FoundationDB Record Layer Documentation

The FoundationDB Record Layer is a Java library that provides a record-oriented data store and a relational database interface built on top of the FoundationDB key/value store. Its key features are:

  • A SQL interface for defining schemas, querying data, and manipulating tables using familiar SQL syntax. Java applications interact with the Record Layer through a standard JDBC interface.

  • A record-oriented Java API. This API defines records using protocol buffers, supports declarative query execution, and offers fine-grained control over storage layout and index maintenance through extension points such as index maintainers and custom query components. It is positioned as a low-level alternative to SQL for advanced use cases; most of its capabilities are expected to migrate to the SQL interface over time.

  • Reusable schema templates, which enable massively multi-tenant architectures where each tenant instantiates their own database based on a shared schema that can safely evolve over time.

  • Rich indexing capabilities, which include value indexes, aggregate indexes, rank indexes, and indexes on the commit time of records. Indexes are akin to materialized views and can be updated incrementally.

  • An advanced type system that, beyond the standard types found in traditional relational databases, includes user-defined struct types, arrays of primitives or complex types, and fixed-dimension numerical vectors for ML embeddings and similarity search.

  • An intelligent query planner, which performs automatic index selection and query optimization, and supports complex correlated subqueries, joins, aggregation with GROUP BY, and ordering (ORDER BY). To limit memory use, query plans avoid in-memory buffers and instead rely heavily on stream-based processing over available indexes.

  • Full ACID transactions, inherited from FoundationDB along with its strong reliability and performance in a distributed setting. To work within the transaction time and size limits, the query API provides continuations that efficiently page through large result sets.

  • A scalable, stateless architecture. Scaling up is as simple as launching more instances, since there is no server-side state—all data resides in FoundationDB. With millisecond-level store initialization and query execution, it is well-suited to multi-tenancy across thousands of independent database instances. Each tenant’s state is encapsulated, and resource consumption is tightly constrained and balanced across users, even under a wide variety of workloads.