A new model version must replace an old one in production. What is the safest approach?
Why A is the answer
Evaluate it, roll out gradually, and keep a way back
Evaluate, then shift traffic gradually, with a rollback path.
In detail
- Evaluate first, roll out gradually, and keep a way back.
- Delete the old version last.
Use case
An illustrative scenario, not a real incident.
A new model version is evaluated, rolled out in stages, and the old one is kept until the new one proves itself.
See it in different ways
flowchart LR
A["Blue: current"] -->|"90%"| C["Traffic"]
B["Green: new"] -->|"10%"| C
C --> D{"Healthy?"}
D -->|Yes| E["Increase green"]
D -->|No| F["Back to blue"]sequenceDiagram
participant E as Evaluation
participant N as New deployment
participant O as Old deployment
participant T as Traffic
E->>N: Passes checks
T->>N: Small share
T->>O: The rest
alt Metrics hold
T->>N: Increase the share
else Metrics fall
T->>O: Roll back
endflowchart LR S0["Evaluate"] --> S1["Gradual rollout"] --> S2["Monitor"] --> S3["Retire the old version"]
Key point: Evaluate, roll out gradually, keep a way back.
Why the other options are wrong
- B. Delete the old deployment first to save costs
- Deleting first removes your way back.
- C. Update it on Friday so that traffic is lower
- A quiet day is not a safety plan.
- D. Swap it directly and watch for complaints
- A direct swap risks everyone at once.