Thanks for the logs. That was really helpful. If you can build a 2.4.55 yourself, could you try the following patch?
h2-access-500-on-rst.diff
Description: Binary data
Analysis: The 500 in the access log happens when a client RST (resets) a stream before the response has really started. The HTTP/2 processing of this is totally fine and as should be. We just have an entry in the access log where there should be none. The cause is code that is supposed to catch a rare error condition. This triggers and produces extra "SOMETHING HAS GONE WRONG" information which then leads to a wrong entry in the access log. The patch avoids this by not entering this code path when the request has been aborted. I hope this solves the wrong access logging completely. Sorry for not having caught this. I do not have test cases that check entries in the access log. Kind Regards, Stefan > Am 18.01.2023 um 18:40 schrieb Stefan Eissing via dev <dev@httpd.apache.org>: > > Helmut, > > thanks for setting this up. If you can go a bit deeper and get me logs with > "http2:trace2" that would be supberb. > > If you want to keep them out of the open, you can mail at ste...@eissing.org > > Thanks a lot! > > Stefan > >> Am 18.01.2023 um 17:00 schrieb Helmut K. C. Tessarek <tessa...@evermeet.cx>: >> >> On 2023-01-18 06:43, Stefan Eissing via dev wrote: >>> In that case, maybe your could run the 2.4.55 mod_http2 with "LogLevel >>> http2:debug" until some 500s show up. That would be very interesting to >>> analyze. >>> You can mail/link me such a log confidential atste...@eissing.org, if you >>> like. >> >> I've setup a test VM and was able to reproduce the problem. However, in my >> test I got only one 500 error, but I do have an access log and an error_log >> set to http2:debug. >> >> Do you want the 2 logs or is Andreas' log sufficient? >> >> Cheers, >> K. C. >> >> -- >> regards Helmut K. C. Tessarek KeyID 0x172380A011EF4944 >> Key fingerprint = 8A55 70C1 BD85 D34E ADBC 386C 1723 80A0 11EF 4944 >> >> /* >> Thou shalt not follow the NULL pointer for chaos and madness >> await thee at its end. >> */ >> >
