failed: StreamRefsSpec

[He-Pin]

[09-01 15:20:20.134] [info] - must pass cancellation upstream across remoting before elements has been emitted *** FAILED *** (26 milliseconds)
[09-01 15:20:20.135] [info]   java.lang.AssertionError: assertion failed: expected Done, found Failure(org.apache.pekko.stream.RemoteStreamRefActorTerminatedException: Remote target receiver of data Some(Actor[pekko://StreamRefsSpec/system/Materializers/StreamSupervisor-2568/$$m-SourceRef-8#-1525367045]) terminated. Local stream terminating, message loss (on remote side) may have happened.)
[09-01 15:20:20.135] [info]   at scala.Predef$.assert(Predef.scala:279)

https://github.com/apache/incubator-pekko/actions/runs/6050882412/job/16421131496?pr=609

[He-Pin]

https://github.com/apache/incubator-pekko/actions/runs/6229433326/job/16907923622#step:6:21256

[He-Pin]

https://github.com/apache/incubator-pekko/actions/runs/6293248461/job/17083517089#step:6:22009

[He-Pin]

https://github.com/apache/incubator-pekko/actions/runs/6312975683/job/17140088676#step:7:13652

[He-Pin]

Root-cause analysis: this is the test marked with an explicit in-code // FIXME https://github.com/akka/akka/issues/30844 (StreamRefsSpec.scala:361, just above the failing case "pass cancellation upstream across remoting before elements has been emitted").

When cancellation is propagated across remoting before any element has been emitted, the KillSwitches.shutdown() races the remote SourceRef actor termination: the local side can observe RemoteStreamRefActorTerminatedException ("message loss may have happened") instead of a clean Done. This is a known upstream product race in stream-ref cancellation propagation that Akka itself left unfixed (akka#30844).

Making the test tolerate both Done and the termination exception would only mask the deliberately-flagged product bug, so a test-only tweak is not appropriate here. The real fix belongs in the stream-ref cancellation path. Leaving open and linked to the FIXME.