Task table full

[Narayanan_2]

Hi,
We are using DS20 Tru64 v4.0 running oracle. Last week the server hang and i had to power off to resolve this. I did a crash dump. When the dump was analysed, i found this statement
"task table full". Can someone advise me what does this mean.

[Michael Schulte zur Sur]

Hi,

please see sysconfig -q proc | grep task
perhaps task-max is to low.

greetings,

Michael

[Hein van den Heuvel]

According to Tony in 1997 (Thanks!) it means:
-----------
The message "task table full" comes from the kernel side of the
fork system call. It can only happen if the kernel fails to allocate
memory (about 2.5KB) for the new task (e.g., the malloc() fails).

Somehow your system is out of physical memory. Use vmstat to find
the "wire" count (pages locked into memory). Chances are good that
this will be a large precentage of total memory and there isn't
enough left for the malloc.

Use vmstat -M to see whos is using the memory.
----------------

You may want to verify the Oracle recommended setting for the sysconfig params: task_max, thread_max,
max-threads-per-user and max-proc-per-user.

Also... you may want to try a more recent Tru64, or at least make sure you have a reasonbably up-to-date patch kit as my archive searches only found 'way old' reference to similar problems.

hth,
Hein.

[Hein van den Heuvel]

Ah! I should have scanned the forum directoy better. There is a recent seemingly related topic. It has an excellent anwer by Joris. Be sure to check it out:
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=261088

Hein.

[Joris Denayer]

Hi,

Increase maxusers in /etc/sysconfigtab.
This will be effective in 99% of the cases.

I hear you say, increase to what value ??


If you run a sys_check on your system, it will probably advise you to increase maxusers to a certain value. This is more or less in function of the amount of memory.


BTW: The default setting of maxusers is 32, which is ways too low for seious production

Joris

[Ralf Puchner]

Hein, the answer given by Joris (your link) is 5.x related not 4.x related.

So increasing maxusers or task-max will solve the problem in 4.x

It is also important to check why number of processes increased - maybe there is a network problem leading to stuck processes waiting for socket close operations or number of users increased over time.

[Joris Denayer]

You say in your first entry, that you did a crash dump.
Do you mean, that you forced the system to crash via the console commands.

In that case, you should have valid core files in /var/adm/crash.

If you have a service contract, you should contact HP services to investigate these files. If there were resource shortages, it will be soon discovered.

Joris

[Ralf Puchner]

Increasing the task-max parameter is a starter but investigating the dump is necessary to get the reason for the task "peak".