and your nodes.
-
Find the correct STONITH agent script: pcs stonith list
-
Find the parameters associated with the device: pcs stonith describe agent_name
-
Create a local copy of the CIB: pcs cluster cib stonith_cfg
-
Create the fencing resource: pcs -f stonith_cfg stonith create stonith_id stonith_device_type [stonith_device_options]
Any flags that do not take arguments, such as --ssl
, should be passed as ssl=1
.
-
Enable STONITH in the cluster: pcs -f stonith_cfg property set stonith-enabled=true
-
If the device does not know how to fence nodes based on their uname, you may also need to set the special pcmk_host_map parameter. See man stonithd
for details.
-
If the device does not support the list command, you may also need to set the special pcmk_host_list and/or pcmk_host_check parameters. See man stonithd
for details.
-
If the device does not expect the victim to be specified with the port parameter, you may also need to set the special pcmk_host_argument parameter. See man stonithd
for details.
-
Commit the new configuration: pcs cluster cib-push stonith_cfg
-
Once the STONITH resource is running, test it (you might want to stop the cluster on that machine first): stonith_admin --reboot nodename
For this example, assume we have a chassis containing four nodes and an IPMI device active on 10.0.0.1. Following the steps above would go something like this:
Step 1: Install the fence-agents-ipmilan package on both nodes.
Step 2: Configure the IP address, authentication credentials, etc. in the IPMI device itself.
Step 3: Choose the fence_ipmilan STONITH agent.
Step 4: Obtain the agent’s possible parameters:
[root@pcmk-1 ~]# pcs stonith describe fence_ipmilan
Stonith options for: fence_ipmilan
ipport: TCP/UDP port to use for connection with device
inet6_only: Forces agent to use IPv6 addresses only
ipaddr (required): IP Address or Hostname
passwd_script: Script to retrieve password
method: Method to fence (onoff|cycle)
inet4_only: Forces agent to use IPv4 addresses only
passwd: Login password or passphrase
lanplus: Use Lanplus to improve security of connection
auth: IPMI Lan Auth type.
cipher: Ciphersuite to use (same as ipmitool -C parameter)
privlvl: Privilege level on IPMI device
action (required): Fencing Action
login: Login Name
verbose: Verbose mode
debug: Write debug information to given file
version: Display version information and exit
help: Display help and exit
power_wait: Wait X seconds after issuing ON/OFF
login_timeout: Wait X seconds for cmd prompt after login
power_timeout: Test X seconds for status change after ON/OFF
delay: Wait X seconds before fencing is started
ipmitool_path: Path to ipmitool binary
shell_timeout: Wait X seconds for cmd prompt after issuing command
retry_on: Count of attempts to retry power on
sudo: Use sudo (without password) when calling 3rd party sotfware.
stonith-timeout: How long to wait for the STONITH action (reboot, on, off) to complete per a stonith device.
priority: The priority of the stonith resource. Devices are tried in order of highest priority to lowest.
pcmk_host_map: A mapping of host names to ports numbers for devices that do not support host names.
pcmk_host_list: A list of machines controlled by this device (Optional unless pcmk_host_check=static-list).
pcmk_host_check: How to determine which machines are controlled by the device.
Step 5: pcs cluster cib stonith_cfg
Step 6: Here are example parameters for creating our STONITH resource:
[root@pcmk-1 ~]# pcs -f stonith_cfg stonith create ipmi-fencing fence_ipmilan \
pcmk_host_list="pcmk-1 pcmk-2" ipaddr=10.0.0.1 login=testuser \
passwd=acd123 op monitor interval=60s
[root@pcmk-1 ~]# pcs -f stonith_cfg stonith
ipmi-fencing (stonith:fence_ipmilan): Stopped
Steps 7-10: Enable STONITH in the cluster:
[root@pcmk-1 ~]# pcs -f stonith_cfg property set stonith-enabled=true
[root@pcmk-1 ~]# pcs -f stonith_cfg property
Cluster Properties:
cluster-infrastructure: corosync
cluster-name: mycluster
dc-version: 1.1.12-a14efad
have-watchdog: false
stonith-enabled: true
Step 11: pcs cluster cib-push stonith_cfg
Step 12: Test:
[root@pcmk-1 ~]# pcs cluster stop pcmk-2
[root@pcmk-1 ~]# stonith_admin --reboot pcmk-2
After a successful test, login to any rebooted nodes, and start the cluster (with pcs cluster start
).
Chapter 9. Convert Cluster to Active/Active
The primary requirement for an Active/Active cluster is that the data required for your services is available, simultaneously, on both machines. Pacemaker makes no requirement on how this is achieved; you could use a SAN if you had one available, but since DRBD supports multiple Primaries, we can continue to use it here.
