UNIX Client Confusion

[GregE]

Solaris 10.

I have the following TSM client scenario...
Server09-CAPS - local
Server10-CAPS - local
Server00-PRISM - virtual IP

The local clients are behind a load balancer, which directs traffic to the primary when users access PRISM.company.com. Currently Server09-CAPS is primary. (Server10-CAPS knows nothing of those SAN-based filesystems, until failover which is manual)

I have setup dsm.opt and dsm_PRISM.opt as well as two stanzas of dsm.sys.
In the stanza Server09 (and Server10), I have specified TCPClientAddress of Server09.company.com )or Server10.company.com)

In the stanza Server00-PRISM, I have specified TCPClientAddress of PRISM.company.com.

All SchedModes are PROMPTED. At TSM console, node Server00-PRISM currently shows:
TCP/IP Name: Server10
TCP/IP Address: <IP address of Server10>

When I run a backup of Server00-PRISM it fails with return code 12, because it's looking to filesystems that do not exist, because Server09 is the current primary server, not Server10. Server10 does knows nothing about those filesystems.

Why is TSM not using the TCPClientAddress I am specifying? We recently moved to these new servers. The old servers were configured exactly the same, yet they backup the virtual node just fine, even though that TSM node's IP Name and IP Address values are the server NOT currently connected to those filesystems.

Any ideas?

 

[sgabriel62]

Have you tried by IP rather and FQDN? Any Domain statements for the associated filesystems?

 

[GregE]

Haven't tested with just IP but it's a pingable IP. It tries the IP of either backend node depending on the IP it's storing for that node. If I've last accessed TSM from current primary node Server09, then the Server00-PRISM scheduled backup runs correctly. If I've last accessed TSM from Server10 then there are errors in Server10's error log when the schedule kicks off showing errors because the domains listed do not exist, and in Server10's visibility, that's correct they don't exist currently. They exist in Server09's visibility.

dsm_PRISM.opt...

SErvername Server00-PRISM
Domain /opt/progress /var/progress/appdata /var/progress/backups /var/progress/dump

dsm.sys stanza ...

SErvername Server00-prism
Nodename Server00-prism
COMMmethod TCPip
TCPPort 1502
tcpadminport 1503
httpport 1582
TCPServeraddress tsm01.company.com
TCPClientaddress prism.company.com
SCHEDMODE prompted
PASSWORDACCESS GENERATE
PASSWORDDIR /opt/tivoli/tsm/client/ba/bin/PRISM
INCLExcl /opt/tivoli/tsm/client/ba/bin/PRISM/inclexcl.PRISM
ERRORLOGName /tools/strg/TSM/logs/dsm_PRISM_error.log
SCHEDLOGName /tools/strg/TSM/logs/dsm_PRISM_sched.log
SCHEDLOGRetention 30
ERRORLOGRetention 30
MANAGEDSERVICES WEBCLIENT SCHEDULE

 

[sgabriel62]

From what Im reading that makes sense. Perhaps your domain statements in your opt files is incorrect. I would gather than 09 and 10 have their own set and the virtual has the filespace in question - so no matter where they are mounted - the virtual node backs it up

 

[GregE]

You're understanding it correctly. That's what's supposed to happen. Server09 and Server10 have include-exclude lists that exclude those filesystems covered by Server00-PRISM's opt file, so there is no duplication on those 4 filesystems.

 

[sgabriel62]

So what did I miss other than you're going to try straight IP. I want to say you do not necessarily need the tcpclientaddress on the local nodes. Only if aliases or used or behind a firewall.

Not to sound ignorant but is Solaris zoned correctly?