solaris 8/9, memcached consumes all CPU after request

Brad Fitzpatrick brad at danga.com
Fri Nov 5 14:14:54 PST 2004


Looks like a libevent bug --- we never do any ioctls in memcached.

I'd suggest rebuilding libevent without devpoll support, so it falls back
on select or poll, but then you'll eat more CPU.  It's probably best to
work with the libevent developer, especially since you have such an
easily-reproducable test case.


- Brad


On Fri, 5 Nov 2004, Jim Lawson wrote:

> Hi memcached at lists.danga.com,
>
> We're trying out memcached on our web farm, which is mixed Solaris and
> Linux, mostly Solaris.  It looks pretty cool.  It seems to work fine
> under Linux (RHEL3, 2.4.21-EL3 kernel.)  However, under Solaris 8 or 9,
> after the first request is answered and the TCP session is closed,
> memcached starts looping and consuming 100% of the CPU it's running on.
>
> Running memcached 1.1.11, libevent 0.9.
>
> Running EVENT_SHOW_METHOD=1 /usr/local/bin/memcached -u nobody returns:
> libevent using: devpoll
>
> I have attached an edited transcript of "truss" against the running
> memcached which shows the behavior.  This may be just a configuration
> issue, but I'm afraid it's out of my league.  Any pointers?
>
> Jim
>
>


More information about the memcached mailing list