On 06.03.2012 10:43, Philip Martin wrote:
> Daniel Shahaf<danielsh_at_elego.de> writes:
>> I believe you cannot assume svn_fs_t's will be short lived, and
>> _certainly_ cannot assume that the access pattern to an svn_fs_t
>> will be in any way related to random clients having random RA
>> sessions; for all the FS API knows, svnserve is a daemon that calls
>> svn_fs_open() once per reboot.
>> And yes, people do this.
> Perhaps the read routines could be changed to always read the
> revprop-gen file and refresh the cache if required. Obviously this
> would involve more IO than the current caching strategy, but it would
> still be more efficient than the the previous non-cached approach.
Actually, this is a good idea -- maybe with
some lower rating on "style" (adding a one-
liner to most of the FS_FS API functions).
And we don't have to actually re-read the
revprop-gen file but only need to invalidate
the cached revprop-gen value. Only the
next time we read some revprop, will the
generation info be read from file again.
I'll implement that in the next couple of days.
Received on 2012-03-07 09:41:23 CET