How Citus Turns PostgreSQL into a Distributed Database

What Is Citus?

Citus is an open‑source extension to PostgreSQL that distributes data and queries across multiple nodes, turning PostgreSQL into a distributed SQL engine with sharding, reference tables, and distributed tables.

Ways to Obtain Citus

Open‑source: download the source code or clone the GitHub repository at https://github.com/citusdata/citus (download page: https://www.citusdata.com/download).

Enterprise software: Citus Enterprise is offered as a subscription with 24×7 support for self‑managed deployments.

Managed service: Azure provides Citus as the built‑in “Hyperscale (Citus)” option of Azure Database for PostgreSQL ( https://docs.microsoft.com/azure/postgresql/hyperscale).

Scalability Limits

Citus scales horizontally by adding worker nodes and vertically by upgrading workers or the coordinator. Reported customer deployments include:

Algolia – ingesting 5‑10 B rows per day.

Heap – >700 B events on a 70‑node Citus cluster storing 1.4 PB of data.

Chartbeat – >2.6 B rows added each month.

Pex – 80 B rows per day on a 20‑node Google Cloud cluster with 2.4 TB memory, 1 280 CPU cores, 80 TB storage, and a plan to grow to 45 nodes.

Mixrank – 1.6 PB of time‑series data.

When to Use Citus

Multi‑tenant Databases

Citus provides full SQL coverage and can support >100 K tenants. It offers tenant isolation, performance guarantees for large tenants, and reference tables to reduce data duplication. Benefits include fast cross‑tenant queries, sharding logic in the database rather than the application, higher data density per PostgreSQL node, sustained performance under high concurrency, rapid metric analysis across customers, easy scaling for new registrations, and isolated resource usage for large customers.

Real‑time Analytics

Citus enables sub‑second response times on large, rapidly growing event streams. Typical use cases are analytical dashboards with sub‑second latency, exploratory queries on ingesting events, large‑scale archival and reporting, and funnel, segmentation, and cohort analyses. The engine parallelizes query execution and scales linearly with the number of workers, preserving SQL semantics while handling high concurrency.

Usage Considerations

Citus extends PostgreSQL’s distributed capabilities but is not a drop‑in replacement for every workload. Selecting appropriate SQL features and evaluating the data model are required. If a workload matches the described multi‑tenant or real‑time analytics patterns, Citus can be applied; otherwise, work‑arounds may be needed for unsupported tools or queries.

When Citus Is Not Suitable

Workloads that do not need a distributed database or that require heavy inter‑node data flow may perform poorly with Citus. Examples include:

Single‑node PostgreSQL deployments that already meet performance needs and have no anticipated growth.

Offline analytics that do not require real‑time ingestion or queries.

Analytical applications without a need to support a large number of concurrent users.

ETL queries that return massive result sets rather than aggregated summaries.