Thanks Yann (the one and only) for quickly adressing this. This explains maybe some of my frustrations in the past with making pools/allocators work in the h2 environment.
-Stefan > Am 06.02.2017 um 01:19 schrieb Yann Ylavic <[email protected]>: > > Hi Stefan, > > On Sun, Feb 5, 2017 at 7:51 PM, Stefan Priebe - Profihost AG > <[email protected]> wrote: >> >> tested your patch against mod_ssl. I haven't seen any pool crashes again >> so it seems to fix this issue. >> >> But two new ones: > > Possibly a promising fix suggested (in another thread) by yet another > talented Stefan :) > He has spotted a possible issue in mpm_event which could affect mainly > mod_http2. > I applied his suggestion in the attached patch, and did some > extrapolation for similar code in mod_h2 (so all possible errors are > mine...). > Would you test this one please? > > @icing: I changed the parent pool of the slave connection from the > mplx pool to the master connection's (ptrans), just to simplify the > allocators in place for now. > I don't see it was needed from a concurrency POV, but if you wanted to > avoid locking when creating slaves' pool we can probably keep the > dedicated allocator finally (to reduce possible contention on > ptrans->mutex? No mutex needed I think since mplx doesn't seem to have > other subpools. Trade off with memory consumption, though). > > > Regards, > Yann. > <ptrans_and_slaves_allocator.patch>
