kanaloa
kanaloa copied to clipboard
Make your service more resilient by providing protection against traffic oversaturation
Needed for downstreams to use 2.12.x Please advice on things to change and version bump.
In some cases load shedding might be more useful than back pressure. We need to implement this based on proved algo.
basically, it should be able to recover from unexpected children death - queueProcessor, queue, metricsCollector, authThrottler, etc To improve metrics collector it needs to be an actor under the dispatcher...
This interval if set below than the updateInterval will basically disable the auththrottle. It should probably just be controlled by the updateInterval so that its action cycle can be in...
It returns a Future that completes on routee ready (i.e. the remote actor ref is available in the cluster)
something like `send[TRequest, TResponse](r: TRequest): Future[TResponse]`, so that the akka is hidden from user.
1. Start from large pool size, and let kanaloa walk down. Less risk of over throttling, but higher risk of backend unprotected from overflow and high latency 2. Start from...
Shouldn't allow invalid config values blow at random places
Sometimes an auto down-size could be handy if we want to keep the autothrottle in effect. (too many workers basically disables autothrottle, and the default downsize time is 72 hour...