memory fragmentation issues in 1.2?
sgrimm at facebook.com
Wed Dec 6 22:31:23 UTC 2006
It's possible, but IMO unlikely, that this is memory fragmentation. What
do your requests look like? How many client connections are open at a
time? Each client connection has some buffer space for incoming
requests, and if you have a lot of clients that can add up to a fair bit
of memory, especially if your application does very large multi-key
"get" requests. Based on our experience that's what I'd say you're most
likely running into; we see our memcached instances exceed their
configured memory limits but it always stabilizes once they've reached a
That said, please try 1.2.1 and see if you see the same problem.
George Schlossnagle wrote:
> I'm seeing some weird behavior on my memcached 1.2 instances. Overtime
> the RSS of my processess is growing and growing. I start out at around
> 800M (my configured memory size), and it monotonically grows (over a
> period of a couple weeks) until it is over double that size and gets
> killed by the OOM killer.
> Trying to duplicate this behavior under valgrind hasn't been very
> fruitful, which makes me suspect it's system memory fragmentation.
> Has anyone else seen this sort of behaviour? Is there some likely EBKAC
> that I might be engaging in that could cause this? I have some ideas
> for working around this, but I'd much rather just be told I'm stupid. :)
More information about the memcached