A way is needed to distinguish expected and unexpected read-not-found counts.
There are situations where an application needs to determine whether a particular key exists and get the record if it does, or discover that the key does not exist (after which it may write a new record, for example); neither situation is an error.
In other situations, record-not-found indicates a true problem which needs to be identified and addressed.
Adding both expected and unexpected fails to client_read_not_found makes it difficult to distinguish the two and thus difficult to identify and rectify problems.
I proposed that a new flag be added to the ‘get’ call. When the key is not found, suppress incrementing client_read_not_found or, alternatively, increment a new flag (something like ‘client_read_success_not_found’).