Informix hangs

[Franjo Franjic]

Dear colleagues,

I hope for some idea what to do with
our Informix database which have very strange behaviour on the new rp5450 server
(with 4 processor and 2 GB RAM) as Informix database dedicated box.
Version of Informix database is 9.21.FC1.X7 and OS is HP-UX 11.0 (64 bit).
Informix chunks use raw devices on VA7100 disk array (via FC switch).
The HP-UX kernel is configured as suggested in readme file for this version
of Informix (shmmax=2147483648, shmmni=512, shmseg=256, semmni=15000,
semmns=30000, semmnu=7500, nfile=120000)
The symptoms are following:
1) In some cases Informix database "hangs" without any message after some
hours after starting the database (regardless if some user sessions are open and active or not).
Database is not accessible by any way (onmonitor, dbaccess, ???) and cannot be stopped.
It seams that Informix processes are become "zombie" processes that could
not be killed. The only help is to completely restart the server. There are no
error messages in Informix logs (online.log) or in system logs.
After system restart, database is brought online without problems.
2) In some other cases, Informix database stops with status "On-Line (CKPT REQ)" from "onstat -u", which looks like that no more free logs are available, but this is not true since "onstat -l" gives list of many free logs.
We have tried different Informix parameters and size of shared memory, but it didn't help.

Thanks in advance,
FF

[Steve Steel]

Hi

http://www.informix.com/informix/resource/pubs/ius/0914uc1/ius0914r.txt

Records this problem in a slightly earlier Informix version.

Suggest you ask informix.



Steve Steel

[Franjo Franjic]

Thanks Steve!
I have informed Informix support two months ago and there was no help yet. According to them in this version the previous similar bug is fixed. I was hopping that someone could have some further idea what to check more ...
I didn't mention that I have installed latest HP-UX Quality Pack from March 2002.

[Steve Steel]

Hi

register at

http://dbforums.com/

Ask there


Steve

[Steve Lewis]

Are you by any chance using /dev/async and KAIOON=1?

We had a problem last year with fibrechannel and informix on HP where the solution was to recreate /dev/async as
mknod /dev/async c 101 0x0

Secondly check out the following bug numbers:

150790 ( rstcb_cleanup() never exits while other threads wait for the mutex; patch 34906 fixed this on our DB9.21.FC5)

149569 (hang casued by mutex waiting for a thread it already owns.)

153689 (Hang at checkpoint caused by deadlock: Thread X inserting serial value waits for lock on index. Thread Y holds lock on index while waiting for the buffer page that
X has just modified.)

Thirdly try turning setting STMT_CACHE to 0.

Lastly we had problems with fuzzy checkpoints and turned those off too.

[Franjo Franjic]

Thanks!

We don't use KAIOON at all.
Regarding bugs you have mentioned,
I'm trying now to verify if these bugs are showing in our case and if there is a fix for them in 9.21FC1