Back to catalog
ClickHouse
Apache-2.0ClickHouse is an open-source columnar database for real-time analytics, running sub-second SQL queries across billions of rows on self-hosted hardware.
Overview
Alexey Milovidov developed ClickHouse at Yandex to handle petabyte-scale event streams for Yandex Metrica without pre-aggregating data. Open-sourced in 2016, the column-oriented engine is engineered around hardware efficiency, streaming ingestion, and vectorized query execution across billions of rows.
Where ClickHouse fits
ClickHouse occupies the real-time analytical tier in modern data architectures, bridging the gap between transactional databases and long-term data lakes. Rather than forcing teams to endure slow queries in big-data engines or costly ingestion bills in cloud data warehouses like Snowflake or BigQuery, ClickHouse ingests streaming records directly while maintaining sub-second query latency. It does not handle operational transactional writes; instead, it consumes from them via Change Data Capture to run aggregations, filtering, and reporting without degrading primary application performance.
Its ecosystem positioning centers on broad data integration capabilities. Native table engines connect directly to Apache Kafka topics, RabbitMQ queues, and object storage buckets in AWS S3 or Google Cloud Storage, parsing Parquet, ORC, JSON, and Protocol Buffers on the fly. External table engines allow direct federated joins across live MySQL and PostgreSQL instances, while modern connectors let developers query Apache Iceberg and Delta Lake datasets directly without intermediate ETL jobs.
What it does well
- Vectorized query execution using SIMD instructions and column-oriented compression (LZ4, ZSTD, Delta) enables scans of hundreds of millions of rows per CPU core per second.
- Single self-contained C++ binary deployment simplifies infrastructure operations by eliminating JVM runtimes and external dependencies.
- Built-in ClickHouse Keeper provides a drop-in C++ consensus alternative to Apache ZooKeeper for managing replication with minimal memory footprint.
- Diverse storage engine family—including MergeTree, ReplacingMergeTree, and AggregatingMergeTree—handles deduplication, incremental rollups, and time-partitioned lifecycles automatically.
- Native Kafka table engines consume event streams continuously and populate materialized views without requiring separate processing clusters.
- Object storage integration decouples compute from storage by tiering cold historical partitions to Amazon S3 or MinIO while retaining hot data on local NVMe drives.
