問題文
A design must choose a checkpoint store for workflows that run across multiple instances behind a load balancer. Which store is appropriate?
選択肢
- The model deployment's own conversation store
- A shared external database such as Azure Cosmos DB, so any instance can read the state written by another instance
- The local file system of each instance
- An in-memory store, since it has the lowest latency and the load balancer can be configured to send every request in a session back to the same instance, so the state is always found where it was written