Fix flaky NettyClientAlpnTest by retrying on port-bind race conditions

[joviegas]

Motivation and Context

NettyClientAlpnTest intermittently fails with IOException: Failed to bind to 0.0.0.0:<port>.

Root cause: BaseMockServer.getUnusedPort() opens a ServerSocket(0) to get a free port number, then closes the socket before the port is actually used. Between that close and the later server.start() bind, the port is unowned and can be taken by another process on the CI host (example an outbound connection's ephemeral source port). When that happens, Jetty's bind fails and the test errors out, unrelated to any ALPN/HTTP2 code being tested.

Modifications

• MockH2Server.java: connectors now bind to port 0, so the kernel assigns a free port atomically at bind time and the allocate-then-free race is gone. start() reads the actual ports back via connector.getLocalPort().
• BaseMockServer.java: removed getUnusedPort() and the constructor that pre-allocated ports; it now only holds the port fields and URI helpers.
• NettyClientAlpnTest.java: no functional change, tests stay plain @Test.

Testing

Verified by forcing a real port collision: retry fired for 3 attempts as configured, then passed once the port was free.

License

• I confirm that this pull request can be released under the Apache 2 license

[dagnir]

Root cause: BaseMockServer.getUnusedPort() opens a ServerSocket(0) to get a free port number, then closes the socket before the port is actually used. Between that close and the later server.start() bind, the port is unowned and can be taken by another process on the CI host (example an outbound connection's ephemeral source port). When that happens, Jetty's bind fails and the test errors out, unrelated to any ALPN/HTTP2 code being tested.

Weird. Can we just fix the server so that it doesn't do this weird allocate then free dance? Forcing every test that uses BaseMockServer to retry doesn't seem very scalable

[joviegas]

Root cause: BaseMockServer.getUnusedPort() opens a ServerSocket(0) to get a free port number, then closes the socket before the port is actually used. Between that close and the later server.start() bind, the port is unowned and can be taken by another process on the CI host (example an outbound connection's ephemeral source port). When that happens, Jetty's bind fails and the test errors out, unrelated to any ALPN/HTTP2 code being tested.

Weird. Can we just fix the server so that it doesn't do this weird allocate then free dance? Forcing every test that uses BaseMockServer to retry doesn't seem very scalable

Done, added the fix to get the port right before the start.
Initially I did that fix but moved to retry since this involved 1 file change and not touching the BaseMockServer.

[github-actions]

This pull request has been closed and the conversation has been locked. Comments on closed PRs are hard for our team to see. If you need more assistance, please open a new issue that references this one.