Consider tuning autovacuum to make it less aggressive
According to https://gitlab.com/gitlab-com/infrastructure/issues/4650 autovacuum is quite aggressive:
- autovacuum is consuming 75% of all the read I/O! That means that approximately, autovacuum is re-reading the whole db every 3h!
- It is consuming approximately 0,5TB each hour.
The scale factors are set to 0.5%, which might be a very low value. Combined with other autovacuum configuration factors, they are driving read I/O too high. Proposal is to tune autovacuum to significantly reduce its I/O, while assuring bloat is still controlled.