Yes. This is likely not the typical workload. Our PG database is only used in research, in "burst" situations - e.g. big batch jobs written to the DB (2 days 400 Million tuples) and read in big chunks (e.g. 400 Million tuples exported in 2 hours to CSV/Postgres FDW). ZFS is on spinning Rust (Sata 6GB), 6x8TB drives in a Raidz2 pool, the ZFS dataset is both compressed and encrypted. In Proxmox, I do not see I/O in any way limited during these burst writes/reads, the bottleneck is the CPU. However, the CPU was the bottleneck also before ZFS, so I cannot say how much impact the compression/encryption has. Other ZFS parameters are default (eg. filesystem recordsize is 128KB - lower values will yield better read speed, but less compression, and we were aiming for a lot of compression).

Likely no impact, but we use Postgres in Docker on unprivileged LXC, the `/data` directory is mounted from the LXC from the host ZFS pool. Since LXC runs all processes on the host, the performance impact is negligible (unlike, e.g. running this in a full VM).

I have the _feeling_ (not actually tested) that the Postgres database is faster with ZFS, since less data needs to be read, especially since we have a lot of sequential scans.

Just looked it up [1]

> LZ4 is lossless compression algorithm, providing compression speed > 500 MB/s per core (>0.15 Bytes/cycle). It features an extremely fast decoder, with speed in multiple GB/s per core (~1 Byte/cycle).

[1]: https://lz4.github.io/lz4/