On 21.01.2016 13:32, Stefan Sperling wrote:
> On Thu, Jan 21, 2016 at 01:20:56PM +0100, Stefan Fuhrmann wrote:
>> TL;DR: Change is effective as per attached test case.
>> No crashes should ever result from it. Some 20% portion
>> of cache memory is guaranteed to work as before which
>> limits potential side effects.
> Sounds like this is related to the yellow box here?
Yes, it is related. The recent trigger was that
mod_dav_svn also causes OOM crashes for complicated
reasons once a directory is no longer cachable.
> We recently ran into a case where the default cache settings resulted
> in performance issues during checkout of a tree with large directories,
> so bad that clients just gave up and disconnected while the checkout
> was still running. Configuring the cache made the system usable.
> If this change avoids this problem by default, I like it.
Well, it pushes the limits of the default (to ~4x
the old value).
In real world applications this means people just
hitting the old limit aren't likely to ever hit
the new limit while those already far beyond it
have fixed their config by now.
Received on 2016-01-21 13:59:10 CET