Joe: with r1892615 I fixed a bug in waiting for the server to become unresponsive after a stop. This *should* make timed waits unnecessary in the http2 test suite.
- Stefan > Am 26.08.2021 um 10:43 schrieb ste...@eissing.org: > > > >> Am 26.08.2021 um 10:27 schrieb Ruediger Pluem <rpl...@apache.org>: >> >> >> >> On 8/26/21 10:21 AM, ste...@eissing.org wrote: >>> >>> >>>> Am 26.08.2021 um 10:16 schrieb Joe Orton <jor...@redhat.com>: >>>> >>>> On Thu, Aug 26, 2021 at 09:51:07AM +0200, ste...@eissing.org wrote: >>>>> https://app.travis-ci.com/github/apache/httpd >>>>> >>>>>> assert env.apache_restart() == 0 >>>>> 3026E assert 1 == 0 >>>>> 3027E + where 1 = <bound method H2TestEnv.apache_restart of >>>>> <h2_env.H2TestEnv object at 0x7ffb03366a60>>() >>>>> 3028E + where <bound method H2TestEnv.apache_restart of >>>>> <h2_env.H2TestEnv object at 0x7ffb03366a60>> = <h2_env.H2TestEnv object >>>>> at 0x7ffb03366a60>.apache_restart >>>>> 3029 >>>>> 3030test/modules/http2/test_002_curl_basics.py:11: AssertionError >>>>> 3031---------------------------- Captured stderr setup >>>>> ----------------------------- >>>>> 3032WARNING: exit 1, stdout: , stderr: (98)Address already in use: >>>>> AH00072: make_sock: could not bind to address [::]:40001 >>>>> 3033(98)Address already in use: AH00072: make_sock: could not bind to >>>>> address 0.0.0.0:40001 >>>>> 3034AH03272: no listening sockets available, shutting down >>>>> 3035 >>>>> >>>>> Looks like the stop/start cycle instead of reload is not timing safe. >>>>> If httpd is no longer responding, it does not mean that the ports are >>>>> available again for the new process. Reloads are just better here, as >>>>> wait loops can be avoided. Hmm....thinking... >>>> >>>> Yeah :( Next build I've pushed has r1892598 reverted and a sleep(2) >>>> after running apachectl, that was sufficient for me when I was testing >>>> locally before. >>> >>> Can we add a configuration directive for this server behaviour? Like >>> >>> AcceptOnGraceful on|off >>> >>> So that people who do not want the server to accept new connections >>> during graceful reloads can influence this? >> >> What about a graceful stop first and a start afterwards? > > Just tested: "graceful-stop" seems to make it worse. Connections are > being served using the old configuration. > >> >> Regards >> >> RĂ¼diger