MonoToListenableFutureAdapter.cancel problems [SPR-17336]

[spring-projects-issues]

Ivan Sopov opened SPR-17336 and commented

org.springframework.messaging.support.MonoToListenableFutureAdapter.cancel method is strange.
Consider adapter for the case that cannot be cancelled.

Future<Void> foo = new MonoToListenableFutureAdapter<>(Mono.empty());
System.out.println(foo.cancel(true));
System.out.println(foo.isCancelled());

This prints:

true
false

This clearly violates cancel method javadoc "Subsequent calls to isCancelled will always return true if this method returned true" And I believe cancel should've returned false itself since it should "return false if the task could not be cancelled, typically because it has already completed normally"

Also curent implementation gives result true several times under race if actual cancellation is happening. Consider the following JCStress-based test:

@JCStressTest
@Outcome(id = "true, false", expect = Expect.ACCEPTABLE)
@Outcome(id = "false, true", expect = Expect.ACCEPTABLE)
@Outcome(expect = Expect.FORBIDDEN, desc = "Other cases are forbidden.")
@State
public class MonoToListenableFutureAdapterCancelTest {
    private final Future<Void> future = new MonoToListenableFutureAdapter<>(Mono.fromFuture(new CompletableFuture<>()));
    @Actor
    public void cancel1(ZZ_Result r) {
        r.r1 = future.cancel(true);
    }
    @Actor
    public void cancel2(ZZ_Result r) {
        r.r2 = future.cancel(true);
    }
}

It fails with "true, true" result while result "false, true" and "true, false" are also observed.

---

Affects: 5.1 GA

Issue Links:

• Support use of reactive clients in @MessageMapping methods [SPR-16634] #21175 Support use of reactive clients in @MessageMapping methods
• DelegatingCompletableFuture.cancel() returns false for already cancelled future [SPR-17337] #21871 DelegatingCompletableFuture.cancel() returns false for already cancelled future

Referenced from: commits c01f350, 928c541, bef22ec

[spring-projects-issues]

Rossen Stoyanchev commented

This type was mostly for internal use, within the ReactorNettyTcpClient until it was made public recently with #21175. Indeed it should not return a hardcoded "true" value so I'm scheduling this for a change.

Can you clarify how you're using and ran into this? Directly or perhaps indirectly via ReactorNettyTcpClient or an @MessageMapping method returning a Mono?

[spring-projects-issues]

Ivan Sopov commented

Sorry - I'm not actually using it. Found it mentioned here - https://docs.spring.io/spring-framework/docs/current/javadoc-api/org/springframework/util/concurrent/ListenableFuture.html

[spring-projects-issues]

Rossen Stoyanchev commented

Ok no worries, thanks.

[spring-projects-issues]

Rossen Stoyanchev commented

Note that this class is now in spring-core. It became public only very recently in 5.1 GA, but remained in spring-messaging. We're fixing this in 5.1.1 by putting it where it belongs in org.springframework.util.concurrent.