Replication Resilience Test Report
ReplMon · faults injected on purpose, scored against targets
1. Executive Summary
2. Results Against Targets
ReplMon's scheduled jobs did the detecting and repairing, with no human in the loop. Bars show the time taken; the black tick is the target. Times from injection.
| Run | Fault | Subscriber | Injected (UTC) | Time to detect | Time to recover | vs previous run | Result |
|---|
3. Where the Time Went, Run by Run
One bar per run, all on the same scale, from injection to verified recovery. Under each: what was broken, the incident ReplMon raised, and the rows it repaired.
Waiting for detectionDetected → repairedRepaired → verified recovery
4. Test Guardrail
Chaos tests run in non-production, or in an agreed game-day window.
- Targets live in
ReplMon.dbo.ChaosTarget; change them to match your SLOs, not to make a run pass. - A failed run stays recorded as Fail, so a regression is visible on the next report.
- Rerun the full set after every cumulative update, schema change or new subscriber, and compare with the previous run.