Architecture Design Questions

Scenario: Microservice IP keeps changing.

Fix: Client-side (Netflix Eureka) vs Server-side (AWS ALB) discovery. Consul/Zookeeper for health checks.

Scenario: One microservice is down, others are waiting and crashing.

Fix: Resilience4j/Hystrix. 3 states: Closed (normal), Open (fail fast), Half-Open (test recovery).

Scenario: One service's thread pool is full, the whole app hangs.

Fix: Isolate resources (threads/connections) like ship bulkheads. If one sinks, the rest survive.

Note: For more insights, read about MLOps & Model Deployment.

Scenario: API fails due to a network blip.

Fix: Retry, but double the delay each time (1s, 2s, 4s). Add Jitter so everyone does not retry at the exact same time (Thundering herd).

Scenario: Need to change the API without breaking old clients.

Fix: URI versioning (/v1/users), Header versioning (Accept: v1), or Query param.

Scenario: Making a choice.

Fix: REST = Simple, public APIs. GraphQL = Flexible queries, mobile apps (fixes over-fetching). gRPC = Internal microservices (Protobuf, HTTP/2, very fast).