Re: Memory Hogging
Florian La Roche (florian@rw22floh.jura.uni-sb.de)
Wed, 26 Jul 1995 11:41:54 +0200
>>with the httpd process, once in a while the system load will shoot up
>>from an average of 0.05 to about 10 and stay there for 5 minutes whilst
>>a httpd process decides that it is not actually going to get any more
>>memory and dies.
>
>>This effect happens both with the self-compiled and pre-compiled binaries, but
>>I figure the problem lies in the kernel since it shouldn't behave that way.
>
>I experience this too. But I think the problem should primarily be looked
>for in the http daemon. The problem at my end is that it only happens
>(at most) once a week. This is too seldom to perform any real diagnostics.
>
>Have you tried attaching an strace to the runaway httpd? What does it
>say? Also, you might try recompiling httpd with debugging information, so
>that you can either attach a gdb to the process or analyse the coredump
>and find out were it is looping in the code.
I have seen the same effect several times with the gopher daemon. Never
with the NCSA httpd. I suspected, that the gopher-daemon is the bad ting,
but maybe the kernel could do something wrong. (Linus? Any Idea?)
(Some timer stuff going mad? Seems to be somewhat related to networking code.)
The symptom was, that the load goes up to about 20. Killing the gopher-daemon
will fix everything. That's what I did about 4 times. About 2 weeks ago,
I have seen the same thing, but after about 2 minutes, everything was ok
again.
Next time, I'll try to do an strace. (I can't always log in in that situation.)
Florian