raboof/pekko
1
0
Fork 0
Code Issues Pull requests Projects Releases 6 Packages Wiki Activity Actions 10

Merge pull request #1473 from drewhk/wip-proper-sync-shutdown-remoting-drewhk

Browse source 
Shutdown of transports  are now synchronous with system shutdown #3368
This commit is contained in:
drewhk 2013-05-23 07:03:24 -07:00
commit beed693c1b
7 changed files with 40 additions and 39 deletions
Download patch file Download diff file Expand all files Collapse all files

35
akka-remote/src/main/scala/akka/remote/Remoting.scala
View file

@ -126,26 +126,23 @@ private[remote] class Remoting(_system: ExtendedActorSystem, _provider: RemoteAc
endpointManager match {
case Some(manager) ⇒
implicit val timeout = ShutdownTimeout
val stopped: Future[Boolean] = (manager ? ShutdownAndFlush).mapTo[Boolean]
def finalize(): Unit = {
eventPublisher.notifyListeners(RemotingShutdownEvent)
endpointManager = None
}
stopped.onComplete {
(manager ? ShutdownAndFlush).mapTo[Boolean].andThen {
case Success(flushSuccessful) ⇒
if (!flushSuccessful)
log.warning("Shutdown finished, but flushing timed out. Some messages might not have been sent. " +
log.warning("Shutdown finished, but flushing might not have been successful and some messages might have been dropped. " +
"Increase akka.remote.flush-wait-on-shutdown to a larger value to avoid this.")
finalize()
case Failure(e) ⇒
notifyError("Failure during shutdown of remoting.", e)
finalize()
}
stopped map { _ ⇒ () } // RARP needs only type Unit, not a boolean
} map { _ ⇒ () } // RARP needs only type Unit, not a boolean
case None ⇒
log.warning("Remoting is not running. Ignoring shutdown attempt.")
Future successful (())
@ -532,12 +529,20 @@ private[remote] class EndpointManager(conf: Config, log: LoggingAdapter) extends
endpoints.prune()
case ShutdownAndFlush ⇒
// Shutdown all endpoints and signal to sender when ready (and whether all endpoints were shut down gracefully)
val sys = context.system // Avoid closing over context
Future sequence endpoints.allEndpoints.map {
gracefulStop(_, settings.FlushWait, EndpointWriter.FlushAndStop)
} map { _.foldLeft(true) { _ && _ } } recover {
case _: AskTimeoutException ⇒ false
} pipeTo sender
def shutdownAll[T](resources: TraversableOnce[T])(shutdown: T ⇒ Future[Boolean]): Future[Boolean] = {
(Future sequence resources.map(shutdown(_))) map { _.foldLeft(true) { _ && _ } } recover {
case NonFatal(_) ⇒ false
}
}
(for {
// The construction of the future for shutdownStatus has to happen after the flushStatus future has been finished
// so that endpoints are shut down before transports.
flushStatus ← shutdownAll(endpoints.allEndpoints)(gracefulStop(_, settings.FlushWait, EndpointWriter.FlushAndStop))
shutdownStatus ← shutdownAll(transportMapping.values)(_.shutdown())
} yield flushStatus && shutdownStatus) pipeTo sender
// Ignore all other writes
context.become(flushing)
}
@ -629,12 +634,6 @@ private[remote] class EndpointManager(conf: Config, log: LoggingAdapter) extends
override def postStop(): Unit = {
pruneTimerCancellable.foreach { _.cancel() }
transportMapping.values foreach { transport ⇒
try transport.shutdown() catch {
case NonFatal(e) ⇒
log.error(e, s"Unable to shut down the underlying transport: [$transport]")
}
}
}
}