I’m wondering if it’s more of a “Chaos Injector” component/service that reads configuration data from the Chaos Control Plane on what to target, with parameters on how/when to do so. That would make the arrow make sense in my mind given it sounds like that’s a solid pattern for scaling these data/control plane flows: https://aws.amazon.com/builders-library/avoiding-overload-in...
This. It's an internal system called ChAP, Chaos Automation Platform. It has the ability to target failure down to specific RPC calls in single instances, using platform components that services consume as the mechanism for doing that injection.