Summary
Create and tune a “QS‑only” availability workload that shows clear benefit without RL

Description
We want a clean, reproducible demonstration that Query Sentinel by itself can improve availability and/or latency under sustained long‑running queries, with RL disabled. This becomes our canonical “QS‑only helps” example.

Scope

1. Workload + config design

• • Use or adapt the many_long_queries availability workload.
  • Tune:
    • QS config (e.g., thresholds, max_operations_to_kill, tick interval).
    • Workload parameters (concurrency, query mix, duration) as needed.

1. Experiments

• • Baseline:
    • QS disabled or dry run.
    • RL off.
  • Treatment:
    • QS enabled (wet run).
    • RL off.
  • For each run, capture:
    • Availability / canary metrics.
    • Latency for short operations.
    • QS metrics (killed_operations, ticks_under_overload, etc.).