On Fri, Mar 15, 2019 at 3:19 PM <r...@apache.org> wrote: > This is an automated email from the ASF dual-hosted git repository. > > remm pushed a commit to branch master > in repository https://gitbox.apache.org/repos/asf/tomcat.git > > > The following commit(s) were added to refs/heads/master by this push: > new 004fe8b Fix regular read taking over async reads > 004fe8b is described below > > commit 004fe8bbab588b97b61f0bb03726deb379d0227b > Author: remm <r...@apache.org> > AuthorDate: Fri Mar 15 15:19:13 2019 +0100 > > Fix regular read taking over async reads > > A call to registerReadInterest with the right timing could cause a > regular read to occur, preventing the use of vectored IO. This is > exhibited in the very elusive client timeout failure seen on CI, for > example here in build #4128: > > https://ci.apache.org/projects/tomcat/tomcat9/logs/4128/TEST-org.apache.coyote.http2.TestHttp2Section_6_7.NIO2.txt
I have been using the cpulimit tool on Linux to be able to reproduce this one and then figure out the right "speed" to do it reliably and be able to debug it. I'm still seeing NIO2 failures that don't occur with NIO, but they seem to be caused by "legitimate" IO server side timeouts. Everything can happen veeeeery slowly with cpulimit and it's possible the read patterns are different between the two. I will continue to investigate though. Rémy
