WARNING: This is the _old_ Lustre wiki, and it is in the process of being retired. The information found here is all likely to be out of date. Please search the new wiki for more up to date information.

Using Red Hat Cluster Manager with Lustre: Difference between revisions

From Obsolete Lustre Wiki
Jump to navigationJump to search
 
(48 intermediate revisions by 2 users not shown)
Line 13: Line 13:
== Preliminary Notes ==
== Preliminary Notes ==


This document is based on the RedHat Cluster version 2.0, which is part of RedHat Enterprise Linux version 5.5. For other Versions or RHEL-based distributions the syntax or methods to setup and run RedHat Cluster may differ.
This document is based on the RedHat Cluster version 2.0, which is part of RedHat Enterprise Linux version 5.5. For other versions or RHEL-based distributions, the syntax or methods to set up and run RedHat Cluster may differ.


In comparison with other HA solutions RedHat Cluster as in RHEL 5.5 is a pretty old HA solution. It is recommended to use other HA solutions like pacemaker, if possible.
In comparison with other HA solutions, RedHat Cluster as in RHEL 5.5 is an old HA solution. We recommend using other HA solutions like Pacemaker, if possible.
 
It is assumed that two Lustre server nodes share a number of Lustre targets. Each Lustre node provides a number of Lustre targets and, in case of a failure, the active/non-failed node takes over the Lustre targets of the failed nodes and makes them available to the Lustre clients.
 
Furthermore, to make sure the Lustre targets are mounted only on one of Lustre server nodes at a time, STONITH fencing is implemented. This requires a way to make sure the failed node is shut down in case of a failure. In the examples below, it is assumed that the Lustre server nodes are equipped with a service processor allowing to shut down a failed node using IPMI. For other methods of fencing, refer to the RedHat Cluster documentation.


== Setting Up RedHat Cluster ==
== Setting Up RedHat Cluster ==
Setting up RedHat Cluster consists of three steps:
* setup ''openais'',
* configure the cluster and,
* start the RedHat cluster services


==== Setting Up the ''openais'' Communication Stack ====
==== Setting Up the ''openais'' Communication Stack ====
Line 45: Line 54:
ringnumber: 0
ringnumber: 0
bindnetaddr: 10.0.0.0
bindnetaddr: 10.0.0.0
mcastaddr: 226.94.1.1
mcastaddr: 226.94.1.1
mcastport: 5405
mcastport: 5405
}
}
}
}
</pre>
</pre>


''Openais'' uses the option ''bindnetaddr'' to determine which interface is to be used for cluster communication. The example above assumes one of the node’s interfaces is configured on the network 10.0.0.0. The value of the option is calculated from the IP address AND the network mask for the interface (IP & MASK) so the final bits of the address are cleared. Thus the configuration file is independent of any node and can be copied to all nodes.
''Openais'' uses the option ''bindnetaddr'' to determine which interface is to be used for cluster communication. In the example shown above, it is assumed that one of the node’s interfaces is configured on the network 10.0.0.0. The value of the option is calculated from the IP address AND the network mask for the interface (IP & MASK) so the final bits of the address are cleared. Thus the configuration file is independent of any node and can be copied to all nodes.


'''2. Create an AIS key
'''2. Create an AIS key
Line 63: Line 72:
== Installing RedHat Cluster ==
== Installing RedHat Cluster ==


The minimum installation of RedHat Cluster consists of the Cluster Manager package ''cman'' and the Resource Group Manager package ''rgmanager''. The ''cman'' package can be found in the RHEL repository. The ''rgmanager'' package is part of the Cluster repository. It can be found on the RHEL DVD in the Cluster sub-directory.
The minimum installation of RedHat Cluster consists of the Cluster Manager package ''cman'' and the Resource Group Manager package ''rgmanager''. The ''cman'' package can be found in the RHEL repository. The ''rgmanager'' package is part of the Cluster repository. It can be found on the RHEL DVD/ISO image in the ''Cluster'' sub-directory and may need to be added to the yum configuration manually. With yum configured correctly RedHat Cluster can be installed using:
<pre>
yum install cman rgmanager
</pre>
If yum is not set up correctly, the rpm packages and their dependencies need to be installed manually.


== Installing the Lustre Resource Skript ==
== Installing the Lustrefs resource script ==


The ''rgmanager'' package includes a number of resource scripts (''/usr/share/cluster'') which are used to integrate resources like network interfaces or file systems with ''rgmanager''. Unfortunately, there is no resource script for Lustre included.


== Configure RedHat Cluster ==
Luckily Giacomo Montagner posted an resource script on the lustre-discuss mailing list:


http://lists.lustre.org/pipermail/lustre-discuss/attachments/20090623/7799de37/attachment-0001.bin


After downloading this file it needs to be copied to /usr/share/cluster/lustrefs.sh.
Make sure the script is executable.


----
== Configure your Cluster ==


RedHat Cluster uses ''/etc/cluster/cluster.conf'' as central configuration file. This file is in XML format. The complete schema of the XML file can be found at http://sources.redhat.com/cluster/doc/cluster_schema_rhel5.html.


The Basic structure of a cluster.conf file may look like this:
<pre>
<?xml version="1.0" ?>
<cluster config_version="1" name="Lustre">
...
</cluster>
</pre>
In this example the name of the cluster is set to ''Lustre'' and the version is initialized as ''1''. If the cluster configuration is updated the config_version attribute must be increased on all nodes in this cluster.
RedHat cluster is usually used with more than two nodes providing resources. To tell RedHat cluster to work with two nodes the following ''cman'' attributes need to be set:
<pre>
  <cman expected_votes="1" two_node="1"/>
</pre>
This tells cman, that there are only two nodes in a cluster and one vote is enough declare a node failed.


