moving the context of issue https://github.com/elastic/elasticsearch/issues/39939 here. Looks like on my machine the assumption specified in https://github.com/elastic/elasticsearch/blob/master/test/framework/src/main/java/org/elasticsearch/transport/AbstractSimpleTransportTestCase.java#L1926-L1934 does not hold, i.e. when the tcp socket backlog is full, my machine (Windows 10, Dell) upon receival of ack message is sending the syn-ack response instead of just ignoring the ack message (which is assumed by the test).
In context of this article http://veithen.io/2014/01/01/how-tcp-backlog-works-in-linux.html looks like the test assumes behavior 1, while on my machine it is behavior 2 or equivalent
Anybody having experience with this? Currently I am just uncommenting these tests before compilation, which is not really good approach. Maybe somebody knows how to adjust this behavior in java on windows 10. Otherwise we may consider adding some config to gradle to mark such tests and exclude them from gradlew check run (which is much less clean, but still better then uncommenting tests manually on some platforms.