Problem in MSCS fail over

[Former Member]

Hi,

Sorry for posting this here but I have a problem in two MSCS clusters 2003 and got very confused. First I want to tell the story and the original problem and then will need opinions and experiences from any one who can share something interesting.

We installed a NW 7.0 in a Windows 2003 Cluster following the Installation guide from SAP and we probably made a mistake but need to read it here because SAP has not given us a convincing answer so really need your help. As many of you know and we discovered, installation for 7.0 is very different from 6.40 when installing in MSCS because SPOF's are installed now in shared resources and Applications servers are installed locally.

The problem that started all this is that when doing fail over from Node A to Node B everything works fine, it takes around 20 seconds to move and then you can access ABAP and Java with no problem. BUT when returning fail over from Node B to node A then after all resources moved then dispatcher fails and the error we found in the logs is that

DP_FATAL_ERROR => DpMsAttach: incorrect value for SAPLOCALHOST

We oppened an OSS to SAP because this is a Productive system that is not yet live in production and after three weeks that they asked to take a look at SAP notes about vitual names and their uses in MSCS. After that time another SAP processor from development team entered the system and told us that installing application servers using SAP group virtual name is not supported so we should reinstall CI and DI so everything could be in working order. His explanation is that when coming from Node B to node A as CI is not up and the virtual hostname is not in node A then dispatcher cannot start. He changed to physical host names and the fail over worked fine.

But the new processor is searching all over again and I now think that they don´t agree between them and from some other opinions I got from different more experienced Basis, Virtual name is not the problem.

So I really need your opinions and experience so I can be more clear about the process, in installation guide they tell to install CI and DI in local disks and do not mention anything about virtual names so they probably assumes that we should use the local physical disks and there our confussion came.

Probably somebody else did the same as we did and using SAP group virtual name is a NO NO procedure, or somebody did it and it worked, or there is an easier solution than reinstalling CI and DI, or probably reinstalling won´t harm anything in JAVA...

Your experiences are pretty usefull to us so please go ahead and write it here.

Thanks,

[Former Member · Accepted solution]

Hi dude!

It is not clear for me... did it fail the ASCS dispatcher, or just one of the dialog application servers? Also, the developer said that it is not a good idea to use the virtual name of the SAP group... where did you use it and you should not?

Anyway if the only failing is the dispatcher, but the service is starting up properly, I would firstly check the start and instance profiles of the instance failing, paying special attention to the SAPLOCALHOST (I guess it should not be the FQDN) and the names of these profiles too.

Cheers!

- J.Garcia

[Former Member · Accepted solution]

Hello,

J Gacia's comment seem to be sensible. I would look at ASCS hidden shares and corresponding permissions. Please check if "saploc" and "SAPMNT" share and security permissions are included for local administrators.

Regards,

Krishna

[Former Member · Accepted solution]

Hi,

It is just faining one of the application servers, developer wrote that when installing CI, Local hostname is written in Database and SDM. We will have to do a Homogeneous system copy to change the name.

The problem is that I used Virtual SAP group name in CI and DI application servers, in SCS and ASCS we used Virtual hostnames and it is OK according to SAP developer.

The Start and instance profiles were checked and everything was fine, just the dispatcher from CI is having problems when comming from Node B to Node A.

Regards

[Former Member · Accepted solution]

Hi,

We already checked shares and permissions and everything was fine. The problem is that SAP Group virtual name should not be used to install CI .

SAP Developer already checked shares and permissions also and it was OK,

Regards,

[Former Member]

Hi,

We did what SAP told us about to deinstall DI and CI (and ERS) and reinstall it again using local hostname doing it as a system copy but not deinstalling DB. We just used export preparation for CI and in target we used CI installation and it bringed back all JAVA configurations and objects.

Now cluster works fine so if somebody has any doubt about if using SAP group virtual name for installation is possible the answer is NO, it cannot be used and is mandatory to use Local hostname OR virtual hostname not related to the virtual SAP Group hostname.

Thanks to those that tried to help.

[Former Member]

Dear AC

There might be problem with "sapstartsrv.exe" file.

Sometimes this file gets corrupted while failover or failback.

Just check the patch level of your kernel if it old then upgrade your kernel patch and try.

This might fix your problem.

Do let us know the status.

Regards

Khalid