Lazy Garbage Collection Question

[Philip Neustrom]

The reaper was necessary because of the server fall-back implemented
in memcached clients.  The individual wanted a safe way to reset a
bunch of machines' caches without worrying about machine X's items
falling to machine Y and then machine X coming back online, receiving
the items and having the items in both X and Y -- which can lead to
out-of-date items being served up.  Realistically it shouldn't be a
problem because you can usually set clients up (no fall-back
temporarily, things like that) so that a restart across a bunch of
servers all at once is possible without a chance for corruption.

--Philip Neustrom

On 4/29/06, Adam Hiatt <adam.hiatt at cnet.com> wrote:
> I've noticed a couple threads in the mailing list archives with
> questions concerning the lazy garbage collection for expiring
> entries. One user attempted to write a 'reaper' agent that
> specifically cleaned up their cache. Now this clearly should not be
> necessary, but from what I read it appears that they felt that it was
> critical because the memcached processes kept using memory until it
> had to swap to disk. I wasn't able to determine this from the
> threads, so, is the consensus that these deployments of memcached are
> configured to use more memory than the system will provide to it and
> that when the configured level is approached the OS is forced to swap
> to disk? Or, is there some actual problem that prevents deallocation
> and causes memory leaks force these swaps. I am concerned because I
> am looking into using memcached in a production environment and on at
> least one of the threads I referenced, I noticed that the
> administrators had to restart the boxes every couple of days or so.
>
> ___________________________
> Adam Hiatt (adam.hiatt at cnet.com)
>
>
>