memory fragmentation issues in 1.2?

Paul T pault12345 at
Thu Dec 7 03:38:14 UTC 2006

One more thing. Your version of libevent might be
leaking (or the way memcached interacts with libevent
might not be good for this particular version of
libevent). The way memcached is using libevents'
timers - that kind of stuff.

If that is the case - most likely you would not be
able to capture it with valgrind easily (and it is
also consistent with the rest of the picture).

There was one very obscure wild pointer once in univca
code and the only way I was able to capture some funny
things in univca/libevent interactions was to hit
univca directly with 30K hps with ab - under valgrind
- and *sometimes* it will show the wild pointer. I
don;t know how can you possibly do that with
memcached, though. I guess one can write a custom
stress-testing tool for memcached.


Want to start your own business?
Learn how on Yahoo! Small Business.

More information about the memcached mailing list