Postgresql database implements MVCC by creating a new row version for any update/delete/or insert. While this is a sound implementation to support concurrent transactions reading old version of the rows, it does have some side effects and this is what I want to discuss in this episode of the backend engineering show.

0:00 Intro

1:00 Postgres MVCC design

6:42 MVCC on other databases

11:15 Side-effects of Row-versioning

17:00 Postgres HOT optimization

19:50 How Index bloat affects Performance

24:20 How Postgres 14 Addresses index bloat

Cards:

14:12 b-tree https://www.youtube.com/watch?v=a1Z40OC553Y

Resources

https://www.postgresql.org/docs/14/btree-implementation.html#BTREE-DELETION