Summary
Run Timelake on a production-like industrial dataset with parallel writers fed from a Kinesis stream, measure ingest, commit contention, latency and cost on real access patterns, and tune the defaults from what it shows. Was P9 of the performance plan, after the scan tuning and the remaining rollup work.
Plan
- Measure: ingest throughput, commit contention between parallel writers (spec 2), query latency, and the monthly bill (requests, storage, compute) on real data and access patterns.
- Then: tune settings and defaults for what they show, rather than for synthetic benchmarks only.
History
- 2026-10-18: planned (P9)
- 2026-10-19: was spec 44; a milestone on the roadmap since the specs were regrouped one per feature.