E9.2-1 v. existing CLUSTER_AUTHORIZE.DAT (IA64, et al.)

[sms]

Code: Select all

   I thought that I had an x86_64 system clustered (briefly) with my
main IA64 system, but now that I've installed (upgrade not allowed)
(the Community) E9.2-1, the systems seem not to notice each other.

   I suspect a mismatched CLUSTER_AUTHORIZATION PASSWORD, but I'm not
sure if I'm looking at the right stuff on the x86_64 system.

   Pre-x86_64, CLUSTER_AUTHORIZE.DAT was just a file, but now it's a
symlink to some exotic thing (which I'm afraid to touch):

V87 $ dire /acl sys$system:cluster_authorize.dat

Directory SYS$COMMON:[SYSEXE]

CLUSTER_AUTHORIZE.DAT;1 ->
 /SYS$MD23108222540A/VMS$COMMON/SYSEXE/CLUSTER_AUTHORIZE.DAT

   But I did rediscover my group number and password, and used SYSMAN
CONFIGURATION SET CLUSTER_AUTHORIZATION to (try to) ensure that everyone
matches.

   DUMP /BYTE /WIDTH=80 on SYS$SYSTEM:CLUSTER_AUTHORIZE.DAT shows the
same stuff on the IA64 system and the x86_64 system, but it's not clear
to me that this is what's actually used when the x86_64 system starts.

   SYS$LOADABLE_IMAGES:SYS$MEMORYDISK.DAT does mention
SYS$SYSTEM:CLUSTER_AUTHORIZE.DAT, but I seem to recall reading firm
instructions not to fiddle with SYS$UPDATE:SYS$MD.COM unless someone at
VSI told me to, so I'm afraid to wreck something.

   So, is DUMP SYS$SYSTEM:CLUSTER_AUTHORIZE.DAT good enough to ensure
compatibility on all systems, or do I need to do more on the x86_64
system?  (What?)

   Is this all documented someplace?  Or, is there some BOOTMGR
diagnostic which would reveal what it's actually using for these
parameters (PROGRESS?)?


   The V9.2 Release Notes doc says:

      2.1.15. OpenVMS Clusters on Virtual Machines

      VSI OpenVMS x86-64 V9.2 supports VirtualBox, KVM, and VMware
      virtual machines in OpenVMS clusters. However, shared disk access
      on virtual machines is not supported. Note that this means no
      shared data disks or cluster common system disks.

   Which sounds to me as if my VMware Fusion couldn't access the
QMAN$MASTER directory on a disk on the IA64 system.  So am I wasting my
time trying to cluster these things?  Or is E9.2-1 better?

[volkerhalle]

sms,

we have a report of a 3-node E9.2-1 cluster with separate system disks, see viewtopic.php?f=37&t=8626 - so the CLUSTER_AUTHORIZE.DAT mechanism must work.

If there is layer 2 network connectivity between your E9.2-1 system and your I64 systems, but the cluster group number and passwords do not match, you should see PEA0 errorlog messages on the running nodes in your cluster. $ SHOW ERROR should also increase for PEA0:

Does the contents of CLUSTER_AUTHORIZE.DAT change, if you locally change your cluster group number or password with SYSMAN ?

I'm assuming you know what you're doing and the E9.2-1 system boots, but does not join the cluster - right ? Did you wait long enough (let's say 5-10 minutes), to see, if there are any 'have connection to' messages ? You have also certainly checked console messages on the running nodes in your cluster.

Please consider to re-check your cluster system parameters on the E9.2-1 system.

You could try to force a crash on the E9.2-1 system (CTRL-P CTRL-P crash (y/n): y), boot with VAXCLUSTER=0 and have a look at the dump.

SDA> SHOW PORT/ADDR=PE_PDT
SDA> SHOW CLUSTER
SDA> SHOW CLUSTER/SCS
SDA> SHOW CONNECTION

However, shared disk access on virtual machines is not supported. Note that this means no shared data disks or cluster common system disks.

I'm pretty sure, that the above mean 'shared PHYSICAL disk access' ! In you case, the E9.2-1 system will happily use MSCP to access the Itanium disk, on which the QMAN$MASTER file lives.

Volker.