RFR: 8304701: Request with timeout aborts later in-flight request on HTTP/1.1 cxn [v3]

[Conor Cleary]

On Mon, 20 Nov 2023 10:11:52 GMT, Conor Cleary <ccleary at openjdk.org> wrote:

>> **Problem**
>> When using HTTP/1.1 with HttpClient, it was observed that requests configured with timeouts at build time fail with a HttpTimeoutException when they are redirected to a separate URI by a server (status code 3xx on first response). What should happen is that the second request response (so after receiving a 3xx code) clears restarts the timer intially set. However, when `responseTimerEvent` is registered for the first time, it is not unregistered and cleared before starting a second timer.
>> 
>> **Solution**
>> This fix addresses the issue by calling `cancelTimer()` in `MultiExchange.java` whenever the `newRequest` reference is set to a non-null value after calling `responseFilters(response)` on the initial response received. This occurs in the case where a status code 3xx is received in the initial response. When `cancelTimer()` is called, it now unreferences `responseTimerEvent` after cancellation operations take place.
>> 
>> While the fix for the issue was relatively straight forward, the regression test is less so. I would point to to the comment located in `RedirectTimeoutTest:L119` for an explanation of the testing method.
>
> Conor Cleary has updated the pull request incrementally with one additional commit since the last revision:
> 
>   8304701: Increased timeout values again for test stability

Increased the value of `TIMEOUT_MILLIS` and `SLEEP_TIME` to help test stability. I will of course keep an eye on this regression test as time goes on but all seems in order now 👍

It didn't seem necessary to increase the number of iterations in the test loop and I don't want to make the test run for any longer than it has to. So, I've left the `ITERATIONS` field unchanged.

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

PR Comment: https://git.openjdk.org/jdk/pull/16689#issuecomment-1818787432