shinrich commented on pull request #7622: URL: https://github.com/apache/trafficserver/pull/7622#issuecomment-886790190
In my production testing, I am having problems with increased ERR_CONN_FAIL from another ATS box acting as origin as described in issue #8163. Adjusting the error rate up helped some, but the ERR_CONN_FAIL was still significantly higher for that origin. The mechanism was the same. The origin was returning GOAWAY with no error which caused all other active streams to be terminated. Finally, Friday I notice that the H2 client logic was using the "Connection: closed" header as described below to start draining the session. https://docs.trafficserver.apache.org/en/latest/admin-guide/plugins/header_rewrite.en.html?highlight=drain#close-connections-for-draining Unfortunately, I cannot get the logic to keep around existing streams until the are finished working. I just move my ERR_CONN_FAIL errors to ERR_CLIENT_ABORT or ERR_READ_TIMEOUT errors. With recent changes to improve header error checking, I've been adding "Connection: close" headers not worrying about the HTTP/2 case since H2 should be ignoring the Connection header. Forgetting about this draining feature. Not sure how to deal with this. Could add H1/H2 checks into the HttpSM to only really add the "Connection: close" for Http/1 connections. Or more likely add a virtual method onto the ProxyTransaction to do that logic. -- 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]