=== Nodes ===
Next the nodes which form the cluster need to be specified. Each cluster node need to be specified separately wrapped in an surrounding ''clusternodes'' tag.
<pre>
  <clusternodes>
    <clusternode name="lustre1" nodeid="1">
      <fence>
        <method name="single">
          <device lanplus="1" name="lustre1-sp"/>
        </method>
      </fence>
    </clusternode>
    <clusternode name="lustre2" nodeid="2">
      <fence>
        <method name="single">
          <device lanplus="1" name="lustre2-sp"/>
        </method>
      </fence>
    </clusternode>
  </clusternodes>
</pre>
Each cluster node is given a name which must be it's hostname or IP address. Additionally a unique node ID needs to be specified.
The ''fence'' tag assigned to each node specifies a fence device to use to shut down this cluster node. The fence devices are defined elsewhere in ''cluster.conf'' (see below for details).


All services that the Pacemaker cluster resource manager will manage are called resources. The Pacemaker cluster resource manager uses resource agents to start, stop or monitor resources.
=== Fencing ===
 
'''''Note:''''' The simplest way to configure the cluster is by using a crm subshell. All examples will be given in this notation. If you understood the syntax of the cluster configuration, you also can use the GUI or XML notation.
 
==== Completing a Basic Setup of the Cluster ====
 
To test that your cluster manager is running and set global options, complete the steps below.


'''1. Display the cluster status.''' Enter:
Fencing is essential to keep data on the Lustre file system consistent. Even with Multi-Mount-Protection enabled, fencing can make sure that a node in an unclear state is brought down for more analysis by the administrator.


To configure fencing, first some fence daemon options need to be specified. the ''fence_daemon'' tag is a direct child of the ''cluster'' tag.
<pre>
<pre>
# crm_mon -1
  <fence_daemon post_fail_delay="0" post_join_delay="3"/>
  <fence_daemon clean_start="0"/>
</pre>
</pre>
Depending on the hardware configuration, these values may differ for different installations. Please see the notes in the cluster_schema_rhel5 document (linked above) for details.


The output should look similar to:
Each Lustre node in a cluster should be equipped with a fencing device. RedHat cluster supports a number of devices. More details on which devices are supported and how to configure them can be found in the cluster schema document.
 
For this example IPMI based fencing devices are used.
The ''fencedevices'' section may look like this:
<pre>
<pre>
============
  <fencedevices>
Last updated: Fri Dec 25 17:31:54 2009
    <fencedevice name="lustre1-sp" agent="fence_ipmilan" auth="password" ipaddr="10.0.1.1" login="root" passwd="supersecretpassword" option="off"/>
Stack&#58; openais
    <fencedevice name="lustre2-sp" agent="fence_ipmilan" auth="password" ipaddr="10.0.1.2" login="root" passwd="supersecretpassword" option="off"/>
Current DC: node1 - partition with quorum
  </fencedevices>
Version&#58; 1.0.6-cebe2b6ff49b36b29a3bd7ada1c4701c7470febe
2 Nodes configured, 2 expected votes
0 Resources configured.
============
 
Online&#58; [ node1 node2 ]
</pre>
</pre>
Every fence device has a number of attributes:
''name'' is used to define a name for this fencing device. This name is referred to in the ''fence'' part of the ''clusternode'' definition (see above). The ''agent'' defines the kind of fencing device to use. In this example an IPMI-over-Lan device is used. The remaining attributes are specific for the ''ipmilan'' device and are self-explanatory.


This output indicates that ''corosync'' started the cluster resource manager and it is ready to manage resources.
=== Resource Manager ===


Several global options must be set in the cluster. The two described in the next two steps are especially important to consider.
The resource manager block of the ''cluster.conf'' is wrapped in a ''rm'' tag:
 
<pre>
'''2. If your  cluster consists of just two nodes, switch the quorum feature off.''' On the command line, enter:
  <rm>
    ..
  </rm>
</pre>
It contains definitions of resources, failover domains, and services.


==== Resources ====
In the ''resources'' block of the ''cluster.conf'' file all Lustre targets of both clustered nodes are specified. In this example, four Lustre object storage targets are defined:
<pre>
<pre>
&#35; crm configure property no-quorum-policy=ignore
    <resources>
      <lustrefs name="target1" mountpoint="/mnt/ost1" device="/path/to/ost1/device" force_fsck="0" force_unmount="0" self_fence="1"/>
      <lustrefs name="target2" mountpoint="/mnt/ost2" device="/path/to/ost2/device" force_fsck="0" force_unmount="0" self_fence="1"/>
      <lustrefs name="target3" mountpoint="/mnt/ost3" device="/path/to/ost3/device" force_fsck="0" force_unmount="0" self_fence="1"/>
      <lustrefs name="target4" mountpoint="/mnt/ost4" device="/path/to/ost4/device" force_fsck="0" force_unmount="0" self_fence="1"/>
    </resources>
</pre>
</pre>
 
To use the ''lustrefs'' resource definition it is essential that the lustrefs.sh script is installed in ''/usr/share/cluster'' as described above. To verify the script is installed correctly and has correct permission run
If your lustre setup comprises more than two nodes, you can leave the no-quorum option as it is.
 