9.1. Install Cluster Filesystem Software
The only hitch is that we need to use a cluster-aware filesystem. The one we used earlier with DRBD, xfs, is not one of those. Both OCFS2 and GFS2 are supported; here, we will use GFS2.
On both nodes, install the GFS2 command-line utilities and the Distributed Lock Manager (DLM) required by cluster filesystems:
# yum install -y gfs2-utils dlm
9.2. Configure the Cluster for the DLM
The DLM needs to run on both nodes, so we’ll start by creating a resource for it (using the ocf:pacemaker:controld resource script), and clone it:
[root@pcmk-1 ~]# pcs cluster cib dlm_cfg
[root@pcmk-1 ~]# pcs -f dlm_cfg resource create dlm ocf:pacemaker:controld op monitor interval=60s
[root@pcmk-1 ~]# pcs -f dlm_cfg resource clone dlm clone-max=2 clone-node-max=1
[root@pcmk-1 ~]# pcs -f dlm_cfg resource show
ClusterIP (ocf::heartbeat:IPaddr2): Started
WebSite (ocf::heartbeat:apache): Started
Master/Slave Set: WebDataClone [WebData]
Masters: [ pcmk-2 ]
Slaves: [ pcmk-1 ]
WebFS (ocf::heartbeat:Filesystem): Started
Clone Set: dlm-clone [dlm]
Stopped: [ pcmk-1 pcmk-2 ]
Activate our new configuration, and see how the cluster responds:
[root@pcmk-1 ~]# pcs cluster cib-push dlm_cfg
CIB updated
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Fri Aug 14 11:19:36 2015
Last change: Fri Aug 14 11:19:28 2015
Stack: corosync
Current DC: pcmk-1 (1) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
8 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
ClusterIP (ocf::heartbeat:IPaddr2): Started pcmk-2
WebSite (ocf::heartbeat:apache): Started pcmk-2
Master/Slave Set: WebDataClone [WebData]
Masters: [ pcmk-2 ]
Slaves: [ pcmk-1 ]
WebFS (ocf::heartbeat:Filesystem): Started pcmk-2
ipmi-fencing (stonith:fence_ipmilan): Started pcmk-1
Clone Set: dlm-clone [dlm]
Started: [ pcmk-1 pcmk-2 ]
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
9.3. Create and Populate GFS2 Filesystem
Before we do anything to the existing partition, we need to make sure it is unmounted. We do this by telling the cluster to stop the WebFS resource. This will ensure that other resources (in our case, Apache) using WebFS are not only stopped, but stopped in the correct order.
[root@pcmk-1 ~]# pcs resource disable WebFS
[root@pcmk-1 ~]# pcs resource
ClusterIP (ocf::heartbeat:IPaddr2): Started
WebSite (ocf::heartbeat:apache): Stopped
Master/Slave Set: WebDataClone [WebData]
Masters: [ pcmk-2 ]
Slaves: [ pcmk-1 ]
WebFS (ocf::heartbeat:Filesystem): Stopped
Clone Set: dlm-clone [dlm]
Started: [ pcmk-1 pcmk-2 ]
You can see that both Apache and WebFS have been stopped, and that pcmk-2 is the current master for the DRBD device.
Now we can create a new GFS2 filesystem on the DRBD device.
This will erase all previous content stored on the DRBD device. Ensure you have a copy of any important data.
Run the next command on whichever node has the DRBD Primary role. Otherwise, you will receive the message:
/dev/drbd1: Read-only file system
[root@pcmk-2 ~]# mkfs.gfs2 -p lock_dlm -j 2 -t mycluster:web /dev/drbd1
It appears to contain an existing filesystem (xfs)
This will destroy any data on /dev/drbd1
Are you sure you want to proceed? [y/n]y
Device: /dev/drbd1
Block size: 4096
Device size: 1.00 GB (262127 blocks)
Filesystem size: 1.00 GB (262126 blocks)
Journals: 2
Resource groups: 5
Locking protocol: "lock_dlm"
Lock table: "mycluster:web"
UUID: 9a72c488-d8a7-24c9-ceee-add7a8ca52c2
The mkfs.gfs2
command required a number of additional parameters:
-
-p lock_dlm
specifies that we want to use the kernel’s DLM.
-
-j 2
indicates that the filesystem should reserve enough space for two journals (one for each node that will access the filesystem).
-
-t mycluster:web
specifies the lock table name. The format for this field is clustername:fsname
. For clustername
, we need to use the same value we specified originally with pcs cluster setup --name
(which is also the value of cluster_name in /etc/corosync/corosync.conf
). If you are unsure what your cluster name is, you can look in /etc/corosync/corosync.conf
or execute the command pcs cluster corosync pcmk-1 | grep cluster_name
.
