On Thu, Apr 28, 2016 at 2:40 AM, Yann Ylavic <ylavic....@gmail.com> wrote: >> >> What I saw was memory nullified, by assumedly, new apr_pcallocs, that should >> not have been freed already. With more likelihood the more load and the less >> logging, of course. > > So I'm rather thinking of a pool lifetime issue in h2 buckets > handling, did you see the comment in ap_bucket_eor_create(), the case > does not seem to be handled in h2_beam_bucket_make() (if that's the > bucket you are talking about).
Hmm, the beam bucket is DATA (not META), so the issue would instead be in h2_bucket_{eos,eoc}_destroy(), since h2_bucket_{eos,eoc}_make() don't seem to take care of the h2_{stream,session} being destroyed outside the bucket handling... Is that possible?