On Fri, 4 Apr 2025 15:27:25 GMT, Daniel Fuchs <dfu...@openjdk.org> wrote:
>> test/jdk/java/net/httpclient/HttpResponseConnectionLabelTest.java line 412:
>>
>>> 410: HttpResponse<String> response1 = client.send(pair.request,
>>> BodyHandlers.ofString(CHARSET));
>>> 411: LOGGER.log("Firing request 2...");
>>> 412: HttpResponse<String> response2 = client.send(pair.request,
>>> BodyHandlers.ofString(CHARSET));
>>
>> In theory, there's no guarantee that these sequential requests will be
>> executed "immediately" one after the other. Internally, in the httpclient
>> implementation, we use idle timeouts to close idle connections. So I'm
>> wondering if it's realistic that there would ever be a case where in some
>> setup (like the CI), these two execute so far apart from each other
>> (`-Xcomp`?) that the connection might have timed out in the meantime and
>> closed? Thus the second request ends up using a different connection and
>> fails this test?
>> Should we perhaps use `othervm` for this test and configure an extremely
>> high idle timeout of connections, through the system properties, to avoid
>> such intermittent failures?
>
> Good point - our default idle timeout is 30s IIRC. There would need to be a
> pause of 30s between the two calls to `client.send` however. Not impossible
> but unlikely. We could either pass a higher timeout (with e.g.
> `-Djdk.httpclient.keepalive.timeout=120`) preemptively, or wait until we see
> the test fail...
Waiting for the test to fail due to this reason and then adjusting the timeout
is fine with me.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/24154#discussion_r2029126877