'''3. In a Lustre setup, fencing is normally used and is enabled by default. If you have a good reason not to use it, disable it by entering:'''
 
<pre>
<pre>
&#35; crm configure property stonith-enabled=false
# /usr/share/cluster/lustrefs.sh --help
usage: /usr/share/cluster/lustrefs.sh {start|stop|status|monitor|restart|meta-data|verify-all}
</pre>
</pre>
Each ''lustrefs'' resource has a number of attributes. ''name'' defines how the resource can be addressed.


After the global options of the cluster are set up correctly, continue to the following sections to configure resources and constraints.
==== Failover Domains ====
 
Usually RedHat cluster is used to provide a service on a number of nodes, where one node takes over the service of a failed node. In this example a number of Lustre targets is provided by each of the Lustre server nodes. To allow such a configuration, the definition of two Failover domains is necessary. The definition of ''failoverdomains'' may look like this:
==== Configuring Resources ====
 
OSTs are represented as Filesystem resources. A Lustre cluster consists of several Filesystem resources along with constraints that determine on which nodes of the cluster the resources can run.
 
By default, the start, stop, and monitor operations in a Filesystem resource time out after 20 sec. Since some mounts in Lustre require up to 5 minutes or more, the default timeouts for these operations must be modified. Also, a monitor operation must be added to the resource so that Pacemaker can check if the resource is still alive and react in case of any problems.
 
'''1. Create a definition of the Filesystem resource and save it in a file such as ''MyOST.res''.'''
 
If you have multiple OSTs, you will need to define additional resources.
 
The example below shows a complete definition of the Filesystem resource. You will need to change the ''device'' and ''directory'' to correspond to your setup.
 
<pre>
<pre>
primitive resMyOST ocf&#58;heartbeat&#58;Filesystem \
  <failoverdomains>
meta target-role="stopped" \
    <failoverdomain name="first_first" ordered="1" restricted="1">
operations $id="resMyOST-operations" \
        <failoverdomainnode name="lustre1" priority="1"/>
op monitor interval="120" timeout="60" \
        <failoverdomainnode name="lustre2" priority="2"/>
op start interval="0" timeout="300" \
      </failoverdomain>
op stop interval="0" timeout="300" \
      <failoverdomain name="second_first" ordered="1" restricted="1">       
params device="device" directory="directory" fstype="lustre"  
        <failoverdomainnode name="lustre1" priority="2"/>
        <failoverdomainnode name="lustre2" priority="1"/>
      </failoverdomain>
    </failoverdomains>
</pre>
</pre>
In this example, two fail-over-domains are created by adding the same nodes to each fail-over-domain, but the nodes are assigned different priorities.


In this example, the resource is initially stopped (''target-role=”stopped”'') because the constraints specifying where the resource is to be run have not yet been defined.


The ''start'' and ''stop'' operations have each been set to a timeout of 300 sec. The resource is monitored at intervals of 120 seconds. The parameters "''device''", "''directory''" and "lustre" are passed to the ''mount'' command.


'''2. Read the definition into your cluster configuration''' by entering:
==== Services ====
As a final configuration step the resources defined earlier are assigned to their fail-over-domain. This is done by defining a service for each of the Lustre nodes in the cluster and assign a domain. For the resources and fail-over-domains defined earlier this may look like this:  
<pre>
    <service autostart="1" exclusive="0" recovery="relocate" domain="first_first" name="lustre_2">
      <lustrefs ref="target1"/>
      <lustrefs ref="target2"/>
    </service>


# crm configure < MyOST.res
    <service autostart="1" exclusive="0" recovery="relocate" domain="second_first" name="lustre_1">
 
      <lustrefs ref="target3"/>
You can define as many OST resources as you want.
      <lustrefs ref="target4"/>
 
    </service>
If a server fails or the monitoring of a OST results in the detection of a failure, the cluster first tries to restart the resource on the failed node. If the node fails to restart it, the resource is migrated to another node.
</pre>
 
In this example ''target1'' and ''target2'' are assigned to the first node and ''target3'' and ''target4'' are assigned to the second node by default.
More sophosticated ways of failure management (such as trying to restart a node three times before migrating to another node) are possible using the cluster resource manager. See the Pacemaker documentation for details.
 
If mounting the file system depends on another resource like the start of a RAID or multipath driver, you can include this resource in the cluster configuration. This resource is then monitored by the cluster, enabling Pacemaker to react to failures.
 
==== Configuring Constraints ====
In a simple Lustre cluster setup, constraints are not required. However, in a larger cluster setup, you may want to use constraints to establish relationships between resources. For example, to keep the load distributed equally across nodes in your cluster, you may want to control how many OSTs can run on a particular node.
 
Constraints on resources are established by Pacemaker through a point system. Resources accumulate or lose points according to the constraints you define. If a resource has negative points with respect to a certain node, it cannot run on that node.
 
For example, to constrain the co-location of two resources, complete the steps below.
 
'''1. Add co-location constraints between resources.''' Enter commands similar to the following:


# crm configure colocation colresOST1resOST2 -100: resOST1 resOST2


This constraint assigns -100 points to resOST2 if an attempt is made to run resOST2 on the same node as resOST1. If the resulting total number of points assigned to reOST2 is negative, it will not be able run on that node.  
=== Start RedHat Cluster ===
Before bringing up RedHat cluster, make sure ''cluster.conf'' is update/edited on both Lustre server nodes. Usually ''cluster.conf'' should be the same on both nodes. The only exception is, if the device paths differ on both nodes.  


