History

The development of SereneDB began in 2024, led by SereneDB GmbH in Germany.

Checkpoints

SereneDB ensures data persistence through consistent checkpoints and a write-ahead log (WAL).

Compression

Concurrency Control

SereneDB uses optimistic concurrency control transactional model. Search indexes over tables are eventually consistent.

Data Model

The system uses a relational data model and aims for high compatibility with PostgreSQL.

Indexes

The default index structures are SereneDB's own columnar storage layer for primary data organization and inverted indexes from IResearch for full-text search. SereneDB also supports dedicated vector indexes for approximate nearest-neighbor (ANN) search, alongside geospatial indexes for location-based queries. SerebeDB also supports secondary ART indexes.

Isolation Levels

The system supports Snapshot Isolation, which is the default level.

Joins

SereneDB leverages the join algorithms provided by its in-house query execution engine, inspired by DuckDB.

Logging

Query Execution

The system uses a query execution engine adapted from DuckDB, modified to read SereneDB's own indexes and tables and to integrate with IResearch throughout. It employs a vectorized execution model, in which operations are processed in batches of values (vectors) on a column-by-column basis.

Query Interface

SereneDB uses SQL as its primary query interface, specifically supporting the PostgreSQL dialect. Its support for the PostgreSQL wire protocol allows it to be used with standard PostgreSQL clients and tools. SereneDB also provides an Elasticsearch REST API — bulk and single-document ingestion, Query DSL search with aggregations, scrolling, and kNN/vector search, counts, and mapping endpoints — so existing Elasticsearch clients and tools work with it directly.

Storage Architecture

SereneDB is a disk-oriented database that supports larger-than-memory datasets. It utilizes a Hybrid Storage Engine that combines disk-based components (its in-house columnar storage layer, search index segment files) with in-memory data structures.

Storage Model

In-house columnar layout, tightly coupled with IResearch, for incremental updates and persistent storage. Columns are split into a set of compressed data blocks for aggregations, filtration, and parallelization.

Storage Organization

The storage is a blend of SereneDB's in-house columnar storage engine, inverted indexes, and a columnar data store, both tightly integrated with IResearch. Included columns are stored using the DuckDB file format. SereneDB also supports remote storage across a range of data sources and communication protocols.

System Architecture

SereneDB runs as a single server process that hosts the entire engine: query processing, columnar table storage, and search indexing, with durability provided by write-ahead logging. The planned distributed architecture follows the separation-of-compute-and-storage model: stateless compute nodes execute queries over a shared storage layer, so compute capacity can be scaled independently of the data, and both table data and search index segments — being immutable, append-only files — map naturally onto shared object storage.

Views

SereneDB supports standard SQL views and Search views to union query to multiple search indexes.