Now we can (re-)populate the new filesystem with data (web pages). We’ll create yet another variation on our home page.
[root@pcmk-2 ~]# mount /dev/drbd1 /mnt
[root@pcmk-2 ~]# cat <<-END >/mnt/index.html
<html>
<body>My Test Site - GFS2</body>
</html>
END
[root@pcmk-2 ~]# chcon -R --reference=/var/www/html /mnt
[root@pcmk-2 ~]# umount /dev/drbd1
[root@pcmk-2 ~]# drbdadm verify wwwdata
9.4. Reconfigure the Cluster for GFS2
With the WebFS resource stopped, let’s update the configuration.
[root@pcmk-1 ~]# pcs resource show WebFS
Resource: WebFS (class=ocf provider=heartbeat type=Filesystem)
Attributes: device=/dev/drbd1 directory=/var/www/html fstype=xfs
Meta Attrs: target-role=Stopped
Operations: start interval=0s timeout=60 (WebFS-start-timeout-60)
stop interval=0s timeout=60 (WebFS-stop-timeout-60)
monitor interval=20 timeout=40 (WebFS-monitor-interval-20)
The fstype option needs to be updated to gfs2 instead of xfs.
[root@pcmk-1 ~]# pcs resource update WebFS fstype=gfs2
[root@pcmk-1 ~]# pcs resource show WebFS
Resource: WebFS (class=ocf provider=heartbeat type=Filesystem)
Attributes: device=/dev/drbd1 directory=/var/www/html fstype=gfs2
Meta Attrs: target-role=Stopped
Operations: start interval=0s timeout=60 (WebFS-start-timeout-60)
stop interval=0s timeout=60 (WebFS-stop-timeout-60)
monitor interval=20 timeout=40 (WebFS-monitor-interval-20)
GFS2 requires that DLM be running, so we also need to set up new colocation and ordering constraints for it:
[root@pcmk-1 ~]# pcs constraint colocation add WebFS with dlm-clone INFINITY
[root@pcmk-1 ~]# pcs constraint order dlm-clone then WebFS
Adding dlm-clone WebFS (kind: Mandatory) (Options: first-action=start then-action=start)
9.5. Clone the IP address
There’s no point making the services active on both locations if we can’t reach them both, so let’s clone the IP address.
The IPaddr2 resource agent has built-in intelligence for when it is configured as a clone. It will utilize a multicast MAC address to have the local switch send the relevant packets to all nodes in the cluster, together with iptables clusterip rules on the nodes so that any given packet will be grabbed by exactly one node. This will give us a simple but effective form of load-balancing requests between our two nodes.
Let’s start a new config, and clone our IP:
[root@pcmk-1 ~]# pcs cluster cib loadbalance_cfg
[root@pcmk-1 ~]# pcs -f loadbalance_cfg resource clone ClusterIP \
clone-max=2 clone-node-max=2 globally-unique=true
-
clone-max=2
tells the resource agent to split packets this many ways. This should equal the number of nodes that can host the IP.
-
clone-node-max=2
says that one node can run up to 2 instances of the clone. This should also equal the number of nodes that can host the IP, so that if any node goes down, another node can take over the failed node’s "request bucket". Otherwise, requests intended for the failed node would be discarded.
-
globally-unique=true
tells the cluster that one clone isn’t identical to another (each handles a different "bucket"). This also tells the resource agent to insert iptables rules so each host only processes packets in its bucket(s).
Notice that when the ClusterIP becomes a clone, the constraints referencing ClusterIP now reference the clone. This is done automatically by pcs.
[root@pcmk-1 ~]# pcs -f loadbalance_cfg constraint
Location Constraints:
Ordering Constraints:
start ClusterIP-clone then start WebSite (kind:Mandatory)
promote WebDataClone then start WebFS (kind:Mandatory)
start WebFS then start WebSite (kind:Mandatory)
start dlm-clone then start WebFS (kind:Mandatory)
Colocation Constraints:
WebSite with ClusterIP-clone (score:INFINITY)
WebFS with WebDataClone (score:INFINITY) (with-rsc-role:Master)
WebSite with WebFS (score:INFINITY)
WebFS with dlm-clone (score:INFINITY)
Now we must tell the resource how to decide which requests are processed by which hosts. To do this, we specify the clusterip_hash parameter. The value of sourceip means that the source IP address of incoming packets will be hashed; each node will process a certain range of hashes.