'''2. After defining all necessary constraints, start the resources.''' Enter:
==== ''cman'' service ====


# crm resource start resMyOST
With ''cluster.conf'' in place of both nodes it's time to start the ''cman'' service.
 
this is done by running
Execute this command for each OST (Filesystem resource) in the cluster.
 
 
----
 
'''Note:''' Use care when setting up your point system. You can use the point system if your cluster has at least three nodes or if the resource can acquire points from other constraints. However, in a system with only two nodes and no way to acquire points, the constraint in the example above will result in an inability to migrate a resource from a failed node.
 
For example, if resOST1 is running on ''node1'' and resOST2 on ''node2'' and ''node2'' fails, an attempt will be made to run resOST2 on ''node1''. However, the constraint will assign resOST2 -100 points since resOST1 is already running on ''node1''. Consequently resOST2 will be unable to run on ''node1'' and, since it is a two-node system, no other node is available.
 
----
 
To find out more about how the cluster resource manager calculates points, see the Pacemaker documentation.
 
==== Internal Monitoring of the System ====
 
In addition to monitoring of the resource itself, the nodes of the cluster must also be monitored. An important parameter to monitor is whether the node is connected to the network. Each node pings one or more hosts and counts the answers it receives. The number of responses determines how “good” its connection is to the network.
 
Pacemaker provides a simple way to configure this task.
 
'''1. Define a ping resource.''' In the command below, the ''host_list'' contains a list of hosts that the nodes should ping.
<pre>
<pre>
# crm configure resPing ocf:pacemaker:pingd \
service cman start
  params host_list=“host1 ...“ multiplier=“10“ dampen=”5s“
# crm configure clone clonePing resPing
</pre>
</pre>
on both clustered nodes. To verify ''cman'' is running ''clustat'' can be used:
<pre>
bash-3.2# clustat
Cluster Status for Lustre @ Tue Dec 14 11:27:36 2010
Member Status: Quorate


For every accessible host detected, any resource on that node gets 10 points (set by the ''multiplier='' parameter). The clone configuration makes the ping resource run on every available node.
Member Name                                      ID  Status
 
------ ----                                      ---- ------
'''2. Set up constraints to run a resource on the node with the best connectivity.''' The score from the ''ping'' resource can be used in other constraints to allow a resource to run only on those nodes that have a sufficient ping score. For example, enter:
lustre1                                            1 Online, Local
 
  lustre2                                            2 Online
  # crm configure location locMyOST resMyOST rule $id="locMyOST" pingd: defined pingd
</pre>
 
To enable the ''cman'' service permanently run:
This location constraint adds the ''ping'' score to the total score assigned to a resource for a particular node. The resource will tend to run on the node with the best connectivity.
 
Other system checks, such as CPU usage or free RAM, are measured by the Sysinfo resource. The capabilities of the Sysinfo resource are somewhat limited, so it will be replaced by the SystemHealth strategy in future releases of Pacemaker. For more information about the SystemHealth feature, see:
[http://www.clusterlabs.org/wiki/SystemHealth www.clusterlabs.org/wiki/SystemHealth]
 
==== Administering the Cluster ====
 
Careful system administration is required to support high availability in a cluster. A primary task of an administrator is to check the cluster for errors or failures of any resources. When a failure occurs, the administrator must search for the cause of the problem, solve it and then reset the corresponding failcounter.
This section describes some basic commands useful to an administrator. For more detailed information, see the Pacemaker documentation.
 
'''Displaying a Status Overview'''
 
The command ''crm_mon'' displays an overview of the status of the cluster. It functions similarly to the Linux top command by updating the output each time a cluster event occurs. To generate a one-time output, add the <br>
option ''-1''.
 
To include a display of all failcounters for all resources on the nodes, add the ''-f'' option to the command. The output of the command crm_mon -1f looks similar to:
 
<pre>
<pre>
============
chkconfig cman on
Last updated: Fri Dec 25 17:31:54 2009
Stack: openais
Current DC: node1 - partition with quorum
Version: 1.0.6-cebe2b6ff49b36b29a3bd7ada1c4701c7470febe
2 Nodes configured, 2 expected votes
2 Resources configured.
============
 
Online: [ node1 node2 ]
 
Clone Set: clonePing
    Started: [ node1 node2 ]
resMyOST      (ocf::heartbeat:filesys): Started node1
 
Migration summary:
* Node node1:  pingd=20
  resMyOST: migration-threshold=1000000 fail-count=1
* Node node2:  pingd=20
</pre>
</pre>


'''Switching a Node to Standby'''
==== ''rgmanager'' service ====
 
With ''cman'' up and running it's time to start the resource group manager ''rgmanager'' by running
You can switch a node to standby to, for example, perform maintenance on the node. In standby, the node is still a full member of the cluster but cannot run any resources. All resources that were running on that node are forced away.
 
To switch the node called ''node01'' to standby, enter:
 
# crm node standby node01
 
To switch the node online again enter:
 
# crm node online node01
 
'''Migrating a Resource to Another Node'''
 
The cluster resource manager can migrate a resource from one node to another while the resource is running. To migrate a resource away from the node it is running on, enter:
 
# crm resource migrate resMyOST
 
This command adds a location constraint to the configuration that specifies that the resource ''resMyOST'' can no longer run on the original node.
 
