ClickHouse

A distributed columnar analytical (OLAP) database — shared-nothing, horizontally scalable across nodes, built for large-scale analytical workloads. The distributed end of the analytical-databases spectrum, and the engine tinybird manages.

Profile

• Distributed, multi-node, server-based — client-server protocol; scales horizontally and vertically (posthog-duckdb-vs-clickhouse).
• MergeTree storage — data in immutable parts with background merges; continuous streaming ingest and insert-time materialized views for pre-aggregation (posthog-duckdb-vs-clickhouse).
• Where it fits: high-ingest pipelines (100K+ rows/s), hundreds of concurrent queries, datasets beyond RAM, fault-tolerant production analytics (tinybird-clickhouse-vs-duckdb).
• Indicative numbers (vendor, tinybird-clickhouse-vs-duckdb): a 1 TB query ~10 s single-node → ~2-3 s on 4 nodes; ingestion ~200K inserts/s on a 16-core node → ~800K/s across 4 nodes.

Operational cost

The flip side of the distributed model is ops: tuning MergeTree variants and managing replicas (historically via ZooKeeper) (posthog-duckdb-vs-clickhouse). That operational burden is exactly what the managed-service layer (tinybird) and, in part, the reach for duckdb in some workloads are responses to. PostHog notes self-managed ClickHouse “would not work for data warehousing for all of our customers,” which pushed them to duckdb for that layer (posthog-duckdb-vs-clickhouse).

Related

duckdb · analytical-databases · tinybird · tinybird-clickhouse-vs-duckdb · posthog-duckdb-vs-clickhouse · synthesis