ClickHouse
ClickHouse is an open-source columnar database designed for low-latency analytical queries over large volumes, available self-hosted or as ClickHouse Cloud.
LicenseOpen source (permissive)· Apache-2.0
DeploymentSelf-hostedManagedHybrid
PricingFreeSubscription
Operational complexityMedium
WorkloadInteractiveStreaming
LanguageC++
Use it when
- Interactive analytics over large event volumes need sub-second scans: user-facing dashboards, observability, clickstream.
- Ingest is heavy and continuous, Kafka included, with materialized views maintaining aggregates at insert time.
- SQL depth matters: a rich analytical dialect, plus MySQL and PostgreSQL wire ports for existing tooling.
- You want an open-source engine with a first-party managed cloud when the operations should go away.
Think twice when
- The workload is update-heavy; mutations rewrite whole parts, so OLTP-shaped change belongs elsewhere.
- Query-time joins across many large tables are the norm; StarRocks makes that its headline, and warehouses absorb it too.
- Extreme per-user query rates with strict SLAs are the entire product; Pinot and Druid are built around exactly that shape.
How it runs
Self-hosted from a single node to replicated shards with Keeper coordinating, MergeTree tables on local disk; ClickHouse Cloud runs SharedMergeTree over object storage with autoscaling compute. Insert-trigger materialized views, refreshable views, and projections carry pre-aggregation. Iceberg and Delta are reachable through DataLakeCatalog.
Details
CompareHow ClickHouse answers the questions OLAP Engines turns on.
OLAP Engines
| How it works | |
| Specialization | General-purpose column store for analytics, logs and observability |
| SQL dialect | ClickHouse SQL; MySQL and PostgreSQL wire ports |
| Joins | Distributed, several join algorithms |
| Updates and deletes | Lightweight UPDATE and DELETE; mutations rewrite parts |
| Ingest to queryable | Immediate on insert; async inserts not queryable until flush |
| Storage format | MergeTree; Iceberg and Delta via DataLakeCatalog |
| Compute and storage | MergeTree on local disk; SharedMergeTree over object storage in Cloud |
| Indexes | Sparse primary key; skipping indexes: minmax, set, bloom filter, text |
| Pre-aggregation | Materialized views as insert triggers; refreshable MVs and projections |
| Running it | |
| Concurrent users | No fixed limit; concurrency traded against per-query parallelism |
Alternatives to ClickHouse
