RFR: 6968351: httpserver clashes with delayed TCP ACKs for low Content-Length [v13]

[robert engels]

On Sun, 21 Apr 2024 18:49:45 GMT, robert engels <duke at openjdk.org> wrote:

>> fix bug JDK-B6968351 by avoiding flush after response headers
>
> robert engels has updated the pull request incrementally with one additional commit since the last revision:
> 
>   remove FixLengthOutputStream changes since not needed

Hi all. I simplified the PR quite a bit by removing the changes to the fixed length output stream by flushing in ServerImpl.

I think all existing "bad" behavior is accounted for and you still have the performance benefit of not sending small packets.

I am still unclear about the concerns regarding "existing code will break" - since if the handler never closes the output stream or the exchange, the connection would be dead and any further requests on it would hang. It seems there are some broken tests cases that exhibit this - which this patch still allows to work - but I don't think the test case behavior could ever be found in the real world.

Let me know if there is anything else I need to do or clarify. I very much appreciate all of the feedback I have received.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/18667#issuecomment-2068165518