To delete this constraint, enter:
 
# crm resource unmigrate resMyOST
 
A target node can be specified in the migration command as follows:
 
# crm configure migrate resMyOST node02
 
This command causes the resource ''resMyOST'' to move to node ''node02'', while adding a location constraint to the configuration. To remove the location constraint, enter the ''unmigrate'' command again.
 
'''Resetting the failcounter'''
 
If Pacemaker monitors a resource and finds that it isn’t running, by default it restarts the resource on the node. If the resource cannot be restarted on the node, it then migrates the resource to another node.
 
It is the administrator’s task to find out the cause of the error and to reset the failcounter of the resource. This can be achieved by entering:
 
# crm resource failcount <resource> delete <node>
 
This command deletes (resets) the failcounter for the resource on the specified node.
 
'''“Cleaning up” a Resource'''
 
Sometimes it is necessary to “clean up” a resource. Internally, this command removes any information about a resource from the Local Resource Manager on every node and thus forces a complete re-read of the status of that resource. The command syntax is:
 
# crm resource cleanup resMyOST
 
This command removes information about the resource called ''resMyOST'' on all nodes.
 
== Setting up Fencing ==
 
Fencing is a technique used to isolate a node from the cluster when it is malfunctioning to prevent data corruption. For example, if a “split-brain” condition occurs in which two nodes can no longer communicate and both attempt to mount the same filesystem resource, data corruption can result. (The Multiple Mount Protection (MMP) mechanism in Lustre is designed to protect a file system from being mounted simultaneously by more than one node.)
 
Pacemaker uses the STONITH (Shoot The Other Node In The Head) approach to fencing malfunctioning nodes, in which a malfunctioning node is simply switched off. A good discussion about fencing can be found [http://www.clusterlabs.org/doc/crm_fencing.html here]. This article provides information useful for deciding which devices to purchase or how to set up STONITH resources for your cluster and also provides a detailed setup procedure.
 
A basic setup includes the following steps:
 
'''1. Test your fencing system manually before configuring the corresponding resources in the cluster.''' Manual testing is done by calling the STONITH command directly from each node. If this works in all tests, it will work in the cluster.
 
'''2. After configuring of the according resources, check that the system works as expected.''' To cause an artificial “split-brain” situation, you could use a host-based firewall to prohibit communication from other nodes on the heartbeat interface(s) by entering:
 
# iptables -I INPUT -i <heartbeat-IF> -p 5405 -s <other node> -j DROP
 
When the other nodes are not able to see the node isolated by the firewall, the isolated node should be shut down or rebooted.
 
== Setting Up Monitoring ==
 
Any cluster must to be monitored to provide the high availability it was designed for. Consider the following scenario demonstrating the importance of monitoring:
 
:''A node fails and all resources migrate to its backup node. Since the failover was smooth, nobody notices the problem. After some time, the second node fails and service stops. This is a serious problem since neither of the nodes is now able to provide service. The administrator must recover data from backups and possibly even install it on new hardware. A significant delay may result for users.''
 
Pacemaker offers several options for making information available to a monitoring system. These include:
*Utilizing the ''crm_mon'' program to send out information about changes in cluster status.
*Using scripts to check resource failcounters.
 
These options are described in the following sections.
 
==== Using ''crm_mon'' to Send Email Messages ====
In the most simple setup, the ''crm_mon'' program can be used to send out an email each time the status of the cluster changes. This approach requires a fully working mail environment and ''mail'' command.
 
Before configuring the ''crm_mon'' daemon, check that emails sent from the command line are delivered correctly by entering:
 
# crm_mon --daemonize –-mail-to <user@example.com> [--mail-host mail.example.com]
 
The resource monitor in the cluster can be configured to ensure the mail alerting service resource is running, as shown below:
 
<pre>
<pre>
primitive resMON ocf&#58;pacemaker&#58;ClusterMon \
service rgmanager start
operations $id="resMON-operations" \
op monitor interval="180" timeout="20" \
params extra_options="--mail-to <your@mail.address>"
</pre>
</pre>
rgmanager will than start to bring up the Lustre targets assigned to each of the Lustre nodes.


If a node fails, which could prevent the email from being sent, the resource is started on another node and an email about the successful start of the resource is sent out from the new node. The administrator's task is to search for the cause of the failover.
==== Verifying RedHat Cluster ====


==== Using ''crm_mon'' to Send SNMP Traps ====
To verify the state of the cluster run ''clustat'' again. With the above configuration the output should look like this:
The ''crm_mon'' daemon can be used to send SNMP traps to a network management server. The configuration from the command line is:
<pre>
bash-3.2# clustat
Cluster Status for Lustre @ Tue Dec 14 13:12:07 2010
Member Status: Quorate


  # crm_mon –-daemonize –-snmp-traps nms.example.com
  Member Name                                    ID  Status
------ ----                                    ---- ------
lustre1                                          1 Online, Local, rgmanager
lustre2                                          2 Online, rgmanager


This daemon can also be configured as a cluster resource as shown below:
  Service Name                      Owner (Last)                      State       
------- ----                      ----- ------                      -----       
service:lustre_1                  lustre1                            started     
service:lustre_2                  lustre2                            started     
</pre>


=== Relocate services ===
It may be necessary to relocate running lustre services manually. This can be done using
''clusvcadm'' as shown in the example below. First the service ''lustre_2'' is assigned to node ''lustre2''. After calling ''clusvcadm -r lustre_2'' this service is relocated to node ''lustre1'', as show in the last ''clustat'' output.
<pre>
<pre>
primitive resMON ocf&#58;pacemaker&#58;ClusterMon \
bash-3.2# clustat
operations $id="resMON-operations" \
Cluster Status for Lustre @ Tue Dec 14 15:00:00 2010
op monitor interval="180" timeout="20" \
Member Status: Quorate
params extra_options="--snmp-trap nms.example.com
</pre>


