Flaky InitOnLineTest: initializeExportMultiSS times out, leaked replication server port fails initializeImport

[vharseko]

Observed failure

CI run on PR #831 (build-maven (ubuntu-latest, 25)): https://github.com/OpenIdentityPlatform/OpenDJ/actions/runs/30817690702/job/91699408826

Two failures in org.opends.server.replication.InitOnLineTest. All other 8 matrix jobs passed, and the PR only touches example code (opendj-ldap-sdk-examples, opendj-embedded-server-examples), so the failure is unrelated to the change:

[ERROR] Tests run: 10, Failures: 2, Errors: 0, Skipped: 0, Time elapsed: 638.9 s <<< FAILURE! -- in org.opends.server.replication.InitOnLineTest
[ERROR]   InitOnLineTest.initializeExportMultiSS:1132->waitForInitializeTargetMsg:1080 » ThreadTimeout Method org.opends.server.replication.InitOnLineTest.initializeExportMultiSS() didn't finish within the time-out 600000
[ERROR]   InitOnLineTest.initializeImport:550->createReplicationServer:493 » Config Replication Server failed to start : could not bind to the listen port : 65523. Error : BindException(Address already in use) Another socket is listening on 127.0.0.1:65523

initializeExportMultiSS hung for the full 600 s TestNG timeout, blocked in a socket read waiting for an InitializeTargetMsg that never arrived:

at org.opends.server.replication.protocol.Session.read(Session.java:437)
at org.opends.server.replication.protocol.Session.receive(Session.java:403)
at org.opends.server.replication.service.ReplicationBroker.receive(ReplicationBroker.java:2457)
at org.opends.server.replication.service.ReplicationBroker.receive(ReplicationBroker.java:2410)
at org.opends.server.replication.InitOnLineTest.waitForInitializeTargetMsg(InitOnLineTest.java:1080)
at org.opends.server.replication.InitOnLineTest.initializeExportMultiSS(InitOnLineTest.java:1132)

Analysis

The second failure is a cascade of the first, not an independent problem:

1. initializeExportMultiSS started RS(8) listening on port 65523 (msgID=204 ... started listening for new connections on address 0.0.0.0 port 65523), then hung waiting for the initialize-target message.
2. On timeout, TestNG killed the test thread, so the test's cleanup never ran and RS(8) kept the port bound.
3. The next test in the class, initializeImport, tried to create its replication server on the same port 65523 and failed with BindException(Address already in use).

So there are two things to address:

• Root flakiness: waitForInitializeTargetMsg can wait forever in ReplicationBroker.receive when the expected message is lost, until the 600 s method timeout kills the whole test.
• Missing cleanup on timeout: a timed-out test method leaves its replication servers running, poisoning the shared port for the remaining tests in the class (same family of problems as Replication tests pin their listen ports for the whole class and rebind them hundreds of times #800 / [#800] Allocate the replication test listen ports per invocation #803, but here the port is leaked within one class rather than across invocations).

Related: #800, #803 (replication test listen port allocation).