# Apache Atlas

> The metadata and governance layer the Hadoop stack was built around

Apache Atlas is an Apache Software Foundation project providing metadata management, classification, and lineage for the Hadoop ecosystem, with an extensible type system and an authorization path through Apache Ranger.

Website: https://atlas.apache.org

## Use it when

- The platform is Hadoop-shaped: Hive, HBase, Kafka, and Storm hooks capture lineage at runtime as queries execute.
- Classifications must enforce rather than describe: tags become Apache Ranger authorization and masking decisions, so a column marked PII is one the engine refuses to serve.
- You need an extensible type system for modelling unusual assets over REST.
- Atlas already ships inside your platform distribution, as it does with Cloudera.

## Think twice when

- You want a modern discovery experience; DataHub and OpenMetadata are built around search, and Atlas is not.
- The stack is cloud warehouses and dbt rather than Hadoop; the connector list is a portrait of an earlier generation.
- Nobody wants to run a graph store, a search index, and Kafka just for a catalog.

## How it runs

Self-hosted Java services: a JanusGraph metadata store over a configurable backend, a separate search index, and Kafka for change notifications. Metadata is pushed in by engine hooks as queries run rather than pulled on an ingestion schedule. Development continues at a slow but real pace, with releases through 2026.

## Details

### General

| Attribute | Value |
| --- | --- |
| License | Open source (permissive) |
| SPDX identifier | Apache-2.0 |
| Deployment | Self-hosted |
| Workload | Interactive, Batch |
| Operational complexity | High |
| Pricing | Free |
| Language | Java |
| Repository | https://github.com/apache/atlas |
| Documentation | https://atlas.apache.org |
| Stars | 2.1k |

### Data Catalogs

| Attribute | Value |
| --- | --- |
| Metadata model | A type system you extend, entity, classification, struct and enum definitions with supertypes and typed attributes, instantiated as entities in a JanusGraph store. Metadata is pushed in by engine hooks as queries run rather than pulled on an ingestion schedule |
| Connectors | Ten bridges and hooks in the tree, and the list is a portrait of its generation, Hive, HBase, Impala, Kafka, Sqoop, Storm, Falcon, HDFS and Couchbase, with a Trino extractor the newest arrival; anything else arrives through the REST API |
| Lineage | Captured at runtime by the engine hooks rather than parsed after the fact (a Process entity records the inputs and outputs of a query as it runs) and classifications propagate along it, so a table marked PII marks the view derived from it too |
| Search and discovery | Basic search over type, attribute and classification filters; a SQL-like DSL for advanced search with grouping and aggregation; relationship search; and saved favourite searches, with free text served by the separate search index |
| Business glossary | First-class: terms grouped into categories that nest into hierarchies, related to one another semantically and mapped onto assets, and managed separately from the metadata they annotate |
| Policy and compliance | The reason to run it, classifications become authorization and masking decisions in Apache Ranger, role- and attribute-based, so a column marked PII is one the engine refuses to serve and one marked NATIONAL_ID can come back as its last four digits. Classifications carry attributes of their own, and Atlas's own API is authorised by either the Ranger authorizer or a simple file-based one |
| Quality checks | None: Atlas describes and classifies data, it does not test it, so anything from data-quality sits beside it rather than inside it |
| Access requests | None: no request, approval or grant flow anywhere. Access is decided in Ranger against the classifications Atlas supplies, and asked for however the organisation already asks |
| Custom metadata | The point of the type system, define new entity types with supertypes and typed attributes over REST, and layer business metadata on top: struct-like attribute sets bound to chosen entity types, assigned per entity through the UI or API and searchable once assigned |
| Billing unit | Nothing to pay: an ASF project with no vendor tier above it, one of only two such rows here. In practice most deployments arrive inside a platform distribution such as Cloudera's and are paid for as part of that |

## Capabilities

- [Data Catalogs](https://matca.io/capabilities/data-catalog)

---

Source: https://matca.io/apache-atlas
Last updated: 2026-08-31T11:40:47.660Z