The MIB of the traps is defined in the ''PCMKR.txt'' file.
Member Name                            ID  Status
------ ----                            ---- ------
lustre1                                  1 Online, Local, rgmanager
lustre2                                  2 Online, rgmanager


==== Polling the Failcounters ====
Service Name                  Owner (Last)                  State       
If all the nodes of a cluster have problems, pushing information about events may be not be sufficient. An alternative is to check the failcounters of all resources periodically from the network management station (NMS). A simple script that checks for the presence of any failcounters in the output of ''crm_mon -1f'' is shown below:
------- ----                  ----- ------                  -----       
service:lustre_1              lustre1                      started     
service:lustre_2              lustre2                      started     
bash-3.2# clusvcadm -r lustre_2 
Trying to relocate service:lustre_2...Success
service:lustre_2 is now running on ldk-2-2-eth2
bash-3.2# clustat
Cluster Status for Lustre @ Tue Dec 14 15:01:00 2010
Member Status: Quorate


  # crm_mon -1f | grep fail-count
  Member Name                            ID  Status
------ ----                            ---- ------
lustre1                                  1 Online, Local, rgmanager
lustre2                                  2 Online, rgmanager


This script can be called by the NMS via SSH, or by the SNMP agent on the nodes by adding the following line to the Net-SNMP configuration in ''snmpd.conf'':
Service Name                  Owner (Last)                  State       
 
  ------- ----                  ----- ------                  -----       
  extend failcounter crm_mon -1f | grep -q fail-count
service:lustre_1              lustre1                      started     
service:lustre_2              lustre1                      started     
</pre>


The code returned by the script can be checked by the NMS using:
== Other tools to use with RedHat Cluster ==


snmpget <node> nsExtend.\”failcounter\”
RedHat cluster is a complex system of programs and services. There are a number of tools available to interact and/or make working with RedHat Cluster easier. In this section a number of these tools are presented. For more details read the man pages.  


A result of ''0'' indicates a failure.
; cman_tool : can be used to manage the cman subsystem. It can be used to add or remove nodes to a cluster configuration
; ccs_tool : may be used to update the configuration of the running cluster
; clustat : show the status of the cluster and if and where services are currently running
; clusvcadm : can be used to enable, disable or relocate services in a cluster
; system-config-cluster : a graphical user interface for cluster configuration

Latest revision as of 08:13, 20 December 2010

(Updated: Dec 2010)

DISCLAIMER - EXTERNAL CONTRIBUTOR CONTENT

This content was submitted by an external contributor. We provide this information as a resource for the Lustre™ open-source community, but we make no representation as to the accuracy, completeness or reliability of this information.

This page describes how to configure and use Red Hat Cluster Manager with Lustre failover. Sven Trautmann has contributed this content.

For more about Lustre failover, see Configuring Lustre for Failover.


Preliminary Notes

This document is based on the RedHat Cluster version 2.0, which is part of RedHat Enterprise Linux version 5.5. For other versions or RHEL-based distributions, the syntax or methods to set up and run RedHat Cluster may differ.

In comparison with other HA solutions, RedHat Cluster as in RHEL 5.5 is an old HA solution. We recommend using other HA solutions like Pacemaker, if possible.

It is assumed that two Lustre server nodes share a number of Lustre targets. Each Lustre node provides a number of Lustre targets and, in case of a failure, the active/non-failed node takes over the Lustre targets of the failed nodes and makes them available to the Lustre clients.

Furthermore, to make sure the Lustre targets are mounted only on one of Lustre server nodes at a time, STONITH fencing is implemented. This requires a way to make sure the failed node is shut down in case of a failure. In the examples below, it is assumed that the Lustre server nodes are equipped with a service processor allowing to shut down a failed node using IPMI. For other methods of fencing, refer to the RedHat Cluster documentation.

Setting Up RedHat Cluster

Setting up RedHat Cluster consists of three steps:

  • setup openais,
  • configure the cluster and,
  • start the RedHat cluster services

Setting Up the openais Communication Stack

The openais package is distributed with RHEL and can be installed using 

rpm -i /path/to/RHEL-DVD/Server/openais0.80.6-16.el5.x86_64.rpm

or 

yum install openais

if yum is configured to access the RHEL repository.

Once installed, the software looks for a configuration in the file /etc/ais/openais.conf .

Complete the following steps to set up the openais communication stack:

1. Edit the totem section of the openais.conf configuration file to designate the IP address and netmask of the interface(s) to be used. The totem section of the configuration file describes the way openais communicates between nodes. 

totem {
	version: 2
	secauth: off
	threads: 0
	interface {
		ringnumber: 0
		bindnetaddr: 10.0.0.0
		mcastaddr: 226.94.1.1
		mcastport: 5405
	}
}

Openais uses the option bindnetaddr to determine which interface is to be used for cluster communication. In the example shown above, it is assumed that one of the node’s interfaces is configured on the network 10.0.0.0. The value of the option is calculated from the IP address AND the network mask for the interface (IP & MASK) so the final bits of the address are cleared. Thus the configuration file is independent of any node and can be copied to all nodes.

