Dual Cluster Locks - A Good Idea?

[Mark DeBoer]

I recently lost a CL disk (see my previous post) at it looks like I will have to have some immediate unscheduled downtime on my two node cluster to fix it. It's obvious that the CL disk is a single point of failure.

I'm thinking of implementing a second CL disk in the same cluster. The thought being that if a single CL disk fails, the alternate CL disk will still be available for the nodes in the cluster. I haven't seen any examples of dual CL disks in a small cluster - only in campus failover situations have I seen this idea discussed.

Is this a good idea? Will this scenario work as I think it will - meaning that I could wait until my regularly scheduled downtime to fix the failed CL disk and the risk of SPOF is reduced?

[James R. Ferguson]

Hi, Mark:

The documentation ( http://docs.hp.com/hpux/onlinedocs/B7660-90001/B7660-90001.html ) I think you're referring to doesn't seem to offer much insight into the "value" of dual cluster locks in "non-campus" situations. (This perception on my part may simply reflect ignorance!). However, the ServiceGuard manual is very explicit in noting that a SINGLE cluster lock is preferred.

By way of attempting to offer some thinking, here:

A mirrored cluster lock disk may increase your level of high-availability.

Document #A4481870 noted a 2-node cluster where the secondary lock disk was experiencing I/O errors leading to a panic of BOTH nodes. The document pointed out that this is possible since the disk is a shared resource.

Importantly, to your question, "LVM Mirroring does not imply Fault Tolerant. It does insure that hard errors are handled (using BBDIR or by disabling the drive), but it can't necessarily handle everything."

An interesting "alternative" is suggested by document #S3100005869. EMC Symmetrix disk arrays can be used as Cluster Lock disks with
PowerPath. With PowerPath, it is possible to have a PVlink fail, but not know it, since
PowerPath effectively hides the failure from ServiceGuard.

...this, for what it's worth...

...JRF...

[Carlos Fernandez Riera]

SECOND_CLUSTER_LOCK_PV is recommended in two node scsi connected clusters.

From man clquerycl: SECOND_CLUSTER_LOCK_PV. These values are only recommended for the 2 node configuration using only internal disks, therefore possibility of both the first cluster lock pv failing at the same instant a node fails.

[Emil Velez]

It should not be used except in a campus cluster environment with 2 data centers and then what you should have is a 3rd center with a system or two to break a network failure tie. When you mirror with LVM it does not mirror the cluster lock information. It is only on that disk. If the cluster lock disk fails fix it prior to other failures.

[melvyn burnard]

The idea of a dual cluster lock disc is aimed at the campus cluster type of configuration.
In the cluster config you have, we do not generally recommend a dual cluster lock set up, as this could lead to what is known as "split brain syndrome".

If your cluster is running, then the missing cl disc will only cause a problem if you were to have one of the nodes fail, causing the remaining node to attempt to grab the cluster lock disc, fail, and therefore TOC itself.

Have you read the suggestion of my colleague Stephen (last response to your previous question) regarding the cminitlock utility?