[2.5.x] Upgrade Akka to version 2.4.4 #6050
Conversation
|
Hey @gmethvin, build is broken: https://travis-ci.org/playframework/playframework/jobs/123501044#L2551 If I recall correctly, this is the same test that broke when I've tried to upgrade Play 2.5.x to Akka 2.4.3 (not sure though). |
|
@marcospereira Yes, I think it is the same test. It passed on master because we removed WebSocket.acceptWithActor, which is what's being tested there. But it could be caused by a legit bug either in Play or in Akka. |
|
Actually, something might be broken on master as well. If I update that test to: "close when the consumer is done" in closeWhenTheConsumerIsDone { implicit app =>
import app.materializer
implicit val as = app.actorSystem
WebSocket.accept[String, String] { req =>
ActorFlow.actorRef { out =>
Props(new Actor() {
out ! Status.Success(())
def receive = PartialFunction.empty
})
}
}
}it still fails, so it has nothing to do with the deprecated methods. Those tests should probably have been updated on master to use |
|
Looks like its a timeout issue: https://travis-ci.org/playframework/playframework/jobs/123505529#L2597 |
This fixes playframework#5883 since Akka 2.4.3 contains akka/akka-core#20031.
| out => | ||
| Props(new Actor() { | ||
| out ! Status.Success(()) | ||
| app.actorSystem.scheduler.scheduleOnce(10.millis, out, Status.Success(())) |
There was a problem hiding this comment.
This fixes the test failure. I suspect the actor is being terminated before the client actually connects to it, so the client is never able to receive the close message. Adding a small delay gives the client enough time to connect. I'm not sure if this was supposed to work before, though.
There was a problem hiding this comment.
Curious to know how did you figured this out. :-)
There was a problem hiding this comment.
Well, it's clear the timeout is because the the client never gets the close frame. So one situation I could think of where that would happen is if the client wasn't connected yet, since the message is being sent when the actor is initialized (probably not a realistic situation anyway...).
Makes sense to me. Looks like ActorFlow is untested. |
Backports #5984 and #6048.