How trouble shoot memory / swap space problems

[Peter Gillis]

Hi,
hpux 11.11v1, model RP2470.
I have seen a message come up on our ux box and I am unsure how/where to start investigating this problem. Would anyone be able to offer any info...
the message I received follows:

PID 27217 received a SIGSEGV for stack growth failure. Possible causes: insufficient memory or swap space, or stack size exceeded maxssiz.

We did have a problem with an mqseries process but that is going now, and since I am unsure what the pid 27217 was (it does not exist now) I cannot say whether this was the cause of the sigsegv message or not. Either way, I should be aware of how to start investigating.
any help much appreciated.
Maria

[Bill Hassell]

If the program is like most Unix code, it probably crashed and left a core file behind. That file will identify the process name and the date on the file will tell you when it crashed.

Ideally the author of the program has provided adequate documentation as to the amount of RAM required by the program so you can change the kernel values as needed. (OK, I know that is a silly ideal...) Start by bumping up maxdsiz to 2Gb and maxdsix_64 to 8Gb or more. These fences cause more troubles than they prevent so set them plenty high. If a bad program is grabbing all your RAM, you'll see it with:

UNIX95=1 ps -e -o vsz,ruser,pid,ppid,args | sort -rn | head -15

[Dennis Handly]

Generally the first two causes in the message are incorrect. Typically this is due to a programming error where there is a recursive stack overflow, so matter how big you make maxssiz, it will still fail. I recently had firefox abort with about 2000 frames of doing something to some "subobject".

>Bill: Start by bumping up maxdsiz to 2Gb and maxdsix_64 to 8Gb or more.

This is maxssiz, not maxdsiz, so this won't help.
What is your current value of maxssiz? If not 80 Mb, I suppose that would be a good value. Unless this is some evil Fortran application.

[mobidyc]

Hello,

if you want to have a process history with their exit status code, you can look at the process accounting.

look at the map page for acct(1M).

generally, your problem is swap or maxssiz problem, run 'swapinfo -tam' for a look your swap.

for maxssiz you'll find more informations at the following url:
http://docs.google.com/View?docid=dc7s6mzr_56cgsg9g

Regards,
Cedrick Gaillard