Well, since this problem is intermittant, and that I can't leave rogue
processes running in production, I have not been able to perform the latest
suggested tests (those against the kernel /proc dir).
As soon as I have additional information, I will surely post to the list.
At 5:39 PM -0400 10/18/99, Steve Reppucci wrote:
>Hmmm. On Friday night one of my servers (Solaris 5.6, modperl 1.21,
>Apache 1.39) was hosed. Logging in to it, I found about 20 httpd children
>spinning, eating cpu cycles.
>
>I didn't dig into it, just killed the parent server (ungracefully) and
>restarted, but mentioning it here in case this is something that's
>happening to a bunch of us running Apache/modperl on Solaris.
>
><Steve Reppucci>
>
>On Mon, 18 Oct 1999, Eric Cholet wrote:
>
>> On Monday, October 18, 1999 5:42 PM, Marc D. Spencer
>>[SMTP:[EMAIL PROTECTED]] wrote:
>> > Hi.
>> >
>> > I have an apache server (1.3.6/Solaris 5.6) running mod_perl/1.21.
>> >
>> > For it, I have a fairly complex handler involving several classes, some
>> > of which bind to libraries (one of them is ImageMagick).
>> >
>> > The server is configured to only allow requests for this module, and
>>does
>> > not serve any static pages.
>> >
>> > At least once a day, I will find an http process monopolizing one of the
>> > 4 CPU's in our production machine (E-450); server-status claims it's in W
>> > mode (waiting for connection) but the log that our module generates
>> > indicates it hangs at an arbitrary point within it's flow.
>>
>> For what it's worth, I've been experiencing the same situation. Happens less
>> often, but that may be just because I have less trafic. Mind that W mode is
>> 'sending reply' which is consistent with the module being hung.
>> I haven't raised this issue on the list because I haven't been able to track
>> it down and therefore I assume it's a bug in my code. Since I do socket
>>stuff
>> I could very well be blocking on a system call.
>>
>>
>> >
>> > We have poured over the code, and over the FAQ's, etc, at the mod_perl
>> > site, but have no ideas beyond a cron to kill these abhorrant processes
>> > when their cume CPU time exceeds some level (not a great measure...we had
>> > to set the threshold very high as to not kill happy processes).
>> >
>> > Can someone shed any light on what may be happening, or a better way to
>> > monitor and kill these children?
>> >
>> > Thanks, in advance.
>> > Marc Spencer
>> > www.kodak.com
>>
>> --
>> Eric
>>
>
>=-=-=-=-=-=-=-=-=-=- My God! What have I done? -=-=-=-=-=-=-=-=-=-=
>Steve Reppucci [EMAIL PROTECTED] |
>Logical Choice Software http://logsoft.com/ |
>508/958-0183 Be Open |
"0201: Keyboard Error. Press F1 to continue."
-- IBM PC-XT Rom, 1982
\---------------------------------------------------------------\
\ Marc D. Spencer Any opinions expressed here \
\ Applications Engineer are my own and are not \
\ Network Services necessarily those of my \
\ Eastman Kodak Company employer. \
\---------------------------------------------------------------\
