[Turck-mmcache-users] Problem after accessing mmcache.php

[Matthias S. <mat...@rp...>]

Hi,

I've just replaced php-accelerator with mmcache this morning on a cluster
of 7 servers, and everything was running fine for over 5h now. The load of
all servers is about the same with both, right now all are between 1 and 3.

But on one, I accessed the mmcache.php page. No problem so far. I saw that
of the 20MB of shm I configured, only 9 were used. With php-accelerator, I
already was able to fill 16MB after a few days, and things got nasty when
it reached the limit, so 20 seems a safe value. To see how things were
evolving, I decided to reload the mmcache.php page... and it just waited...
so I tried again a few times, thinking it could be my browser or proxy, but
still nothing. I then had a look at my terminal in which I had an ssh
session to the server with a "top" running. Well, it froze with 0% idle on
both CPUs (they were 60% idle just before), and the load had quickly raised
from 1.5 to 10.

No signs of quick heavy swapping, and right now the server is only
responding to pings. I'm pretty sure it's related to mmcache, since those
are very reliable servers (uptime of 180 days, since the last kernel
upgrade). Right now, it's unbound from the web cluster, so no more web
traffic is sent ot it, but still isn't reachable through ssh... I think
I'll have to reboot it.

Are there any known issues that could have caused that? Maybe me insisting
on loading the page again and again? Any help welcome. Also, if there is
anything I can do to troubleshoot the problem (apart from killing another
server ;-)), let me know.

I'm running Red Hat Linux 7.3 with all updates, apache 1.3.27 and php 4.1.2
with custom modules on Compaq dual P3 servers.

Matthias

-- 
Clean custom Red Hat Linux rpm packages : http://freshrpms.net/
Raw Hide 20030616 running Linux kernel 2.4.20-20.1.2002.nptl
Load : 0.41 0.51 0.45