On August 31, 2019 7:31:12 AM UTC, Dave Thorn <da...@fysh.org.INVALID> wrote: >On Sat, Aug 31, 2019 at 05:13:06AM +0000, Mark Thomas wrote: >> On August 30, 2019 8:34:02 PM UTC, Dave Thorn ><da...@fysh.org.INVALID> wrote: >> > >> >That looks, to me, like the stream is missing the [\r][\n] between >the >> >B50 chunk and the 3FC0 count. >> > >> >I'm not clear on where the responsibilities lie in producing that >> >chunked output - Spring in my webapp, or Tomcat as my container? > >> I'd also recommend repeating your test with Wireshark to confirm >> that the server doesn't send it and that you aren't seeing a client >> side bug. > >Thanks. The only problem with that is the intermittency of the >problem. The server processes a hundred thousand requests a day, or >so, and the problem only occurs a scant handful of times in that. >I'll talk to our admin and see how feasible it is to do that and still >be able to process the results.
I was thinking run Wireshark on your client machine rather than the server. >There is also a haproxy between client and server. You wouldn't think >it goes fiddling around in the middle of a response, thought. I'm >running some tests today that go straight to the server, skipping the >proxy, but again with the intermittency... Indeed. Chunking is sufficiently fundamental that I'm surprised that any component has a bug in this area. >> I took a quick look at Tomcat's ChunkedOutputFilter and I don't see >> any obvious way the \r\n could be skipped but a strange edge case >> isn't impossible. > >Looking at my access log file for yesterday, the test scenario made >5375 requests, and if I look for those in the log, they all have the >same response size: > ># cat /opt/apache-tomcat-8.5.8/logs/localhost_access_log.2019-08-30.txt >| grep 'various-identifiers' | awk '{print $10}' | sort | uniq -c > 5375 970304 > >Could that suggest that Tomcat responded with the same response every >time? If it dropped a "[\r][\n]" then that response size should be >different, shouldn't it? I'd need to check that and I'm on my phone right now. I don't recall which byte count gets logged. > Although I suppose if it could be being >corrupted by a bug then it might be possible that the response size is >calculated in a way that isn't affected by that hypothetical bug? That is certainly possible. Mark --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org