On 17.01.2019 15:05, Yann Ylavic wrote: > On Thu, Jan 17, 2019 at 2:55 PM Stefan Eissing > <stefan.eiss...@greenbytes.de> wrote: >> >>> Am 17.01.2019 um 14:04 schrieb Yann Ylavic <ylavic....@gmail.com>: >>> >>> On Thu, Jan 17, 2019 at 1:56 PM Branko Čibej <br...@apache.org> wrote: >>>> On 17.01.2019 13:55, Branko Čibej wrote: >>>>> On 17.01.2019 13:21, Yann Ylavic wrote: >>>>>> On Thu, Jan 17, 2019 at 1:02 PM Yann Ylavic <ylavic....@gmail.com> wrote: >>>>>>> On Thu, Jan 17, 2019 at 12:50 PM Branko Čibej <br...@apache.org> wrote: >>>>>>>> Other than that, unlimited max-free is wrong in most cases, so why not >>>>>>>> set the default to something sane instead? >>>>>>> Agreed, something like 10 pools (80K)? >>>>>> Probably at bit too agressive... >>>>>> >>>>>> In httpd that's 2MB, which default is used in svn? >>>>> ./subversion/include/svn_pools.h >>>> Huh, something broke my MUA ... there should be some more on that line: >>>> >>>> #define SVN_ALLOCATOR_RECOMMENDED_MAX_FREE (4096 * 1024) >>> OK, so I'd propose 3072 * 1024 :) >> And so, the binary search for the optimal value started... > Dichotomy is not the worst search algorithm, though :) > > I suppose there is as much optimal value as applications though, on > the APR side it's more of a safeguard than anything el
Indeed. Our documentation should recommend to downstream users to set the max-free to a value that's reasonable for their application. It wouldn't hurt to hint at what the tradeoffs are. -- Brane > se I suppose.