Using Red Hat Enterprise Linux HA with Caché

Red Hat Linux

When configuring Linux on the nodes in the cluster, use the following guidelines:

1. All nodes in the cluster must have identical userids/groupids (that is, the name and ID number must be identical on all nodes); this is required for Caché.

2. Sufficient kernel resources (shmmax, and so on) must be configured on the nodes.

3. Only the minimum necessary Linux packages should be installed. Anything started past run level 3 (such as graphical consoles, and so on) is unnecessary for most Caché-based application and database servers. Each unnecessary package lengthens the time to boot and slows recovery after failures.

Red Hat Enterprise Linux HA

Install and test a standard shared disk cluster service using the following guidelines:

1. Configure the <clusternodes> such that each has one vote. Be sure to use fully qualified domain names as the name of each node.

2. Configure the <fence_daemon> with no post_fail_delay. This speeds the time to recovery.

3. For a two-node cluster with quorum disk, the <cman> configuration should include the following:

   expected_votes="3" two_node="0"
   

   

4. Configure <totem> to have a token timeout equal to or greater than the <cman> quorum_dev_poll value.

5. Modify <cman> timeouts to be sure to account for speed of quorum disk.

6. Configure at least two fencing devices, with a direct power fencing option if possible for the fastest fencing. Be sure to use the passwd_script parameter rather than a plain text password in your cluster.conf file.

7. Simplify the cluster.conf to minimize errors due to complexity. Simple clusters consistently work best and are easiest to maintain.

8. The <ip> stanza is used to configure the virtual IP address resource. It is recommended that monitor_link="1". Be sure the application, all interfaces, any ECP clients, and so on use the virtual IP address as configured here to connect to the Caché resource.

9. Test loss of private and public network connectivity to be sure no split-brain occurs.

10. Add all the <lvm> and <fs> stanzas to the <service> stanza in the cluster.conf file required for Caché and the application. The following are recommended for the <fs/> stanza(s):

    force_fsck = "1"
    force_unmount = "1"
    self_fence = "1"
    

    

Installing the cache.sh Script

To install the Caché control script:

1. Locate the files in dev/cache/HAcluster/RedHat/ under the Caché installation directory.

2. Copy or place the cache.sh file in /usr/share/cluster and make sure the permissions, owner and group are identical to the other files in that directory.

Patching the cluster.rng Script

Patches are provided for the cluster.rng file as supplied with RHEL 5.5, 5.6, 5.7, or 5.8 or RHEL 6.2, 6.3, or 6.4. The patch simply adds the definition of the <cache /> resource to the list of cluster resources.

When using RHEL 6, patching is required since a cluster.conf file that includes the <cache /> resource cannot validate without the patched cluster.rng in place. If your system is running a version newer than RHEL 6.4, contact the InterSystems Worldwide Response Center (WRC)Opens in a new tab for information about options for patching your cluster.rng.

The cluster.rng file is typically found in either the /usr/share/cluster or the /var/lib/cluster directory. Be sure to patch the actual source file (and not simply a linked file).

1. Place the appropriate patch file (either cluster.rng.patch or cluster.rng.patchRHEL6) on the system in /tmp/.

2. Change to the appropriate directory (/var/lib/cluster or /usr/share/cluster) and use the following command to patch the existing cluster.rng file:

   patch -b cluster.rng /tmp/cluster.rng.patchRHEL6

   Note:

   cluster.rng.patch patches unaltered RHEL 5.5, 5.6, 5.7, or 5.8 cluster.rng files. cluster.rng.patchRHEL6 patches unaltered RHEL 6.2, 6.3, or 6.4 cluster.rng files.

Using the <cache /> Resource

The following is a sample resource set up to control an instance named cacheprod:

<cache  __enforce_timeouts="1" cleanstop="1" name="cacheprod">
        <action name="status" interval="30s"/>
        <action name="start" timeout="10m"/>
        <action name="stop" timeout="5m"/>
</cache>

In this sample, a service starts the cacheprod instance of Caché and waits up to 10 minutes for the resource start to return. The service is marked as failed if the start does not complete in 10 minutes. After successfully starting all the resources, status checks occur every 30 seconds.

If a clean service stop is not configured (cleanstop=”0”), the cacheprod instance is stopped with a ccontrol force cacheprod quietly. This results in a fast stop and protects against clean stop hangs due to unavailable disk or other Linux resources, ensuring a fast transition to the other node during faults that do not crash the system but result in a node transition. If cleanstop=”0” is used, system managers should manually stop Caché (ccontrol stop) before any manual service stop or service relocations. This ensures a clean stop during routine maintenance work.

In this example a clean service stop (cleanstop="1") is configured, the cacheprod instance is stopped with a ccontrol stop cacheprod quietly. If that fails or generates an error, a ccontrol force cacheprod quietly is attempted. Note though, that a hung ccontrol stop hangs forever.

In this example, however, with __enforce_timeouts=”1”, the stop must complete in five minutes. A stop that does not return within that timeout period causes the whole service to be marked as failed.

