shinrich opened a new pull request #7828: URL: https://github.com/apache/trafficserver/pull/7828
This fixes an issue introduced in PR #7418. I just filed issue #7827 about it. The 9.1.x servers would consistently have slightly more ERR_CONNECT_FAILS than its 9.0.x peer. I finally got one of the transactions in a packet trace and it looks like the origin had closed the connection. Then a short while later ATS would start sending traffic on the connection. Then the resets would fly and the transaction would fail due to an EOS in state_read_response_header. I finally tracked it down to a change in how the session is released to the pool. In 9.0.x, the release involves doing do_io_read to the pool with max int bytes against the server session's buffer. In 9.1.x the buffer is null, so the vc is not enabled. So the EOS is not delivered while the session is in the pool. I confirmed this with additional warning messages in a test build. With this change we add back the buffer into the do_io_read. The EOS's are being delivered while the session is in the pool. With this change the ERR_CONNECT_FAILS are consistent between the 9.1.x box and its 9.0 peer in production traffic. This closes #7827. -- 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. For queries about this service, please contact Infrastructure at: [email protected]
