+ tl;dr - Ceph (Bluestore) (via Rook) on top of ZFS (ZFS on Linux) (via OpenEBS ZFS LocalPV) on top of Kubernetes. It’s as wasteful as it sounds – 200TPS on pgbench compared to ~1700TPS with lightly tuned ZFS and stock Postgres. The setup is at least usable and can get up to 1000TPS (2 replica ceph block pool) with synchronous_commit=off and some other less risky optimizations. Check out the numbers at the end in context and since this post is very Kubernetes heavy, if you don’t run Kub...
