Using Veritas Cluster Server for UNIX HA environment and J2EE

[Former Member]

For several years, we have been using the Veritas Cluster Server (VCS) to manange our HA environment. To simplify it:

It basically is laid out as two identical hardware enviornments. Host A at one site and Host B at another. There is also an EMC frame at each site with the respecitve host. Host A and B have physically seperate names, but there is a DNS alias that they are referenced by. This DNS is switched to point to whatever host is "live".

These hosts are connected via an EMC SRDF link that provides synchronous updates to the disks at the block level. So, when SAP is running on host A and the DB writes changes, the SRDF link makes the same change on host B before Oracle returns the transaction. We have approx. a 5ms average wait time for this, which hasn't impacted productive use at all.

When Host A fails, we can immediately fail the ENTIRE enviornment over to host B in under 15 minutes. We can then bring up SAP on host B with no loss of transactional data if it was a non_DB failure. If it was a DB failure and the changes were not rolled back, then we lose any transaction that was in the middle of a write at the time of failure.

So far, this has been our setup in an ABAP only environment. Now we're ready to take a NetWeaver 2004s system into Prod and are having questions around the config to make this work in the VCS setup we currently use.

Since the failover is a complete disk replication between the two sites, we have copies of profiles in the /sapmnt/SID/profile directory that have each PHYSICAL host's name in it. So, the profile directory would include SID_DVEBMGS00_hosta and a copy of that with 'hostb' in the end. The actual parameters in the profile use the alias for most of the settings.

This method won't work with J2EE. We want a similar failover environment where we can literally take the NW04s system down, failover the filesystems to host B and bring it back up. But since host B will have a physcially different name, sparks will fly.

Suggestions?

[Former Member · Accepted solution]

Hi Dave,

To implement a failover environment with SAP NetWeaver 2004s, see the guide "SAP Web Application Server in Switchover Environments (6.40)" available on the Service Marketplace at http://service.sap.com/ha -> Media Library -> Documentation -> Switchover. Even though it says 6.40, the concepts are applicable to 2004s.

Best Regards,

Matt

[Former Member · Accepted solution]

thanks Matt. I'll check that out. My biggest concern is in handling the host name aliasing. Our J2EE is pretty straightforward. Good to hear from you again!

[Former Member · Accepted solution]

Follow up question -

I am able to successfully fail over the ABAP portion of the ABAP+J2EE environment we have. This is done by altering the profile parameters in the DEFAULT profile to reflect the cluster alias rather than the actual host name.

But in the confitool of the J2EE, there are some fields that have the actual hostname hardcoded. These are in the DB, so they are host independent:

Global Disp. config>managers>ClusterManger:

ms.host = <host>

Global Disp. config>managers>LockingManger:

enq.profile.filename = /usr/sap/<SID>/SYS/profile/<SID>_DVEBMGS##_host

enqu.host = <host>

Global Server Config>managers>Cluster Manager:

ms.host = <host>

Global Server Config>managers>Locking Manager:

enq.profile.filename = /usr/sap/<SID>/SYS/profile/<SID>_DVEBMGS##_host

enqu.host = <host>

I suspect I can change the physical host names to the alias, but how can the J2EE know which profile to look at based on what physical host it's running on? ABAP knows by just reading the profile that has the real host name in it.