Note the following when using the <cache /> resource:

1. The <cache /> resource takes the following parameters:

   • name: The name of the Caché instance being controlled by the service.

   • __enforce_timeouts: A generic rgmanager setting; as a general rule, it is recommended that you set this = “1” to ensure that a start or stop of the service does not hang indefinitely.

   • cleanstop: If false (= “0”) a cluster service stop or relocate immediately forces Caché down rather than attempting a controlled stop of Caché; while the stop is faster, a force of Caché may lengthen startup time as transactions roll back, and so on

2. Action timeouts are honored only if __enforce_timeouts="1". Generic rgmanger start and stop timeouts should be configured based on application needs, as follows:

   • <action name="start" timeout="10m"/>: the start timeout should be long enough to allow the replay of the CACHE.WIJ and journal files, but not so long that severe cluster problem notifications are delayed while waiting for a start that will never complete.

   • <action name="stop" timeout="5m"/>: the stop timeout comes into play only if the service is stopped (as opposed to a fencing event that cuts power). This can happen at boot time or as part of cluster reorganization or during a manual service stop.

     Note:

     If the start or stop timeout is reached the service is marked as FAILED and manual intervention is required to reenable it.

   • <action name="status" interval="30s"/>: As a general rule, set the status check interval to 30 seconds; more frequent status checks use CPU and may affect cluster responsiveness.

3. The status check uses ccontrol qlist [instance] to see if the instance is down or running. If it is in either of those states, the status check passes and no action is taken; this lets you stop Caché manually, if necessary, without stopping the service and without accidentally triggering a failover event.

4. The <cache /> resource is always started after the <lvm>, <fs> and <ip> resources, and stopped before those resources; this is without regard to placement within the <service> stanza of the cluster.conf file. For more complex cluster configurations, parent and child relationships may need to be configured, but a typical two-node cluster does not require the added complexity.

Installing a Single Instance of Caché

To install a single instance of Caché, use the following procedure.

1. Bring the service that contains all the disk and IP resources online. This mounts the entire set of disks required for Caché and the application to run (for installation, data files, journal files, and any other disk required for the application in use) on one node.

   1. Check the file and directory ownerships and permissions on all mount points and subdirectories.

   2. Prepare to install Caché by reviewing the “Installing Caché on UNIX® and Linux” chapter of the Caché Installation Guide.

2. Create a link from /usr/local/etc/cachesys to the shared disk. This forces the Caché registry and all supporting files to be stored on the shared disk resource you have configured as part of the service group.

   A good choice is to use a ./usr/local/etc/cachesys/ subdirectory under your installation directory.

   For example, assuming Caché is to be installed in /cacheprod/cachesys/, specify the following:

   mkdir –p /cacheprod/cachesys/usr/local/etc/cachesys
   mkdir –p /usr/local/etc/
   ln –s /cacheprod/cachesys/usr/local/etc/cachesys /usr/local/etc/cachesys
   
   

   

3. Run Caché cinstall on the node with the mounted disks. Be sure the users and groups (either default or custom) are available on all nodes in the cluster, and that they all have the same UIDs and GIDs.

4. Stop Caché and relocate the service to the other node. Note that there is no Caché resource configured, so the service does not control Caché yet.

5. On the second node in the cluster, create the link in /usr/local/etc/ and the links in /usr/bin for ccontrol and csession:

   mkdir –p /usr/local/etc/
   ln –s /cacheprod/cachesys/usr/local/etc/cachesys/ /usr/local/etc/cachesys/
   ln –s /usr/local/etc/cachesys/ccontrol /usr/bin/ccontrol
   ln –s /usr/local/etc/cachesys/csession /usr/bin/csession
   
   

   

6. Add the Caché resource to the cluster.conf file and configure according to the needs of your application. At minimum, add the following line to the <service> stanza:

   <cache name=”instance” __enforce_timeouts=”[1/0]” cleanstop=”[1/0]” />
   

   

   where

   • instance is the name of the instance specified during Caché cinstall (step 3)

   • __enforce_timeouts can be specified as 1 or 0, with 1 recommended

   • cleanstop=”1” indicates that cache.sh stop should issue a ccontrol stop while cleanstop=”0” indicates that cache.sh stop should issue a ccontrol force

   In addition, you can optionally specify any or all of the following stanzas within the <cache> stanza (see Installing and Using the <cache /> Resource).

   <action start timeout=10m/>
   <action stop timeout=5m/>
   <action status interval=30s/>
   

   

7. If editing the cluster.conf file directly:

   1. Update the version of the cluster.conf file

   2. Validate cluster.conf (ccs_config_validate)

   3. Distribute cluster.conf to all nodes (cman_tool version –r)

8. Restart the rgmanager daemon (service rgmanager restart) on all nodes and test.

Installing Multiple Instances of Caché

To install multiple instances of Caché, use the following procedure.

1. Bring the service that contains all the disk and IP resources online. This mounts the entire set of disks required for Caché and the application to run (for installation, data files, journal files, and any other disk required for the application in use) on one node.

