Strange startup behavior after installing SGLX_00096

[Joerg Schulz]

Hi,

after installing the ServiceGuard patch starting of the package failed. As per patch documentation a new package control script was created and applied.
In the package logfile I found, that the function "customer_defined_run_cmds" is performed after start of the services and not before as described.
Is this a known behavior? How can I perform programs before the services are started?

Thanks,
Joerg

[Steven E. Protter]

Shalom Joerg,

There is a prescribed upgrade procedure.

Did you use it and install pre-requisite OS patches?

SEP

[Joerg Schulz]

Hello Steven,

in the patch description I can't find any dependencies (all: None; OS: SLES9)

And yes, I followed the installation instructions and the special instructions too.

I didn't get any error during installation, applying the new configuration and starting the node. Rejoining into the cluster without error.

Thanks,
Joerg

[Serviceguard for Linux]

Jeorg,

Put in a call to support.

I'm pretty sure there was no change in the package control script that would affect the startup.

If you saved the original package control script then you can post both the old and the new. If you don't want to post them then you may need to give them to support for help.

By the way, there is a Serviceguard for Linux category in teh ITRC.

http://forums1.itrc.hp.com/service/forums/categoryhome.do?categoryId=555

[Joerg Schulz]

Hello,

sorry about using the wrong category. I wanted to post in the serviceguard category, but I had made a mistake.

Attached you will find the old and the new control script. I have transfered the configuration from the old version to the new one without changes. The output of diff shows only differences of the template changes.

Thanks,
Joerg

[Serviceguard for Linux · Accepted solution]

If you look in the new script you can see on line 1451 (new file) that the customer defined run commands is called and then the services are started. The difference is that in the old version, there are other functions in between.

Question: Is this causing a problem, or do you just need to understand it?

Speculation: Maybe the timing difference between the run commands and services start is causing problems - or at least causing what you are seeing in the logs. Remember then when you do the run_commands, some commands may "fork" and exit before they have really done much to start.

Comment: generally, you should be able to use your old package control scripts if that helps.

[Joerg Schulz]

I see. So the old configuration is only working, because the function "verify_ha_server" consumes enough time, that the commands in customer_defined_run_cmds could be finished.

I think the best way is to modify customer_defined_run_cmds, in order to guarantee that all commands are finished if the function returns.

Thank you very much for your help.

[Joerg Schulz]

Hi,

today I tried to modify the content of the function "customer_defined_run_cmds" (the cluster in still in production, so I have to wait for a production break). All I tried didn't work. At last I removed the critical commands from the function and put them into the service script. Into the function "customer_defined_run_cmds" I added an echo command as first command. The output of this echo command appears into the package logfile: after startup of the services!

Any suggestions?