問題文
A solution must resume long-running workflows after an instance is replaced, keep per-user context isolated between tenants, and let other systems query the current task status. Which state design satisfies all three?
選択肢
- One shared store for all three purposes with a single retention policy
- Workflow state in a shared external store, tenant-partitioned long-term memory, and structured task status outside the conversation
- Workflow state in memory on each instance, with tenant isolation enforced by the prompt and task status published from the same in-memory structure
- All state in the conversation history, with compaction to control its size