問題文
One of an agent's five tools depends on a third-party service that has begun failing for minutes at a time. During those windows every request that touches the tool waits for the full timeout, and the agent's overall latency collapses. Which measure addresses the latency collapse?
選択肢
- Increase the timeout so that slow responses eventually succeed.
- Move the tool call into a background job and return a placeholder answer to the user immediately, so that the user-facing latency is unaffected and the real answer can be delivered later when the third-party service recovers.
- Stop calling the failing tool for a short period after repeated failures and fail that path fast, then probe before resuming normal calls.
- Retry the failing call much more aggressively so that one of the attempts lands during a brief moment when the third-party service is healthy.