Summary
Refactor and simplify the runtime statistics across the pool and scheduler logic:
- Eliminate the generic
RuntimeStatistic abstraction.
- Rename fields like
statistic and last_updated to accurately reflect their meaning and usage.
- Retain only what’s needed to track average batch processing time per worker.
- Update the batch scheduler to align synchronization using this simplified model.
Background
This is a follow-up to #196. The current implementation carries abstraction and naming baggage. The goal is to clean up the statistics logic, make the code more intuitive and easier to maintain, and clarify the intended use of the average batch time to align synchronization across workers.