DEPLOYMENT ARCHITECTURES

CHECK YOUR UNDERSTANDING • Why separate who hosts the harness from who hosts the model? ◦ Answer: They are independent decisions with different trade-offs in control, cost, privacy, and effort. • What does a queue do for scaling? ◦ Answer: It holds work until a worker is free, so bursts of requests are not lost. • Why keep separate secrets for each environment? ◦ Answer: So a test key can never touch real customer data. • What triggers an automatic rollback? ◦ Answer: Pre-set conditions such as rising error rates or cost, which send traffic back to the old version. • Why is a gradual rollout safer than releasing a change to everyone at once? ◦ Answer: If something is wrong, only a small group is affected and the change can be stopped or reversed before it spreads.