shinrich commented on issue #8200: URL: https://github.com/apache/trafficserver/issues/8200#issuecomment-904652782
Working with this again yesterday with my very active H2 session between two ATS boxes. Finally put an assert on ATS2 when the max_concurrent limit is set so I could look at the state of the stuck streams. The problem appears to be that a number of streams had stalled due to empty session windows. So ATS1 had decremented its peer stream count (presumably timed out/client aborted), but it did not send a reset to ATS2 so it could clean up as well. I would assume the ATS2 would time the stream out eventually as well. I added the session/stream window split logic and increased the session window and the starvation went away. Hitting the max_concurrent limit is really bad because the next frame on that stream will take down the entire session. However @maskit seems to indicate in issue #5675 that we can just ignore such frames instead of sending a connection error. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
