Home > Positioning > Subjects > Apache Avro > Implementations and use

Implementations and use

Avro is a specification with many implementations. Because reading and writing are driven by a schema rather than by generated code, an implementation in any language need only interpret the schema and follow the encoding and resolution rules.

Implementations

The Java implementation is the reference and the most complete; Java and Python carry the project’s primary getting-started guides. The Apache project releases official SDKs, versioned together with each Avro release, for Java, C, C++, C#, Python, Ruby, Perl, PHP, Rust, and JavaScript.

• Rust — the official crate is apache-avro, now maintained in its own repository.
• JavaScript — the project publishes avro-js on npm. Distinct from it is avsc, a widely used third-party pure-JavaScript implementation by Matthieu Monsch; the two are separate projects and worth not conflating.

Beyond the official SDKs, community implementations exist for languages including Go, Scala, Haskell, and Elixir.

Confluent Schema Registry

The most prominent user of Avro outside its own toolchain is the Confluent Schema Registry, the de facto way Avro is used with Apache Kafka. It stores schemas centrally, assigns each a numeric ID, and enforces compatibility rules as schemas evolve.

It is worth being precise that the Registry’s wire format and compatibility framing are Confluent-specific, not part of Avro itself. A Confluent-serialized message is a single magic byte (0x00), then a four-byte big-endian schema ID, then the Avro-encoded payload — a reference to a schema held in the Registry rather than the schema-bearing object container file or the RPC handshake that Avro defines natively. Its compatibility modes — BACKWARD (the default), FORWARD, FULL, their transitive variants, and NONE — are the Registry’s policy layer enforced ahead of publication, built on top of Avro’s resolution rules but not specified by Avro. The principle that the schema travels with the data is preserved in spirit — the ID travels, and the schema is one lookup away — but the carrier is Confluent’s, not Avro’s.

Where Avro is used

Avro’s centre of gravity is data infrastructure. It is the common message serialization format in the Kafka ecosystem, a first-class format in Apache Spark and across the Hadoop ecosystem it came from, and a frequent choice for data lakes, event streams, and data pipelines — settings where data outlives the code that wrote it and schemas change while old data and old readers remain in service.

Sources

• Apache Avro documentation and downloads.
• Confluent Schema Registry documentation — wire format and compatibility modes.