I'm curious about whether using Postgres native partitioning was considered? If so, I'd like to hear the rationale for not doing it. It seems like you made the

I'm curious about whether using Postgres native partitioning was considered? If so, I'd like to hear the rationale for not doing it. It seems like you made the database sharding decision as a result of vacuum taking too long. Partitioning the problematic table (assuming you're storing "blocks" in 1 giant table) would have enabled per-partition vacuuming, which will avoid long running vacuum processes.