2. Create an AIS key 

# /usr/sbin/ais-keygen
OpenAIS Authentication key generator.
Gathering 1024 bits for key from /dev/random.
Writing openais key to /etc/ais/authkey.

Installing RedHat Cluster

The minimum installation of RedHat Cluster consists of the Cluster Manager package cman and the Resource Group Manager package rgmanager. The cman package can be found in the RHEL repository. The rgmanager package is part of the Cluster repository. It can be found on the RHEL DVD/ISO image in the Cluster sub-directory and may need to be added to the yum configuration manually. With yum configured correctly RedHat Cluster can be installed using: 

yum install cman rgmanager

If yum is not set up correctly, the rpm packages and their dependencies need to be installed manually.

Installing the Lustrefs resource script

The rgmanager package includes a number of resource scripts (/usr/share/cluster) which are used to integrate resources like network interfaces or file systems with rgmanager. Unfortunately, there is no resource script for Lustre included.

Luckily Giacomo Montagner posted an resource script on the lustre-discuss mailing list:

http://lists.lustre.org/pipermail/lustre-discuss/attachments/20090623/7799de37/attachment-0001.bin

After downloading this file it needs to be copied to /usr/share/cluster/lustrefs.sh. Make sure the script is executable.

Configure your Cluster

RedHat Cluster uses /etc/cluster/cluster.conf as central configuration file. This file is in XML format. The complete schema of the XML file can be found at http://sources.redhat.com/cluster/doc/cluster_schema_rhel5.html.

The Basic structure of a cluster.conf file may look like this: 

<?xml version="1.0" ?>
<cluster config_version="1" name="Lustre">
...
</cluster>

In this example the name of the cluster is set to Lustre and the version is initialized as 1. If the cluster configuration is updated the config_version attribute must be increased on all nodes in this cluster. RedHat cluster is usually used with more than two nodes providing resources. To tell RedHat cluster to work with two nodes the following cman attributes need to be set: 

  <cman expected_votes="1" two_node="1"/>

This tells cman, that there are only two nodes in a cluster and one vote is enough declare a node failed.

Nodes

Next the nodes which form the cluster need to be specified. Each cluster node need to be specified separately wrapped in an surrounding clusternodes tag. 

  <clusternodes>
    <clusternode name="lustre1" nodeid="1">
      <fence>
        <method name="single">
          <device lanplus="1" name="lustre1-sp"/>
        </method>
      </fence>
    </clusternode>
    <clusternode name="lustre2" nodeid="2">
      <fence>
        <method name="single">
          <device lanplus="1" name="lustre2-sp"/>
        </method>
      </fence>
    </clusternode>
  </clusternodes>

Each cluster node is given a name which must be it's hostname or IP address. Additionally a unique node ID needs to be specified. The fence tag assigned to each node specifies a fence device to use to shut down this cluster node. The fence devices are defined elsewhere in cluster.conf (see below for details).

Fencing

Fencing is essential to keep data on the Lustre file system consistent. Even with Multi-Mount-Protection enabled, fencing can make sure that a node in an unclear state is brought down for more analysis by the administrator.

To configure fencing, first some fence daemon options need to be specified. the fence_daemon tag is a direct child of the cluster tag. 

  <fence_daemon post_fail_delay="0" post_join_delay="3"/>
  <fence_daemon clean_start="0"/>

Depending on the hardware configuration, these values may differ for different installations. Please see the notes in the cluster_schema_rhel5 document (linked above) for details.

Each Lustre node in a cluster should be equipped with a fencing device. RedHat cluster supports a number of devices. More details on which devices are supported and how to configure them can be found in the cluster schema document. For this example IPMI based fencing devices are used. The fencedevices section may look like this: 

  <fencedevices>
    <fencedevice name="lustre1-sp" agent="fence_ipmilan" auth="password" ipaddr="10.0.1.1" login="root" passwd="supersecretpassword" option="off"/>
    <fencedevice name="lustre2-sp" agent="fence_ipmilan" auth="password" ipaddr="10.0.1.2" login="root" passwd="supersecretpassword" option="off"/>
  </fencedevices>

Every fence device has a number of attributes: name is used to define a name for this fencing device. This name is referred to in the fence part of the clusternode definition (see above). The agent defines the kind of fencing device to use. In this example an IPMI-over-Lan device is used. The remaining attributes are specific for the ipmilan device and are self-explanatory.

Resource Manager

The resource manager block of the cluster.conf is wrapped in a rm tag: 

  <rm>
    ..
  </rm>

It contains definitions of resources, failover domains, and services.

Resources

In the resources block of the cluster.conf file all Lustre targets of both clustered nodes are specified. In this example, four Lustre object storage targets are defined: 

    <resources>
      <lustrefs name="target1" mountpoint="/mnt/ost1" device="/path/to/ost1/device" force_fsck="0" force_unmount="0" self_fence="1"/>
      <lustrefs name="target2" mountpoint="/mnt/ost2" device="/path/to/ost2/device" force_fsck="0" force_unmount="0" self_fence="1"/>
      <lustrefs name="target3" mountpoint="/mnt/ost3" device="/path/to/ost3/device" force_fsck="0" force_unmount="0" self_fence="1"/>
      <lustrefs name="target4" mountpoint="/mnt/ost4" device="/path/to/ost4/device" force_fsck="0" force_unmount="0" self_fence="1"/>
    </resources>

To use the lustrefs resource definition it is essential that the lustrefs.sh script is installed in /usr/share/cluster as described above. To verify the script is installed correctly and has correct permission run 

# /usr/share/cluster/lustrefs.sh --help
usage: /usr/share/cluster/lustrefs.sh {start|stop|status|monitor|restart|meta-data|verify-all}

Each lustrefs resource has a number of attributes. name defines how the resource can be addressed.

