In one of our applications, we have a massive Materialized View that refreshes three times a day, and takes seven hours to refresh. (Not ideal, I know). This perplexed me, because I surely thought that users and sessions could not access this materialized view while it was being refreshed, but apparently they can!. (The type of refresh is a complete refresh)
During a complete refresh, to my understanding, the existing dataset is dropped and the query is then re-executed. If this is true, then how are users/other sessions able to access the materialized view while the materialized view is being refreshed?
There are two different ways that a complete refresh can happen-- an atomic refresh or a non-atomic refresh. An atomic refresh simply issues a DELETE to delete all the rows in the materialized view and then does an INSERT to insert the new data. This is all within a single transaction so Oracle's standard multi-version read consistency architecture lets Oracle show other sessions the old data until the refresh completes. In a non-atomic refresh, Oracle does a TRUNCATE on the materialized view and then a direct-path INSERT to insert the new data. This is substantially more efficient but since TRUNCATE is DDL, it means that the old data is not visible to other sessions during the refresh.