Data Warehouses & Analytical Databases

This is the OLAP-shaped sibling of Databases & ORMs — engines optimized for scans, aggregates, and joins over millions-to-billions of rows. For the lakehouse / table-format layer underneath, see Lakehouse Table Formats. For dashboards that read from these, see BI & Dashboards.

Embedded / single-node

• ★ DuckDB — in-process columnar SQL; reads Parquet / CSV / JSON / Arrow / Postgres / S3 directly; runs in browsers via DuckDB-Wasm. MIT. The default "I just want analytics SQL on a file" pick in 2026; pairs with Polars for Python ergonomics.
• chDB — embedded ClickHouse in a single binary / Python module; same engine as ClickHouse server. Apache 2.0.
• SQLite + extensions — sqlite-vss, FTS5, sqlean get you surprisingly far for small analytical workloads on existing app DBs.

Columnar OLAP (clustered)

• ★ ClickHouse — fastest open-source columnar DB for most analytical workloads; Apache 2.0 (the OSS server is permissive — note ClickHouse Inc. paid Cloud and the ClickHouse Cloud terms are separate). Excellent for events, logs, product analytics.
• StarRocks — MPP columnar; Apache 2.0; strong on real-time joins and high-concurrency BI; growing fast.
• Apache Doris — sibling fork of StarRocks; Apache 2.0; popular in CN ecosystem.
• Greenplum — Postgres-flavored MPP; Apache 2.0. Older but mature for batch warehousing.
• CedarDB — academic / commercial Postgres-compatible columnar engine; closed source today, watch for OSS.

Real-time / sub-second analytics

• ★ Apache Druid — sub-second slice-and-dice over event streams; column store + bitmap indexes; ideal for "show me this hour's funnel." Apache 2.0.
• ★ Apache Pinot — same niche as Druid, LinkedIn-origin; very low latency for user-facing analytics. Apache 2.0.
• Rockset — was hosted-only; acquired by OpenAI in 2024 and effectively retired for new customers. Mention only to redirect.
• Tinybird — hosted ClickHouse + ingestion APIs; generous free tier for small workspaces.

Postgres-flavored analytics

• ★ ParadeDB — Postgres extension turning Postgres into a search + analytics engine (pg_search, pg_analytics). PostgreSQL license. Run analytical SQL where your app data already lives.
• Hydra (formerly Hydra Postgres) — columnar Postgres extension; AGPL-ish; less active in 2026.
• Citus — distributed Postgres; Microsoft-owned, AGPL-style; sharding more than columnar.
• TimescaleDB — see Time-Series Databases.

Cloud warehouses (paid, mention for context)

• Snowflake, BigQuery, Redshift, Databricks SQL — paid; usually where teams end up when "self-host analytics" outgrows the team. BigQuery's free tier (1 TB queries / 10 GB storage per month) is the most usable hobby option.
• MotherDuck — DuckDB-as-a-service with hybrid local/cloud queries; small free tier.
• Databend — open source Snowflake-shaped warehouse; Apache 2.0; managed Cloud free tier.

License & licensing watch-outs

• ClickHouse OSS is Apache 2.0; ClickHouse Cloud is paid SaaS — no AGPL/BSL switch as of 2026.
• StarRocks / Doris / Druid / Pinot are all Apache 2.0; safe to embed.
• ParadeDB ships under the PostgreSQL license; embeddable in commercial products.
• MotherDuck is closed-source SaaS; DuckDB itself remains MIT.

Pick this if…

• Default analytical SQL on Parquet / CSV / S3, single node: DuckDB.
• Default OSS clustered warehouse for events: ClickHouse.
• User-facing real-time analytics (sub-second): Druid or Pinot; ClickHouse if you can tolerate ~1s.
• High-concurrency MPP for BI dashboards: StarRocks.
• Stay on Postgres, add columnar: ParadeDB.
• Tiny project, big query bills not yet justified: BigQuery free tier or MotherDuck free tier.