That feeds into the Grand Solution to make h2 workers part of the MPM. Not sure how to approach that.
The common way today is that master connections do not return from run_connection if they are busy handling requests. If that takes longer than H2Timeout to produce results, it may happen nevertheless, but that should then drag down the conn anyway, I suppose. > Am 30.12.2015 um 19:42 schrieb Yann Ylavic <[email protected]>: > >> On Wed, Dec 30, 2015 at 5:12 PM, <[email protected]> wrote: >> Author: icing >> Date: Wed Dec 30 16:12:35 2015 >> New Revision: 1722369 >> >> URL: http://svn.apache.org/viewvc?rev=1722369&view=rev >> Log: >> fixes after fuzzing tests, changed H2KeepAliveTimeout default > [] >> >> Modified: httpd/httpd/trunk/modules/http2/h2_workers.c >> URL: >> http://svn.apache.org/viewvc/httpd/httpd/trunk/modules/http2/h2_workers.c?rev=1722369&r1=1722368&r2=1722369&view=diff >> ============================================================================== >> --- httpd/httpd/trunk/modules/http2/h2_workers.c (original) >> +++ httpd/httpd/trunk/modules/http2/h2_workers.c Wed Dec 30 16:12:35 2015 >> @@ -272,7 +272,7 @@ h2_workers *h2_workers_create(server_rec >> apr_atomic_set32(&workers->max_idle_secs, 10); >> >> apr_threadattr_create(&workers->thread_attr, workers->pool); >> - apr_threadattr_detach_set(workers->thread_attr, 0); >> + apr_threadattr_detach_set(workers->thread_attr, 1); > > Hmm, aren't worker threads joined? > How does it work with eg. graceful restarts where (active) workers > need to be waited before the child exits? > > Regards, > Yann.
