Hi Jing,
Thanks for the input, that's exactly how I have imagined the standard
deployment process.
With regards to the Exactly-Once guarantee as per my understanding:
Current Exactly-Once semantics are performed via Two Phase Commit that is done
at both Source and Sink (plus any checkpoint ena
Hi Kevin,
Thanks for your proposal. It is quite interesting. We had a similar
solution back to the Apache Storm era many years ago. The job deployment
with operators itself could be considered as physical distributed task
containers. There will be a management tool, like a control plane, to
deploy
