Redshift stored procs best practice for transactions locking tables and avoiding serialization isolation violations

I am facing occasional concurrency issues resulting in Serialization isolation violations where an sproc is invoked simultaneously for different tasks where the pseudo code for sp_main looks like -- BEGIN LOCK process_table ; -- has an IDENTITY event_id INSERT process_table VALUES(var_work, GETDATE()) ; SELECT INTO var_eid MAX(event_id) FROM process_table ; COMMIT ; END ; IF var_work = 'do_lots' THEN CALL sp_slow_proc ; -- Perform lots of slower statements that do not reference process_table ELSE CALL sp_quick_proc ; -- inserts new rows to process_table using a BEGIN LOCK COMMIT END construct END IF ; BEGIN LOCK process_table ; UPDATE process_table SET ended_at = GETDATE() WHERE event_id = var_eid ; COMMIT ; END ; It seems the slow running process always the one that chokes! I'm 99% confidant this can happen even if the fast process has finished (airflow job concluded having started after the slow process was commenced). Questions are: #1. is this an expected code layout ? #2. should the LOCK move outside the BEGIN END ? #3. when is a LOCK released - on COMMIT or ???