[root@pcmk-1 ~]# pcs -f loadbalance_cfg resource update ClusterIP clusterip_hash=sourceip
Load our configuration to the cluster, and see how it responds.
[root@pcmk-1 ~]# pcs cluster cib-push loadbalance_cfg
CIB updated
[root@pcmk-1 ~]# pcs status
Cluster name: mycluster
Last updated: Fri Aug 14 11:32:07 2015
Last change: Fri Aug 14 11:32:04 2015
Stack: corosync
Current DC: pcmk-1 (1) - partition with quorum
Version: 1.1.12-a14efad
2 Nodes configured
9 Resources configured
Online: [ pcmk-1 pcmk-2 ]
Full list of resources:
WebSite (ocf::heartbeat:apache): Stopped
Master/Slave Set: WebDataClone [WebData]
Masters: [ pcmk-1 ]
Slaves: [ pcmk-2 ]
WebFS (ocf::heartbeat:Filesystem): Stopped
ipmi-fencing (stonith:fence_ipmilan): Started pcmk-1
Clone Set: dlm-clone [dlm]
Started: [ pcmk-1 pcmk-2 ]
Clone Set: ClusterIP-clone [ClusterIP] (unique)
ClusterIP:0 (ocf::heartbeat:IPaddr2): Started pcmk-1
ClusterIP:1 (ocf::heartbeat:IPaddr2): Started pcmk-2
PCSD Status:
pcmk-1: Online
pcmk-2: Online
Daemon Status:
corosync: active/disabled
pacemaker: active/disabled
pcsd: active/enabled
If desired, you can demonstrate that all request buckets are working by using a tool such as arping
from several source hosts to see which host responds to each.
9.6. Clone the Filesystem and Apache Resources
Now that we have a cluster filesystem ready to go, and our nodes can load-balance requests to a shared IP address, we can configure the cluster so both nodes mount the filesystem and respond to web requests.
Clone the filesystem and Apache resources in a new configuration. Notice how pcs automatically updates the relevant constraints again.
[root@pcmk-1 ~]# pcs cluster cib active_cfg
[root@pcmk-1 ~]# pcs -f active_cfg resource clone WebFS
[root@pcmk-1 ~]# pcs -f active_cfg resource clone WebSite
[root@pcmk-1 ~]# pcs -f active_cfg constraint
Location Constraints:
Ordering Constraints:
start ClusterIP-clone then start WebSite-clone (kind:Mandatory)
promote WebDataClone then start WebFS-clone (kind:Mandatory)
start WebFS-clone then start WebSite-clone (kind:Mandatory)
start dlm-clone then start WebFS-clone (kind:Mandatory)
Colocation Constraints:
WebSite-clone with ClusterIP-clone (score:INFINITY)
WebFS-clone with WebDataClone (score:INFINITY) (with-rsc-role:Master)
WebSite-clone with WebFS-clone (score:INFINITY)
WebFS-clone with dlm-clone (score:INFINITY)
Tell the cluster that it is now allowed to promote both instances to be DRBD Primary (aka. master).
[root@pcmk-1 ~]# pcs -f active_cfg resource update WebDataClone master-max=2
Finally, load our configuration to the cluster, and re-enable the WebFS resource (which we disabled earlier).
[root@pcmk-1 ~]# pcs cluster cib-push active_cfg
CIB updated
[root@pcmk-1 ~]# pcs resource enable WebFS
After all the processes are started, the status should look similar to this.
[root@pcmk-1 ~]# pcs resource
Master/Slave Set: WebDataClone [WebData]
Masters: [ pcmk-1 pcmk-2 ]
Clone Set: dlm-clone [dlm]
Started: [ pcmk-1 pcmk-2 ]
Clone Set: ClusterIP-clone [ClusterIP] (unique)
ClusterIP:0 (ocf::heartbeat:IPaddr2): Started
ClusterIP:1 (ocf::heartbeat:IPaddr2): Started
Clone Set: WebFS-clone [WebFS]
Started: [ pcmk-1 pcmk-2 ]
Clone Set: WebSite-clone [WebSite]
Started: [ pcmk-1 pcmk-2 ]
Testing failover is left as an exercise for the reader. For example, you can put one node into standby mode, use pcs status
to confirm that its ClusterIP clone was moved to the other node, and use arping
to verify that packets are not being lost from any source host.
You may find that when a failed node rejoins the cluster, both ClusterIP clones stay on one node, due to the resource stickiness. While this works fine, it effectively eliminates load-balancing and returns the cluster to an active-passive setup again. You can avoid this by disabling stickiness for the IP address resource:
[root@pcmk-1 ~]# pcs resource meta ClusterIP resource-stickiness=0