unable to umount FS or kill-9 ARCH0 process

[Pudovchenko]

There is HA cluser under HP ServiceGuard 11.11 PA-RISC. Afrer the disk array was accidentally disconnected in one node, the database was in down state. But the Oracle ARCH0 process was in memory and we cannot kill it with kill-9
command.
Kill-9 issued under Oracle & Root accounts, but whithout usefull result.
Therefore we cannot umount arch file system from one node, so that connect it
to second node. There no force umount in HPUX.
Therefore, we cannot mount arch file system in second node,and cannot start DB on second node.

Anybody knows what to do ?
I need umount the arch FS or kill ARCH0 process.
It is beyond my understanding, I need more comprehension of OS internals.

[Arunvijai_4]

Hi,

Try to use # fuser -ku

-Arun

[Peter Godron]

Hi,
what is the parent process id of the arch0 (third column)?
ps -ef | grep -i arch0

If it is 0, it will be very difficult to remove and normally requires reboot.

You could try and start, without opening the DB on second node and then disable archive.

[Cheryl Griffin]

kill -9 is not recommended and probably caused the problem. kill -15 would have been the recommended command but at this point it may be too late. A reboot will clean this up.

[Steven E. Protter]

Shalom,

Cheryl is right. I've caused this problem in the lab with kill -9

The situation is beyond hope of correction without a reboot.

SEP

[Pudovchenko]

After accidentally loosing connectionn to disk array in one node the ServiceGuard stops cluster package (Oracle DB in our case) in this node. But it cannot finish this package because ora_arch0 "keeps" the file system.
Is only reboot helps to us ?

[Bill Hassell]

A reboot is required. The ARCH0 process is hung waiting on I/O that will never complete. The kill command only signals the process to stop -- it performs no other action on the process. So if the process is waiting on on I/O, the signal is not meaningful. The I/O is most likely in a corrupt state so even though the array is reconnected, the I/O has been forgotten by HP-UX and the program can never be killed.

[Alexander_118]

This process it's ORA_ARC0.
As I understand, this process dumps Oracle logs from cache to the file. But Log file is located on he disk array.
If I break FC cable from server to the disk array, cluster start down database and can't finish down because this process (ora_arc0) try write log to the file and can't.
If only one way - reboot server, it's very badly. Only one reason to switch from one node to another - complectly break server.

[Pudovchenko]

Thanks to all !