ADSM/TSM & ServiceGuard

[Justin Willoughby]

I have an problem whenever I do a fail-over from one node to another and then fail back to the original node ADSM thinks all (or most) of the files on the filesystem that failed over changed and then tries to back them up. This takes a very long time to backup all the files on the filesystem that is failed over with the package.

Is anyone using ADSM (now TSM) to backup files on a MC/ServiceGuard cluster?

If so, has anyone seen this happen before?

- Justin

[Justin Willoughby]

I see Kurt Beyers has the same problem as I do. It does not look like he found a solution.

http://forums.itrc.hp.com/cm/QuestionAnswer/1,1150,0x9a9dee3e323bd5118fef0090279cd0f9,00.html

Has anyone found a workaround for this problem?

- Justin

[Kurt Beyers. · Accepted solution]

For backing up Service Guard clusters you can install one TSM scheduler in each physical node and one TSM scheduler for each package. The package's scheduler runs with the same nodename in both nodes so when you perform an
incremental backup you only back up the files that have changed since the last backup. A truly cluster aware backup and restore ;-)

dsm.sys:
-------------
Servername tsm
tcpserveraddress tsmserver
tcpport 1500
passwordaccess generate
schedmode prompted
domain /internal_filesystems
[...]

Servername tsm_packA

tcpserveraddress tsmserver
tcpport 1500
passwordaccess generate
schedmode prompted

nodename clustername_packA
domain /shared_disk_filesystems_packageA


dsm_packA.opt:
-----------------------
Servername tsm_packA

To start the scheduler as cluster service:
dsmc schedule -optfile=/opt/tivoli/tsm/client/ba/bin/dsm_packA.opt
If you don't want to syncronice dsm.sys/opt files place it on shared disks.

When you want to perform a restore:
dsmc restore -optfile=/opt/tivoli/tsm/client/ba/bin/dsm_packA.opt

Remember to register the cluster node (cluster_packA) in the TSM Server and
set the password in the client (dsmc
--optfile=/opt/tivoli/tsm/client/ba/bin/dsm_packA.opt)


Kurt

[Justin Willoughby]

Kurt,

Does this setup you suggest work when backing up the same node. That is, my problem is I fail-over to the 2nd node and then back to the first between backups. The ADSM/TSM backup takes a really really long time on the 1st node even though the same files are there as before. The LV was mounted on the 2nd node but were failed back to the 1st node before the backup kicked off again.

Will your sugestion work for this situation?

Also I would assume you need to use exclude statements in the consign files so the package filesystem is not backed up twice.

[Kurt Beyers.]

Since you use the IP address of the package, you're incremental backups are indeed with reference to the package itself. So after a failover, the backup is incremental with respect to when the package was running on the first node. Make sure to include the domain option in the dsm.sys file for your package where you specify which file systems the package consists of. The same goes for your node itself where you have to specify the local file systems that belong to the node. All the incremental backups are with respect to the ones specified in the node option. I did a few tests on a system before it went into production and everything went smooth ;-)

[Justin Willoughby]

Kurt,

I am just wondering about some thing in your example.

You have:

Servername tsm
[...]
Servername tsm_packA

Should this 2nd Servername not point to tsm. I thought that Servername is the adsm/tsm server? We have one adsm server running on an AIX box. The HP/UX box has the adsm client running on it. Do the Servername only look at the part before the "_" for the actual adsm server?

Thanks again,

- Justin

[Kurt Beyers.]

You have to create one dsm.opt (client user option file) for each node and each package. This dsm.opt file contains nothing but:

Servername tsm_packA

for example for the package.

When you launch the scheduler or the backup gui with this specific option file, it will look in the client system option file (dsm.sys) for the entry tsm_packA:

...
Servername tsm_packA

tcpserveraddress IP_tsmserver
tcpport 1500
passwordaccess generate
schedmode prompted
nodename clustername_packA
domain /shared_disk_filesystems_packageA
...

The 'real' TSM server there is specified by the tcpserveraddress and it's IP address (your AIX box). With the option domain you specify the filesystems the package consists of etcetera. It is a bit confusing, but TSM likes to get things complicated.

Summarizing: tsm_packA is a virtual name you use to specify your package and the real TSM server is specified with it's IP address in the tcpserveraddress.

[Justin Willoughby]

Thanks Kurt,

I think I am starting to figure this out... haha

Here is my orginal dsm.opt:

===============
SErvername adsm-host
===============

And my orginal dsm.sys:

===============
SErvername adsm-host
COMMmethod TCPip
TCPPort PortHere
TCPServeraddress IPAddressHere
* compression on
passwordaccess generate
* include /opt/caitdm/.../* standard_14
* include /opt/syb/.../* standard_14

include /.../* standard_14
schedlogname /usr/adsm/dsmsched.log
schedlogretention 3
errorlogname /usr/adsm/dsmerror.log
errorlogretention 7
preschedulecmd "/usr/adsm/serverdown.cmd"
postschedulecmd "/usr/adsm/serverup.cmd"
===============

So I should create a new opt file such as dsm_caipkg.opt that looks like this:
===============
SErvername adsm-host_caipkg
===============

And my new dsm.sys should look like:
===============
SErvername adsm-host
COMMmethod TCPip
TCPPort PortHere
TCPServeraddress IPAddressHere
passwordaccess generate
domain /
exclude /opt/caitdm
* include /.../* standard_14
schedlogname /usr/adsm/dsmsched.log
schedlogretention 3
errorlogname /usr/adsm/dsmerror.log
errorlogretention 7

SErvername adsm-host_caipkg
COMMmethod TCPip
TCPPort PortHere
TCPServeraddress IPAddressHere
passwordaccess generate
nodename caitdm_caipkg
domain /opt/caitdm

schedlogname /usr/adsm/dsmsched_caipkg.log
schedlogretention 3
errorlogname /usr/adsm/dsmerror_caipkg.log
errorlogretention 7

preschedulecmd "/usr/adsm/serverdown.cmd"
postschedulecmd "/usr/adsm/serverup.cmd"
===============

Thanks again for all your help!

- Justin

[Kurt Beyers.]

Thats correct. One dsm.opt file for each node or package and one global dsm.sys file.

[Justin Willoughby]

Great!

In your example you did not have an exclude but I am pretty sure you need to exclude your domain so it does not get backed up twice right? Is there anything in my configuration above I should change?

Also did you modify your package script so it starts the 'dsmc schedule =optfile=" so it is only running node the package is running on?

You would also need to modify the startup and shutdown of the package to do this correct?

Thanks again Kurt,

- Justin