Failover Domains

Usually RedHat cluster is used to provide a service on a number of nodes, where one node takes over the service of a failed node. In this example a number of Lustre targets is provided by each of the Lustre server nodes. To allow such a configuration, the definition of two Failover domains is necessary. The definition of failoverdomains may look like this: 

  <failoverdomains>
     <failoverdomain name="first_first" ordered="1" restricted="1">
        <failoverdomainnode name="lustre1" priority="1"/>
        <failoverdomainnode name="lustre2" priority="2"/>
      </failoverdomain>
      <failoverdomain name="second_first" ordered="1" restricted="1">         
        <failoverdomainnode name="lustre1" priority="2"/>
        <failoverdomainnode name="lustre2" priority="1"/>
      </failoverdomain>
    </failoverdomains>

In this example, two fail-over-domains are created by adding the same nodes to each fail-over-domain, but the nodes are assigned different priorities.


Services

As a final configuration step the resources defined earlier are assigned to their fail-over-domain. This is done by defining a service for each of the Lustre nodes in the cluster and assign a domain. For the resources and fail-over-domains defined earlier this may look like this: 

    <service autostart="1" exclusive="0" recovery="relocate" domain="first_first" name="lustre_2">
      <lustrefs ref="target1"/>
      <lustrefs ref="target2"/>
    </service>

    <service autostart="1" exclusive="0" recovery="relocate" domain="second_first" name="lustre_1">
      <lustrefs ref="target3"/>
      <lustrefs ref="target4"/>
    </service>

In this example target1 and target2 are assigned to the first node and target3 and target4 are assigned to the second node by default.


Start RedHat Cluster

Before bringing up RedHat cluster, make sure cluster.conf is update/edited on both Lustre server nodes. Usually cluster.conf should be the same on both nodes. The only exception is, if the device paths differ on both nodes.

cman service

With cluster.conf in place of both nodes it's time to start the cman service. this is done by running 

service cman start

on both clustered nodes. To verify cman is running clustat can be used: 

bash-3.2# clustat 
Cluster Status for Lustre @ Tue Dec 14 11:27:36 2010
Member Status: Quorate

 Member Name                                      ID   Status
 ------ ----                                      ---- ------
 lustre1                                             1 Online, Local
 lustre2                                             2 Online

To enable the cman service permanently run: 

chkconfig cman on

rgmanager service

With cman up and running it's time to start the resource group manager rgmanager by running 

service rgmanager start

rgmanager will than start to bring up the Lustre targets assigned to each of the Lustre nodes.

Verifying RedHat Cluster

To verify the state of the cluster run clustat again. With the above configuration the output should look like this: 

bash-3.2# clustat
Cluster Status for Lustre @ Tue Dec 14 13:12:07 2010
Member Status: Quorate

 Member Name                                    ID   Status
 ------ ----                                    ---- ------
 lustre1                                           1 Online, Local, rgmanager
 lustre2                                           2 Online, rgmanager

 Service Name                       Owner (Last)                       State         
 ------- ----                       ----- ------                       -----         
 service:lustre_1                   lustre1                            started       
 service:lustre_2                   lustre2                            started

Relocate services

It may be necessary to relocate running lustre services manually. This can be done using clusvcadm as shown in the example below. First the service lustre_2 is assigned to node lustre2. After calling clusvcadm -r lustre_2 this service is relocated to node lustre1, as show in the last clustat output. 

bash-3.2# clustat
Cluster Status for Lustre @ Tue Dec 14 15:00:00 2010
Member Status: Quorate

 Member Name                            ID   Status
 ------ ----                            ---- ------
 lustre1                                   1 Online, Local, rgmanager
 lustre2                                   2 Online, rgmanager

 Service Name                   Owner (Last)                  State         
 ------- ----                   ----- ------                  -----         
 service:lustre_1               lustre1                       started       
 service:lustre_2               lustre2                       started       
bash-3.2# clusvcadm -r lustre_2  
Trying to relocate service:lustre_2...Success
service:lustre_2 is now running on ldk-2-2-eth2
bash-3.2# clustat
Cluster Status for Lustre @ Tue Dec 14 15:01:00 2010
Member Status: Quorate

 Member Name                            ID   Status
 ------ ----                            ---- ------
 lustre1                                   1 Online, Local, rgmanager
 lustre2                                   2 Online, rgmanager

 Service Name                   Owner (Last)                  State         
 ------- ----                   ----- ------                  -----         
 service:lustre_1               lustre1                       started       
 service:lustre_2               lustre1                       started

Other tools to use with RedHat Cluster

RedHat cluster is a complex system of programs and services. There are a number of tools available to interact and/or make working with RedHat Cluster easier. In this section a number of these tools are presented. For more details read the man pages.

cman_tool
can be used to manage the cman subsystem. It can be used to add or remove nodes to a cluster configuration
ccs_tool
may be used to update the configuration of the running cluster
clustat
show the status of the cluster and if and where services are currently running
clusvcadm
can be used to enable, disable or relocate services in a cluster
system-config-cluster
a graphical user interface for cluster configuration
Retrieved from "http://wiki.old.lustre.org/index.php?title=Using_Red_Hat_Cluster_Manager_with_Lustre&oldid=12054"