Rollout Strategies: Shadow, Canary, Blue-Green, A/B

Four patterns, four risk profiles, four feedback signals.

0/1 done

Patterns and their signals

Four patterns

Blue-Green — two identical environments; switch traffic in one move. Cheap, simple, risky.
Canary — gradually shift % of traffic; rollback on bad signal. Best general-purpose pattern.
Shadow — candidate receives a copy of traffic, responses discarded. Zero user risk, no business signal.
A/B (Experiment) — 50/50 random split, statistical comparison on a business metric.

What signal you get

Pattern	Tells you about	Doesn't tell you about
Blue-Green	Errors / latency under full load	Slow degradations
Canary	Errors / latency on small slice	Behaviour on tail traffic
Shadow	Inference parity, perf	Business impact
A/B	Business outcome	Performance at full scale

Most teams chain them: shadow → canary → A/B → full.

Analogy

Theatre analogy: shadow is rehearsal (full performance, empty house), canary is the preview night (small audience), A/B is two simultaneous productions in different cities, blue-green is opening night — and the curtain has only one position.

Reflect

Sketch the rollout for your next model.

›Which patterns will you chain, and in what order?
›What rollback signal aborts the canary?
›Who watches the dashboard during the canary window?

Reading in progress · 0 of 1 activity done