Akka Typed Actor supervision configuration missing explicit restart strategy

Answer

Relying on the parent’s default stop strategy is not safe when a child may fail frequently; each failure leaves a stopped actor instance reachable until its mailbox is drained and the JVM garbage‑collects it, which can cause an unbounded growth of stopped actors under sustained load.

Confirmed facts

• In Akka Typed, a child created with Behaviors.setup inherits the parent’s supervision strategy unless explicitly wrapped with Behaviors.supervise(...).onFailure(SupervisorStrategy.restart).
• If the parent’s default is SupervisorStrategy.stop, each failure stops the child; the stopped instance remains in the system until its mailbox is processed and then reclaimed by GC.
• Applying an explicit restart strategy causes the failed child to be immediately restarted, preserving the same ActorRef and keeping the live actor count bounded to the number of configured children.

Steps to verify the bound

1. Create a parent supervisor with a known pool of typed children (e.g., using a router or a fixed number of Behaviors.setup children).
2. Wrap each child with Behaviors.supervise(...).onFailure(SupervisorStrategy.restart) (or leave it to inherit the stop strategy for a control run).
3. Inject failures at a steady rate (e.g., throw an exception in the child’s message handler) for a sustained period (e.g., 5 minutes at 100 failures/sec).
4. Monitor the number of live ActorRef instances via Akka’s ActorSystem.eventStream (subscribe to ActorRef events) or via JMX/Lightbend Telemetry.
5. Confirm that with the restart strategy the live‑actor count stays constant and does not show a monotonic increase; with the stop strategy observe a rising count of stopped actors until a GC cycle reduces it.