Concurrent fallback collectors all listen for ovos.skills.fallback.pong on the same bus topic. Under load, one collector can consume a pong that belongs to another fallback request, which can make the real request time out or select the wrong fallback state.
What I need here:
- stamp each fallback ping with a request id
- ignore pongs for other request ids
- keep legacy pongs without an id working
Concurrent fallback collectors all listen for
ovos.skills.fallback.pongon the same bus topic. Under load, one collector can consume a pong that belongs to another fallback request, which can make the real request time out or select the wrong fallback state.What I need here: