2) all reads from Mongo
3) in the background start data migration, one key/document at a time; transform Mongo document into RDBMS table/rows and insert... have mongo document id as primary key for main table... for every document migration/insert, lock the corresponding RDBMS row, this will ensure data consistency...
4) create code path to read from RDBMS.
5) test it in stage for different fail-case scenarios...
6) ideally, do canary type push to single server etc in prod to see if read from RDBMS seems fine (this assumes #3 is done)
actual steps may vary... but these are high-level steps....hope it helps...