Akka.Persistence.Query.TCK Bug: TestActor acknowledges delete whether it actually happened or not

[Aaronontheweb]

Version: v1.4.0-beta2

Found this while investigating akkadotnet/Akka.Persistence.MongoDB#77 (comment)

Issue is this:

Error in this case appears to be a stale read coming from the MongoDB journal:

akka.net/src/core/Akka.Persistence.TCK/Query/CurrentEventsByPersistenceIdSpec.cs

Lines 113 to 127 in 4f984bc

	[Fact]
	public virtual void ReadJournal_CurrentEventsByPersistenceId_should_return_remaining_values_after_partial_journal_cleanup()
	{
	var queries = ReadJournal.AsInstanceOf<ICurrentEventsByPersistenceIdQuery>();
	var pref = Setup("h");
	pref.Tell(new TestActor.DeleteCommand(2));
	AwaitAssert(() => ExpectMsg("2-deleted"));
	var src = queries.CurrentEventsByPersistenceId("h", 0L, long.MaxValue);
	src.Select(x => x.Event).RunWith(this.SinkProbe<object>(), Materializer)
	.Request(1)
	.ExpectNext("h-3")
	.ExpectComplete();
	}

Setup writes 3 events to the journal - arrange phase of the test deletes the first two - assertion expects only the third one back. Instead we got the first event straight away.

Sounds to me like Mongo being Mongo, but I'll take a look real fast.

And determined cause:

Ah, the problem is actually with the TestActor in the Akka.Persistence.Query.TCK

akka.net/src/core/Akka.Persistence.TCK/Query/TestActor.cs

Lines 44 to 47 in 4f984bc

	case DeleteCommand delete:
	DeleteMessages(delete.ToSequenceNr);
	Sender.Tell($"{delete.ToSequenceNr}-deleted");
	break;

We send a reply back that the message has been deleted without actually knowing if the journal has done that or not.

Proposed fix: Probably need to wait until we get a DeleteMessagesSuccess or a DeleteMessagesFailure message back, since DeleteMessages(long seqNo) is an asynchronous operation.