Re: [HACKERS] Re: Backend core dump, Please help, Urgent!

Thanks for the patch. I think I'm going to upgrade to FreeBSD-3.3 and
PG-6.5.3 tonight. Will I still need the patch with 6.5.3? I'm also going
to do a connection test on another offline server to see if it is indeed a
load problem. I'll post the results if anyone is interested.
Thank you for the help,
Matthew

At 08:43 PM 12/15/99 +0900, Tatsuo Ishii wrote:
>> Sure sounds like a corrupted-data problem. Can you use gdb on the
>> corefiles to get a backtrace of what they were doing?
>>
>> > My biggest hang-up is why all of a sudden?
>>
>> Good question. We'll probably know the answer when we find the problem.
>
>Besides the problem Tom has pointed out its possibility, there is a
>known problem with 6.5.x on FreeBSD. It would be rather important,
>since it results in a core dump as well. The problem occurs while a
>backend is waiting for acquiring a lock. Thus it tends to happen on
>relatively heavy load (I observed the problem starting with 4
>concurrent transactions). As far as I know, Linux does not have the
>problem at all, but FreeBSD does. I'm not sure about other
>platforms. Solaris seems to be not suffered.
>
>You could try following patch. It was made for 6.5.3, but you could
>apply it to 6.5.1 or 6.5.2 as well. Current has been already fixed
>with more complex and long-term-aid solution. But I would prefer to
>minimize the impact to existing releases. Keeping that in mind, I have
>made the patch the simplest.
>--
>Tatsuo Ishii
>
>---------------------------- cut here -----------------------------
>*** postgresql-6.5.3/src/backend/storage/lmgr/lock.c~ Sat May 29 15:14:42 1999
>--- postgresql-6.5.3/src/backend/storage/lmgr/lock.c Mon Dec 13 16:45:47 1999
>***************
>*** 940,946 ****
> {
> PROC_QUEUE *waitQueue = &(lock->waitProcs);
> LOCKMETHODTABLE *lockMethodTable = LockMethodTable[lockmethod];
>! char old_status[64],
> new_status[64];
>
> Assert(lockmethod < NumLockMethods);
>--- 940,946 ----
> {
> PROC_QUEUE *waitQueue = &(lock->waitProcs);
> LOCKMETHODTABLE *lockMethodTable = LockMethodTable[lockmethod];
>! static char old_status[64],
> new_status[64];
>
> Assert(lockmethod < NumLockMethods);