Death Watch

​

API

• pid.Watch: Register to receive Terminated when cid stops.
• pid.UnWatch: Cancel the watch for cid.
• ctx.Watch: ctx.Self().Watch(cid).
• ctx.UnWatch: ctx.Self().UnWatch(cid).

​

Terminated message

// ActorPath returns the path of the terminated actor; TerminatedAt the time it stopped.
func (x *Terminated) ActorPath() Path
func (x *Terminated) TerminatedAt() time.Time

case *actor.Terminated:
    name := msg.ActorPath().Name()
    // remove from local state, restart a replacement, etc.

• Terminated is sent after the stopped actor’s PostStop has returned, and after its name is free, so a watcher may spawn the same name again as soon as it receives the message.
• It is a control message: in the watcher it is handled ahead of the user messages still queued, and a watcher that is suspended at that moment still receives it.
• A restart is not a death: restarting a watched actor sends no Terminated, and the watch stays in place.
• Watching an actor that has already stopped registers nothing, and no Terminated ever comes.
• A parent watches each of its children automatically.

​

When to use

• Dependencies: Watch a child or peer; when it stops, recreate or fail over.
• Resource cleanup: Release references to the stopped actor.
• Supervision: Custom logic when a watched actor dies.

​

Automatic unwatch

​

Example

func (a *ParentActor) Receive(ctx *ReceiveContext) {
    switch msg := ctx.Message().(type) {
    case *SpawnWorker:
        pid := ctx.Spawn("worker", NewWorkerActor())
        ctx.Watch(pid)
        a.workers[pid.ID()] = pid
    case *actor.Terminated:
        delete(a.workers, msg.ActorPath().String())
        // optionally respawn: the name is free by now
    default:
        ctx.Unhandled()
    }
}