user1016739 wrote:If this is correct, there shouldn't be any sequence number collisions happening. How about posting a proof of that sequence numbers are overshooting the already existing PK values.
Here is an example of sequence definition, insert statement and what two sessions execute concurrently:
CREATE SEQUENCE <schema_name>.tab1_id_sq START WITH 1 increment by 1 nomaxvalue order cache 20;
INSERT INTO <schema_name>.tab1 (id) VALUES (<schema_name>.tab1_id_sq.nextval);
Let's say there are two sessions, S1 and S2. If S1 is running "INSERT INTO schema1.tab1 (id) VALUES (schema1.tab1_id_sq.nextval);", then S2 is running "INSERT INTO schema2.tab1 (id) VALUES (schema2.tab1_id_sq.nextval);", so they don't access the same table or the same sequence, but two tables and two sequences in different schemas.