Serviceguard for Linux Support of RedHat 3 Update3

[Serviceguard for Linux]

Serviceguard for Linux is now supported on Redhat 3 Update 3.

WARNING:

There is a rare problem where Serviceguard for Linux will stop responding to commands from other systems. The command executed on that other system will hang.

Corrupted UDP Messages can cause a system call to wait indefinitely.

This is caused by a change in the Linux 2.6 kernel and also in Red Hat Enterprise Linux 3 Update 3.

This is triggered by certain, very rare, failures in a network environment that result in corrupted UDP messages. These are corupted in a way which is not detected by hardware. These corrupted messages, when combined with a change in UDP on the versions of Linux listed above, result in a hang of the system call, “gethostbyaddr”. Serviceguard for Linux version A.11.15.04, and later, use this call and can be affected. This system call uses UDP network messages and corrupting these messages in a specific way can trigger the problem.

Red Hat is expected to fix this with the next update for Red Hat Enterprise Linux 3.
SUSE is expected to fix this with the next ServicePack for SLES 9.

Since Serviceguard is not yet supported with SLES 9, only the workaround for Red Hat EL3 is described.

Work Around:

You can use Red Hat EL3 Update2. You can wait for the next update for Red Hat EL3 that has the permanent fix. Or you can minimize the need to DNS lookups by:

1. The file /etc/nsswitch.conf lists what order the various mechanisms are used to do look-ups. There should be a line in that file of the form:
hosts: file dns

2. Add entries for all nodes that may send commands to a Serviceguard node in the /etc/hosts file. This is only a problem when a Serviceguard node is sent a command and it checks which host sent it to then check the “rights” of the caller.

a. Update the /etc/hosts file of Serviceguard Linux node on a subnet so that all nodes on that subnet, that are running Serviceguard, are listed. We currently recommend that all of the nodes in a cluster list the other nodes of that cluster in the /etc/hosts file. This workaround expands that recommendation.

b. Since Serviceguard Manager may send commands to Serviceguard nodes, and these are checked using the same mechanisms, these nodes should also be listed in the /etc/hosts file. (Note that the /etc/hosts file on systems running Serviceguard Manager and not running Serviceguard do not need this change.)

The permanent fix will be provided by Red Hat. This is expected in the next Update at a date not yet determined.

NOTE: Other applications that use “gethostbyaddr”, or other systems calls using UDP messages in a similar fashion, can potentially be

[Steven E. Protter]

I have several RH ES 3.0 servers in my labs.

If you need testers to assist with Quality control I might be interested in assisting in this product.

Regards,

SEP

[Serviceguard for Linux]

Appreciate the offer. I think I figured out your email. If we can take you up on your offer, I'll contact you. :)

[Balaji N]

count me in if you need to do some testing. its some time since i played with service guard.
-balaji

[Steven E. Protter]

There is little hope of figuring out my email.

Contact:Dan Gazzaniga for my personnel email or fill out the form at http://www.isnamerica.com/isninquire.shtml

The mail form has nothing to do with your request, so ignore the questions, but I can post the link in forums without the spammers getting my email address.

SEP