DanielLeens commented on issue #12087: URL: https://github.com/apache/seatunnel/issues/12087#issuecomment-5541702076
Thanks @JeremyXin for turning the broader autoscaling discussion in #11663 into a phased STIP. I checked current dev and the open PR set: Zeta has the scheduling and telemetry building blocks, but there is no AutoScaler implementation or active autoscaling PR. This should therefore remain a design-first proposal before Phase 1 code is started. Please cross-link #11663 as the umbrella and make these Phase 1 contracts normative in this STIP: recommendation identity must include masterEpoch and a monotonic generation, stale generations must be rejected, sample freshness and clock-skew tolerance must be defined, stabilization windows must use an unambiguous time model, and the REST/metric output must expose the contributing signals and blocking conditions. Recommendations should be idempotent and must not start cooldown until an external actuator reports convergence. Phase 1 must remain advisory-only: no worker membership mutation, drain, or running-job rescale. The first tests should cover active-master handover, stale or duplicate recommendations, missing metrics, and restart recovery. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
