topic Re: How fail save are cluster logicals in Operating System - OpenVMS

How fail save are cluster logicals

Hans Adriaanse — Mon, 30 Jan 2006 04:30:03 GMT

I am using cluster logicals to create a small clusterwide database. What happens if during the creation of a logical the node crashes. Is the logical completely set or completely not set. Or is it possible that only half the data is in the logical ??
I am using OpenVMS 7.3-2.

Thanks in advance.

Re: How fail save are cluster logicals

Jan van den Ende — Mon, 30 Jan 2006 04:43:35 GMT

Hans,

I can hardby be considered an authority on this subject (maybe John G can jump in?) but I can only see one logical reasoning here:
IF the node on which the command is issued crashes during the processing of the command, then
a. the command has NOT yet reached the phase of communicating with the cluster. Effectively nothing has happened wrt the logical.
b. the cluster communication has been sent. Now if it is received by any other node, then that node ensures clusterwide consistency.

As said, I am not the expert, but it is hard to imagine anything not guaranteeing integrity in so important a piece of the OS EVER leaving Engeneering!

Proost.

Have one on me.

jpe

Re: How fail save are cluster logicals

Wim Van den Wyngaert — Mon, 30 Jan 2006 05:32:00 GMT

Hans,

Read this.

http://groups.google.be/group/comp.os.vms/browse_frm/thread/603cf1f8c7b097f6/f4016929246845f6?lnk=st&q=%22cluster+logical%22+vms&rnum=7&hl=en#f4016929246845f6

Wim

Re: How fail save are cluster logicals

Wim Van den Wyngaert — Mon, 30 Jan 2006 05:33:31 GMT

And note poster 12 concerning LNM$M_INTERLOCKED.

Wim

Re: How fail save are cluster logicals

Ian Miller. — Mon, 30 Jan 2006 08:07:39 GMT

In the referenced thread there is an entry by Rob Brooks which I think answers your question.

"In the second case (the simple defining of a single clusterwide logical via $CRELNM (which DCL DEFINE ultimately uses, of course)], the replication is synchronous; the $CRELNM call does not complete until all of the SCS messaging needed to deal with replication has completed. If a node is unable to replicate the logical name, it will die on its sword and crash. The point is to guarantee that the logical name is faithfully replicated across the cluster."

Re: How fail save are cluster logicals

Hans Adriaanse — Mon, 30 Jan 2006 08:11:15 GMT

Thank you all, I can now build my reliable small cluster database.