below + POD wakeup Did not force the path yet where the listener is started (or fold in the scoreboard change )
On Tue, Mar 12, 2024 at 10:54 AM Eric Covener <cove...@gmail.com> wrote: > > On Tue, Mar 12, 2024 at 10:30 AM Eric Covener <cove...@gmail.com> wrote: > > > > On Tue, Mar 12, 2024 at 10:19 AM Yann Ylavic <ylavic....@gmail.com> wrote: > > > > > > On Tue, Mar 12, 2024 at 3:03 PM Eric Covener <cove...@gmail.com> wrote: > > > > > > > > On Tue, Mar 12, 2024 at 8:48 AM Yann Ylavic <ylavic....@gmail.com> > > > > wrote: > > > > > > > > > > Maybe it could be: > > > > > if (threads_created) { > > > > > > > > not listener_started? > > > > > > > > threads_started>0 could just mean we had no scoreboard issues but > > > > pthread_create failed on anything but the first thread. > > > > > > Don't we want the workers to gracefully stop whenever at least one was > > > created, or we may deadlock in clean_child_exit()? > > > > Yes, I see what you mean now. I will try to force some pthread_create > > errors w/ the patch soon. > > It seems like if there is no listener yet, the > signal_threads(ST_GRACEFUL) will fail to interrupt the queue (because > it defers to the woken up listener) > Maybe we do it inline before apr_thread_exit/ > > > > -- > Eric Covener > cove...@gmail.com -- Eric Covener cove...@gmail.com
start-threads-2.diff
Description: Binary data