On 19/11/2019 11:16, M. Manna wrote: > Sorry Dumb question earlier - It is in Coyote jar. > > We are currently in a rollback state in our prod, but what you reported > above sounds promising. If revert the Coyote jar resolves issues, which is > the earliest/latest version we can revert to?
I wouldn't recommend reverting just the Coyote JAR. It might work, but such configurations are a) untested and b) unsupported. 8.5.42 is the most recent version without the regression until 8.0.50 is released. Alternatively, reducing maxConnections (default is 8192) to 2048 or possibly 3072 should also work around it. Mark > > Thanks, > MAnna > > On Tue, 19 Nov 2019 at 10:32, M. Manna <manme...@gmail.com> wrote: > >> Mark, >> >> On Mon, 18 Nov 2019 at 19:28, Mark Thomas <ma...@apache.org> wrote: >> >>> On 18/11/2019 14:14, Mark Thomas wrote: >>>> On 18/11/2019 12:06, M. Manna wrote: >>>>> Mark and others, >>>>> >>>>> On Mon, 18 Nov 2019 at 12:01, Mark Thomas <ma...@apache.org> wrote: >>>>> >>>>>> OK, it looks like I can reproduce this. >>>>>> >>>>>> Steps to reproduce: >>>>>> >>>>>> - Windows 2016 Server fully patched >>>>>> - Java 1.8.0u144 >>>>>> - Install Tomcat 8.5.45 from windows installer >>>>>> - Add tcnative-1.dll (64-bit) from Tomcat Native 1.2.23 >>>>>> - Modify server.xml to use Http11AprProtocol on port 8080 >>>>>> - Make a single request >>>>>> >>>>>> I then see 1 core running at 100% until the connection times out after >>>>>> 20s. Make another request and a core goes back up to 100% for 20s (the >>>>>> default keep-alive time out). >>>>>> >>>>> >>>>> I have also successfully reproduced this with making a single request >>>>> (sorry for not replying in the weekend). Not sure how your graph looked >>>>> like, but the Jvisualvm showed me a Sinusoidal modulation curve as >>> soon as >>>>> the request hit the server. and it didn't go down at all. >>>> >>>> I see similar behaviour on Windows 7 but the the CPU usage drops after >>> ~5s. >>>> >>>> A binary search indicates that the issue was introduced with this >>> commit: >>>> >>>> >>> https://github.com/apache/tomcat/commit/fffb08790e642e03f00c5f96a3a61ee09a2c8342 >>>> >>>> (this is for 9.0.x - 8.5.x and 7.0.x had similar commits) >>>> >>>> However, that code was removed when APR was switched to a single poll >>>> set. >>> >>> Ah ha. It was removed in 9.0.x but not in 8.5.x (only 9.0.x switched to >>> a single Poller) so it does look like this change is responsible. >>> >>> Thanks for Debugging this. Are you saying that we will have a patch for >> this? If so, which jar could we replace in 8.5.45 until >> we move to new tomcat? >> >> Thanks, >> >> >>> Mark >>> >>> --------------------------------------------------------------------- >>> To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org >>> For additional commands, e-mail: users-h...@tomcat.apache.org >>> >>> > --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org
