running cluster_config.com on existing system

[Karen Lee_3]

we are going to build a cluster out of two servers that were previously stand alone boxes.

I was playing around with cluster_config on a development Openvms server and managed to delete my directory - no real harm since we are rebuilding this server anyway.

My question is, do I have to run cluster_config on the first node of the cluster? My boss recommended I just change VAXCLUSTER to 2 and VOTES to 1 and just reboot. will that work? what's up with deleting the directory?

[Robert Gezelter]

Karen,

Since we do not have a log of the running of CLUSTER_CONFIG.COM, I would hazard a guess that a "REMOVE" was done, and the node-specific directory (which on a non-clustered system is generally SYS0) was "cleaned up". Oops.

A quick flip through CLUSTER_CONFIG.COM shows some things which are specifically done in the case when the node is not currently clustered: it says "either add to an existing cluster OR create a new cluster" [emphasis mine].

I don't have an unclustered machine to play with at this instant, but I would use CLUSTER_CONFIG.COM as described in the OpenVMS Cluster Systems Manual (available online in PDF at http://h71000.www7.hp.com/doc/731FINAL/DOCUMENTATION/PDF/OVMS_731_CLUSTER_SYSTEMS.PDF , it is also available on the the OpenVMS Documentation WWW site in HTML format).

In particular, Example 8-11 speaks to your question. A new cluster SHOULD have a explicitly selected Cluster GROUP and PASSWORD. Using the SYSMAN CONFIGURE SHOW CLUSTER_AUTHORIZATION command on a node properly configured for cluster operation will show a Cluster Group Number and other information. The same command on an unclustered system will give an error.

It is true that you can turn clustering on and off by hitting the VAXCLUSTER parameter, but in that case the data is already there.

BTW, I do suggest Group numbers be set to something other than the default. I have seen LAN connectivity surprises when two groups, unknown to each other, select the same group number. For that reason, "1" is generally a BAD choice.

I hope that the above is helpful.

- Bob Gezelter, http://www.rlgsc.com

[Andy Bustamante]

In addition to Bob's reply, you need to plan out the storage and cluster interface configuration. Assuming you're using ethernet for cluster communications, you also need to modify INTERCONNECT="NI" and set votes and expected votes. Do these systems share storage? Will one or the other server disk and or tape devices? Do the systems provide boot load services? Will your cluster have a quorum disk? CLUSTER_CONFIG.COM will step you through these questions, in addition to setting a cluster number/password as Bob mentions. You can set things manually as your boss says, but expect to set more than two parameters.

A hint which may or may not set off Bob's security sense, document the cluster number and password in a modparams.dat comment.

If these systems are going into production use, I'd also recommend a redundant interface for cluster traffic. This can be as simple as a cross over cable between two NICs. You don't want your cluster hanging because the network guys decided to upgrade the firmware on a switch "real quick during lunch."


Andy

[Peter Quodling]

Karen,

as bob points out, exact commands, or logs would be invaluable in diagnosing this. (Oh for the days of VAX cluster console, that did such things automagically).

Cluster_config.com is a very simple piece of DCL, and what's more it's quite easy to read, in order to determine what it is doing, (or has done), in order to modify the cluster status of a system.

read the code...

All will be revealed.

Peter Q

[Uwe Zessin]

If I remember correctly, you can use CLUSTER_CONFIG to turn a non-clustered node into the first cluster member. Then you use this node to add roots for the other member(s). The nice thing is that CLUSTER_CONFIG already knows a lot of questions to ask for the upcoming configuration.

About deleting [SYS0] ...
Isn't that a fantastic opportunity to test your backup/restore procedures ? ;-)

[Karen Lee_3]

thanks everyone for the good advice. this is really odd, I'm new to this particular site and when mgmt. started talking about clustering the ds20e and the es40, i assumed it had never been clustered. but when i do the command in sysman,


SYSMAN> config show cluster_auth
Node USPAX1: Cluster group number: 1
Multicast address: AB-00-04-01-01-01

this is supposedly an uncluster machine, but it appears that somewhere down the line, someone ran cluster config.

if this is the case, can I just change the sysgen params and reboot?

[Robert Gezelter]

Karen,

Yes, if everything else has been setup correctly (which is a VERY BIG if).

- Bob Gezelter, http://www.rlgsc.com