SOP: Big-bang rollout vs phased/canary release

SOP: Big-bang rollout vs phased/canary release

Run this before pushing any change to all users at once.

Stage 1 — Score the blast radius (owner: release owner)
☐ How many users hit the change immediately?
☐ How fast can you detect a problem — minutes, or only via complaints?
☐ How reversible is it (config toggle vs data migration)?

Stage 2 — Choose
☐ Phased/canary: 1% → 10% → 50% → 100%, watching metrics at each gate. Use for anything user-facing or hard to reverse. Cost: slower, needs segmenting and monitoring.
☐ Big-bang: one shot to everyone. Only use for trivial, instantly reversible, or internal-only changes. Cost: a bad change hits 100% before you notice.

Stage 3 — Gates (owner: release owner)
☐ Define the metric and threshold that halts promotion.
☐ Name who can call the rollback and how.
☐ Done-when: each stage has a go/no-go check, not a vibe.

Trigger: any production release. Done-when: rollout shape and halt-metric set.

Save this. Run it every time.