How do we accept a new cluster?
A cluster that boots has passed one check. Acceptance asks whether the complete system can do the intended work.
Build acceptance evidence in layers
- Inventory and versions
Is the delivered configuration correct
- Devices and links
Do components pass the agreed checks
- Representative workload
Does useful work meet its targets
- Controlled recovery
Can the service recover as agreed
Test from small to large
Confirm inventory and supported versions. Check devices and links. Then test representative workloads and approved recovery scenarios.
Write the pass criteria before running the test: input, concurrency, duration, expected behavior, and measured result.
Burn-in is sustained activity that may expose problems. It does not certify every application or failure mode.
Keep the baseline
Save the accepted topology, versions, and measurements. They give you something useful to compare after a change.
Look at individual nodes and paths, not only the average. Include cold starts and agreed degraded conditions. Disruptive tests belong on authorized capacity with a reviewed plan and stop conditions.
Try it yourself
A fictional cluster meets average throughput, but one node is 30% slower. Would you accept it without investigation?
Check your answer
No. A synchronized job may wait for that node. Investigate against the agreed criteria and record any explicit risk decision. Do not rewrite success after seeing the result.