Analytics in Postgres –A decade in the making
summary
Postgres gains columnar scans by embedding a vectorized engine rather than reimplementing one: pg_mooncake provides columnstore tables backed by DuckDB’s execution engine and reaches top 10 ClickBench performance. The build uses Arrow/Parquet-like on-disk formats, logical replication or WAL decoding to keep row tables synced, and object-store-native storage for scale. Remaining issues include Postgres feature compatibility, update invalidation of column caches, and CDC deduplication when users still replicate to systems like ClickHouse.
AI-written summary · about
chapters (11)
- 0:00Analytics in PostgreSQL
- 1:26PostgreSQL data stack limits
- 2:56PostgreSQL analytics history
- 4:23Defining PostgreSQL analytics
- 7:58Columnar storage and vectorized execution
- 11:53Columnar integration challenges
- 13:39Row-store to columnar replication
- 16:10Building analytics from scratch
- 19:41Future beyond analytics
- 20:01PG Moonake and closing
- 20:30Q&A
related talks
49mCan PostgreSQL Compete with Analytics Specialists?
34mAnalytics and AI with PostgreSQL - Mission Possible!
21mWhat can Postgres learn from DuckDB?
19mFrom OLTP to OLAP: Is PostgreSQL Eating Analytics Too?
34mAnalytics and AI with PostgreSQL - Mission Possible!
27mBuilding a PostgreSQL data warehouse
24mpg_duckdb: Ducking awesome analytics in Postgres
28mColumnstore Revolution