問題文
The nightly embedding refresh for the maintenance knowledge base has a narrow window, and about one call in twenty to the model endpoint fails transiently. The model definition already sets a retry count, but for this one job the team wants a higher number of retries without changing the behavior of the interactive lookups that use the same model. What lets them do that?
選択肢
- Raising the retry count in the model definition and lowering it again after the job finishes, because the setting is global to the model.
- Creating a second external model that points at the same endpoint with a higher retry count and granting the nightly job EXECUTE on that second definition only, because the retry count cannot be varied per call.
- Passing the retry count in the optional parameters of the generating call, because a value supplied there overrides the value defined in the external model at run time.
- Wrapping the call in a retry loop inside a scalar function, because the function boundary isolates the retry policy from other callers.