←back to thread

Pg_parquet: An extension to connect Postgres and parquet

(www.crunchydata.com)
157 points craigkerstiens | 2 comments | 17 Oct 24 16:18 UTC | HN request time: 0.618s | source
Show context
linuxhansl ◴[17 Oct 24 20:55 UTC] No.41873697[source]▶
Parquet itself is actually not that interesting. It should be able to read (and even write) Iceberg tables.

Also, how does it compare to pg_duckdb (which adds DuckDB execution to Postgres including reading parquet and Iceberg), or duck_fdw (which wraps a DuckDB database, which can be in memory and only pass-through Iceberg/Parquet tables)?

replies(3): >>41874044 #>>41874177 #>>41876793 #
fulafel ◴[18 Oct 24 06:21 UTC] No.41876793[source]▶
Having the famously crashy DuckDB share a process and memory with PostgreSQL doesn't seem like the most robust setup.
replies(2): >>41876976 #>>41880138 #
1. memhole ◴[18 Oct 24 15:09 UTC] No.41880138[source]▶
Famously crashy? Any incidents you can share? I’ve only had good experiences is why I ask.
replies(1): >>41885827 #
2. fulafel ◴[19 Oct 24 05:25 UTC] No.41885827[source]▶
I haven't used it, but have gone through their issue tracker and seen a lot of comments to this effect here and on other internets places. (Unverifiable: also some hearsay from colleagues)

Not saying they're doing it wrong, it just seems they have some different stability vs performance tradeoffs than PG.