Good news, the regressions seen in the Insert Benchmark for Postgres versions 14.x and 15beta1 have been fixed. Just in time to free up my home test cluster for 15beta2.
smalldatum.blogspot.com/2022/06/insert
1
10
Conversation
Replying to
Thanks! It is fairly common practice to defensively run VACUUM manually after bulk loading + index builds, right before the start of the benchmark proper. This avoids inter-run noise from autovacuums that started before (but also ran during) benchmark, to varying degrees.
Replying to
I learned that lesson early on with PG and the insert benchmark. Blocking vacuum is run before the first read+write benchmark step.
1