RFR: 8375352: java/net/httpclient/ConnectTimeoutWithProxy*.java tests fail on EC2 [v2]

Volkan Yazici vyazici at openjdk.org
Thu Jan 29 12:26:31 UTC 2026


On Thu, 29 Jan 2026 12:05:59 GMT, Daniel Fuchs <dfuchs at openjdk.org> wrote:

> It's a bit unfortunate that we don't have a means to test this reliably on windows. Do we really need to allow for backlog + 512 sockets? Can we go with backlog + 8 instead?

`backlog + 2` already works in all supported platforms. 512 is a hard limit to give up on exhausting the admission queue. In practice, max. 2 extra sockets do the job.

> If we're now using the loopback there should no longer be any "No route to host" or "Unresolved address" exceptions, should it?

I'd guess so — hard to say for all platforms.

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

PR Comment: https://git.openjdk.org/jdk/pull/29431#issuecomment-3817318841


More information about the net-dev mailing list