On Mon, Jul 10, 2023 at 10:53 AM Ruediger Pluem <rpl...@apache.org> wrote: > > On 7/10/23 10:11 AM, Joe Orton wrote: > > On Fri, Jul 07, 2023 at 03:52:46PM -0000, yla...@apache.org wrote: > >> Author: ylavic > >> Date: Fri Jul 7 15:52:45 2023 > >> New Revision: 1910861 > >> > >> URL: http://svn.apache.org/viewvc?rev=1910861&view=rev > >> Log: > >> ab: Fix accounting of started connections. > >> > >> Revert when a kept alive connection is aborted on read. > >> Stop the polling loop when there is nothing to poll anymore, it's simpler. > > > > Not related to this change but I noticed that trunk ab has stopped > > failing if trying to connect to a closed port, strace looks like: > > > > epoll_wait(3, [{events=EPOLLOUT|EPOLLERR|EPOLLHUP, data={u32=33073216, > > u64=33073216}}], 2, 30000) = 1 > > epoll_wait(3, [{events=EPOLLOUT|EPOLLERR|EPOLLHUP, data={u32=33073216, > > u64=33073216}}], 2, 30000) = 1 > > ... > > > > With HUP set in the returned events it enters the first if() at > > https://github.com/apache/httpd/blob/trunk/support/ab.c#L2523 but > > STATE_CONNECTING is not handled there and so it loops indefinitely. > > That's as far as I debugged it, sorry. Any ideas? > > I would guess that the below fixes the loop, but I am not sure if it is the > correct fix. > > Index: support/ab.c > =================================================================== > --- support/ab.c (revision 1910910) > +++ support/ab.c (working copy) > @@ -2534,6 +2534,11 @@ > case STATE_READ: > read_response(c); > break; > + case STATE_CONNECTING: > + if (rtnew & APR_POLLHUP) { > + try_reconnect(c, APR_ENOPOLL); > + } > + break; > } > > continue;
I think this is not reached with rtnevents == POLLOUT|POLLHUP because it takes the first POLLIN|POLLHUP continue-branch. I moved the check for POLLOUT first in r1910911, which fixed the issue for me. Regards; Yann.
