Within the catalog cache we retry on a couple of different error types which aren't usually considered retriable and we don't retry on errors which are in the "retriable" error category. Based on recent discussion, it seems that it isn't very well defined what transient errors a caller of the catalog cache (or the filtering refresh) API might encounter.
There may be use cases (as in SERVER-107952) in which the caller of a refresh might want to retry on a failed refresh. Right now, it is not at all obvious what errors that would include.

This investigation is to determine what types of errors are actually considered transient for a catalog cache refresh and perhaps to provide a way in the code to identify those errors.