We have been seeing an issue where reads on a table seem to slowly degrade over time. When we run VACUUM ANALYZE on the table the operations perform much better. It usually takes about a week or two for the reads to degrade to a point where we have to manually run VACUUM ANALYZE. This has been happening to a handful of our tables. The mutation operations ran against these tables are mainly update operations. Very few insert and delete operations are ran against these tables. Also a jsonb object stores most of the data within each record. The size of the jsonb object can vary widely between records. We thought the autovacuum facility would have prevented us from having to periodically run VACUUM ANALYZE. Is this a bug or is it a configuration issue on our tables?