PostgreSQL 10 High Performance - Third Edition

autovacuum is too disruptive

The opposite case to the previous is also possible: autovacuum can easily consume too much I/O capacity on a server without a lot of it to spare. In that case, increases of autovacuum_vacuum_cost_delay to as high as 100 ms are common. This is the preferred way to deal with the situation where VACUUM seems to be consuming too many resources. Increasing the cost delay to make autovacuum's actions less disruptive is the way you should be dealing with the situation where it seems you can't run VACUUM during your busy times.