RFR: 6968351: httpserver clashes with delayed TCP ACKs for low Content-Length [v4]

[robert engels]

On Fri, 19 Apr 2024 17:24:33 GMT, Daniel Jeliński <djelinski at openjdk.org> wrote:

>>> The problem with closing a chunked stream is that it flushes both before and after the final chunk. I assume it may also cause delays.
>>> 
>>> The only case when you don't need to close the exchange or the output stream is when your response contains no content.
>> 
>> I don't understand this in reference to the other tests failing because they are not closing the output stream or ending the exchange. If that is incorrect behavior - because they are sending content - then those tests are invalid.
>> 
>> Still I have a fix for this I believe.
>
> ah no, this is not related to the test failures; this is only about the original issue (small sends result in problems with delayed acks)

> The problem with closing a chunked stream is that it flushes both before and after the final chunk. I assume it may also cause delays.
> 
> The only case when you don't need to close the exchange or the output stream is when your response contains no content.

Can you point me to the code covering the before and after the final chunk? I would think the before flush should be removed.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/18667#discussion_r1572720529