For databricks connector DELTA tables, currently fivetran performs 4 steps
between step 1 and step 4 there may be several minutes and the historical table is in a transient state. if any queries hit the table during this period, the results will be incomplete.
I believe these can be combined into a single transactional MERGE statement (I'm not sure why there are delete/write steps). This is for Databricks DELTA tables.