Introduce a cache in Mixer to elide calls into adapters for CHECK variety adapters. We can cache evaluation results based on the input instances given to the adapter. This supports need to be configurable per adapter (on/off, cache size, TTL).
Additionally, or instead of, we could put a cache higher up in the pipeline that's based on input attributes. Based on the set of incoming attributes to Mixer, we can promptly respond to Check calls.
We should measure the effectiveness of either approach and decide if either or both are desirable.
Introduce a cache in Mixer to elide calls into adapters for CHECK variety adapters. We can cache evaluation results based on the input instances given to the adapter. This supports need to be configurable per adapter (on/off, cache size, TTL).
Additionally, or instead of, we could put a cache higher up in the pipeline that's based on input attributes. Based on the set of incoming attributes to Mixer, we can promptly respond to Check calls.
We should measure the effectiveness of either approach and decide if either or both are desirable.