OpenMetadata takes the schema-first approach to model metadata as a Single Source of Truth with clear vocabulary for the system. First, the Entities, Types, and relationships between the entities in the system are identified and schemas are defined for them. Code is then generated from these schemas for implementation. We use JSON schema as the Schema Definition Language as it offers several advantages:

Easy to describe the structure and semantics of metadata models with readable documentation that is both human and machine consumable.
Common types can be developed once and can be reused as building blocks in other schemas and become the basis of vocabulary development.
Models can include structural validation, such as required/optional fields, default values, allowed values, regex that not only serve as automated testing and validation but also as documentation of API contract.
A rich set of tools are available that supports JSON schema support for generating code and validation from JSON schema in various languages, reducing the manual boilerplate coding.
Supports rich formats to convert schema types into native standard types during code generation, such as URI, date, and time.

Reference

JSON schema specification version Draft-07 to 2019-099
JSON schema 2 POJO tool used for Java code generation from JSON schema
Data model code generator for generating python code from JSON schema

Schemas

{% page-ref page="entities/" %}

{% page-ref page="types/" %}

Version Note

The schemas linked above follow the JSON Schema Spec version: http://json-schema.org/draft-07/schema#