Apache Avro

Avro extension

This Apache Druid extension enables Druid to ingest and understand the Apache Avro data format. This extension provides two Avro Parsers for stream ingestion and Hadoop batch ingestion. See Avro Hadoop Parser and Avro Stream Parser for more details about how to use these in an ingestion spec.

Additionally, it provides an InputFormat for reading Avro OCF files when using native batch indexing, see Avro OCF for details on how to ingest OCF files.

Make sure to include druid-avro-extensions in the extensions load list.

Avro Types

Druid supports most Avro types natively, there are however some exceptions which are detailed here.

Unions

Druid has two modes for supporting union types.

The default mode will treat unions as a single value regardless of the type it is populated with.

If you wish to operate on each different member of a union however you can set extractUnionsByType on the Avro parser in which case unions will be expanded into nested objects according to the following rules:

  • Primitive types and unnamed complex types are keyed their type name. i.e int, string
  • Complex named types are keyed by their names, this includes record, fixed and enum.
  • The Avro null type is elided as its value can only ever be null

This is safe because an Avro union can only contain a single member of each unnamed type and duplicates of the same named type are not allowed. i.e only a single array is allowed, multiple records (or other named types) are allowed as long as each has a unique name.

The members can then be accessed using a flattenSpec similar other nested types.

Binary types

bytes and fixed Avro types will be returned by default as base64 encoded strings unless the binaryAsString option is enabled on the Avro parser. This setting will decode these types as UTF-8 strings.

Enums

enum types will be returned as string of the enum symbol.

Complex types

record and map types representing nested data can be ingested using flattenSpec on the parser.

Logical types

Druid doesn’t currently support Avro logical types, they will be ignored and fields will be handled according to the underlying primitive type.