On Sun, 18 Sep 2022 11:52:28 GMT, Jaikiran Pai <j...@openjdk.org> wrote:
> Can I please get a review of this test only change which proposes to fix the > recent intermittent failures in `RmiBootstrapTest` reported in > https://bugs.openjdk.org/browse/JDK-8030616? > > The test has been intermittently failing with `cannot find a free port after > 10 tries`. The tests uses the `jdk.test.lib.Utils.getFreePort()` utility > method to get a free port to use in the tests. That port is then used to > create and bind a JMX connector server. The issue resides in the fact that > the `getFreePort` utility uses loopback address to identify a free port, > whereas the JMX connector server uses that identified port to bind to a > non-loopback address - there's logic in `sun.rmi.transport.tcp.TCPEndpoint` > line 117 which calls `InetAddress.getLocalHost()` which can and does return a > non-loopback address. This effectively means that the port that was > identified as free (on loopback) may not really be free on (some other > address) and a subsequent attempt to bind against it by the connector server > will end up with a `BindException`. > > Data collected in failures on the CI system has shown that this is indeed the > case and the port that was chosen (for loopback) as free was already used by > another process on a different address. The test additionally has logic to > attempt retries upto a maximum of 10 times to find a new free port on such > `BindException`. Turns out the next free port (on loopback) returned by > `jdk.test.lib.Utils.getFreePort()` is incremental and it so happens that the > systems where this failed had a process listening on a range of 10 to 12 > ports, so these too ended up with `BindException` when the JMX connector > server used that port for a different address. > > The commit here makes sure that the JMX connector server in the test is now > configured to loopback address as the address to bind to. That way the test > uses the correct address (loopback) on which the port was free. > > The commit also has a change to the javadoc of the test utility > `jdk.test.lib.Utils.getFreePort()` to clarify that it returns a free port on > loopback address. This now matches what the implementation of that method > does. > > Multiple test runs after this change hasn't yet run into the failure. Hello Serguei, Thank you for your review. I was about to integrate this when I just noticed that I had unintentionally included a new empty file in this commit. I've now updated this PR to remove that stray file (and no other changes). Could you please review the current state of this PR once more? ------------- PR: https://git.openjdk.org/jdk/pull/10322
