[Info-vax] Cluster node fails to boot from shadow-set member

[Michael Moroney]

Richard B <richard.brock57 at gmail.com> writes:

>HI all,

>I'm having a serious issue trying to boot a node of a three-node cluster.  =
>The cluster is comprised of 2x rx2600 i2 machines and 1x rx2660.  Each node=
> is running VMS v8.4 and has internal-only storage.  Each node has its own =
>SYSTEM disk but the 2 rx2600 i2's system disk is shadowed between the two n=
>odes:

>Node 1:  DSA0:  $1$DKA0:
>Node 2: DSA0:  $2$DKA0:

>Node 2 boots fine and mounts the virtual device DSA0: with its physical uni=
>t at $2$DKA0:

>It then forms a cluster just fine with the third, rx2660, node.

>When I try to boot Node 1 I get a bugcheck during boot.  The output is:

>%CNXMAN,  Now a VMScluster member -- system RYEIP1
>%SHADOW-F-/-CONFIGSCAN, enabled automatic disk serving

>**** OpenVMS I64 Operating System V8.4     - BUGCHECK ****

A setup like that with two nodes with a system disk shadowset with non-shared
members will not work. When a node is down, its disk gets removed from the shadowset
(as seen by the other node) and when it tries to return, its system disk is out of
date, and shadowing won't accept it.

You really need shared storage of some sort.

I have heard of a *hack* that sort of allows this.  Sort of.  Each system MSCP
serves its drive to the other. Each system is set up to boot over the network from
the other node's disk. As long as the other node is up (and the quorum issue has
been dealt with), it will boot, add its own disk as a copy target and once the
shadow copy completes (can take a while!), all is well.

Much better to add some sort of shared storage (fibrechannel).

Obviously, this will not work if both nodes are down. Each will be waiting for the
other, and nothing will happen. One of the nodes will have to be manually booted 
from its own disk.