   1. Check the file and directory ownerships and permissions on all mount points and subdirectories.

   2. Prepare to install Caché by reviewing the “Installing Caché on UNIX® and Linux” chapter of the Caché Installation Guide.

2. Run Caché cinstall on the node with the mounted disks. Be sure the users and groups (either default or custom) are available on all nodes in the cluster, and that they all have the same UIDs and GIDs.

3. The /usr/local/etc/cachesys directory and all its files must be available to all nodes at all times. To enable this, after Caché is installed on the first node, copy /usr/local/etc/cachesys to each node in the cluster. The following method preserves symbolic links during the copy process:

   cd /usr/local
   rsync -av -e ssh etc root@node2:/usr/local/
   

   

   If rsync is not available, tar or other options may exist.

4. Verify that ownerships and permissions on the cachesys directory and its files are identical on all nodes

   Note:

   In the future, keep the Caché registries on all nodes in sync using ccontrol create or ccontrol update or by copying the directory again; for example:

   ccontrol create CSHAD directory=/myshadow/ versionid=2013.1.475
   

   

5. Stop Caché and relocate the service to the other node. Note that there is no Caché resource configured, so the service does not control Caché yet.

6. On the second node in the cluster, create the links in /usr/bin for ccontrol and csession:

   ln –s /usr/local/etc/cachesys/ccontrol /usr/bin/ccontrol
   ln –s /usr/local/etc/cachesys/csession /usr/bin/csession
   
   

   

7. Stop Caché and disable the service.

8. Add the Caché resource to the cluster.conf file and configure according to the needs of your application. At minimum, add the following line to the <service> stanza:

   <cache name=”instance” __enforce_timeouts=”[1/0]” cleanstop=”[1/0]” />
   

   

   where

   • instance is the name of the instance specified during Caché cinstall (step 3)

   • __enforce_timeouts can be specified as 1 or 0, with 1 recommended

   • cleanstop=”1” indicates that cache.sh stop should issue a ccontrol stop while cleanstop=”0” indicates that cache.sh stop should issue a ccontrol force

   In addition, you can optionally specify any or all of the following stanzas within the <cache> stanza (see Installing and Using the <cache /> Resource).

   <action start timeout=10m/>
   <action stop timeout=5m/>
   <action status interval=30s/>
   

   

9. If editing the cluster.conf file directly:

   1. Update the version of the cluster.conf file

   2. Validate cluster.conf (ccs_config_validate)

   3. Distribute cluster.conf to all nodes (cman_tool version –r).

10. Restart the rgmanager daemon (service rgmanager restart) on all nodes and test.

When you are ready to install a second instance of Caché within the same cluster, follow these additional steps:

1. Configure the cluster.conf file to add the <ip>, <fs>, and <lvm> resources associated with the second instance of Caché.

2. Bring the service online so the disks are mounted on one of the nodes.

3. Be sure the users and groups to be associated with this new instance are created and synchronized between nodes.

4. On the node with the mounted disk, run ccinstall following the procedures outlined in the "Installing Caché on UNIX® and Linux" chapter of the Caché Installation Guide.

5. Stop Caché.

6. Synchronize the Caché registry using the following steps:

   1. On the install node run

      ccontrol list
      

      

   2. Record the instance name, version ID and installation directory of the instance you just installed.

   3. On the other node, run the following command to create the registry entry, using the information you recorded from the recently installed instance:

      ccontrol create inst_name versionid=vers_ID directory=inst_dir
      

      

7. Edit cluster.conf to add the <cache /> resource to the service controlling this instance.

8. Start the service and verify that Caché starts.

Failure Testing

Typical full scale testing must go beyond a controlled service relocation. While service relocation testing is necessary to validate that the cluster.conf file and the service resource scripts are all functioning properly, you should also test responses to simulated failures. Be sure to test failures such as:

• Loss of public and private network connectivity to the active node

• Loss of disk connectivity

• Hard crash of active node

Testing with an application load builds confidence that the application will recover in the event of actual failure.

Try to test with a heavy disk write load. During heavy disk writes the database is at its most vulnerable.

Caché handles all recovery automatically using its CACHE.WIJ and journal files but testing a crash during an active disk write ensures that all file system and disk devices are properly failing over.

Software and Firmware Updates

Keep software patches and firmware revisions up to date. Avoid known problems by adhering to a patch and update schedule.

Monitor Logs

In RHEL 5, the /var/log/messages file is the default location for cluster-related messages. Keep an eye on this log for issues or cluster problems. The <cache /> resource also logs messages during start and stop events in the system log at log level 5.

In RHEL 6, the messages are also in various logs in the /var/log/clusters/ directory. Use the <log> stanza in the cluster.conf file to increase logging level during testing and debugging.

Use the Caché console log, the Caché Monitor and the Caché System Monitor to be alerted to problems with the database that may not be caught by the cluster software. (See the chapters “Monitoring Caché Using the Management Portal”, “Using the Caché Monitor” and “Using the Caché System Monitor” in the Caché Monitoring Guide for information about these tools.)