trouble shooting a possible kernel param problem

[Tim Killinger]

I'm digging to help our application troubleshoot an issue on our HP 9000 server running UX 11.0 .... I have little to go on so far, but the error messages are like:

" semop() in SEM_op(SEM_GRAB) [errno = 28 : No space left on device]
12:57:26 [unknown-?]: SEM_op() Error. UNLOCK - not ours to unlock [8985] !
sm: Waiting for running system to stabilize...done.
12:57:27 [unknown-?]: StartWrite(): Error starting write [errno = 28 : No space left on device]
12:57:27 [unknown-?]: FATAL ERROR in Dictionary(): StartWrite(): Error starting write"


Does this ring ANY bells out there?

[Denver Osborn]

"erno 28" is a full filesystem. Check your filesystems and correct the prob. If the app requires additional space for the fs it's writing to, increase it.

Hope this helps,
-denver

[Tim Killinger]

I should have mentioned it, but we have no full file systems.... does errno = 28 come form UX?

We're trying to figure out if the message is complaining about a UX issue or perhaps it is coming from SunOne - our webserver software...

[A. Clay Stephenson · Accepted solution]

Eventhough the errno is ENOSPC (28), this is NOT a filesystem full error.

Now for a quick lesson on how to do this:
1) vi /usr/include/sys/errno.h and find the entry for 28. You will note that 28 corresponds to the 'ENOSPC' define. That define is important because all the system call errors use the define rather than the numerical value.

2) Now man 2 semop (the system call in question) and find which conditions can cause errno to be set to ENOSPC.


The answer is the limit of the number of processes requesting a SEM_UNDO would be exceeded thus you need to increase the kernel parameter semmnu.


Now, wasn't that easy? This is essentially the way you do it for any failing system call.

[Steven E. Protter]

Sounds like a failed semaphore operation.

What are the current settings of shmmax, shmseg and related parameters? Consider an increase if called for.

I'm wondering what kind of HP-UX server we're talking about here. Not the model number,the use. Are we running an oracle database(the pac man of semaphore using aps).

I think A Clay's diagnostic method is exceelent, but frankly the proposed solution is over my head.

Adding this post to favorites.

SEP

[Tim Killinger]

yes, Steven... This machine is running an moderate sized database AND Sun's iPlanet webserver (which I believe I am learning is also a PACMAN of semaphores) 10 gig ram, 4 processors and a va7100 array.


Clay's response fixed me, but I'm always looking for insight on the tuning issues with this sort of server.....

Thanks!