BoilStream — Streaming Ingestion Lakehouse

BoilStream is a multi-tenant DuckDB server written in Rust with native DuckLake integration. Ingest, transform, aggregate, search, and consume streaming data — all with SQL.

Download, start, and connect with any Postgres-compatible BI tool. Data streams to S3 as DuckLake-managed Parquet files in real-time.

Key Features

Streaming Data Flow — Ingest → Transform → Aggregate → Search → Consume

• Multi-tenant DuckDB — Full tenant isolation (secrets, attachments, DuckLakes, filesystem)
• Streaming views — Continuous row-by-row SQL transforms on ingested data
• Materialized views — Tumbling/sliding window aggregations with DuckDB SQL
• Full-text search — Integrated Tantivy indexing with hot/cold tiered storage and multilake_search() SQL function
• Real-time consumption — SSE push with Arrow IPC batches via @boilstream/consumer JS SDK
• DuckLake integration — Embedded PostgreSQL catalog, 1s hot tier commits, cold tier hydration (>1GB/s)
• DuckLake vending — Credential vending for native DuckDB, DuckDB-WASM, and in-server queries
• Enterprise auth — Entra ID SAML, SCIM provisioning, MFA/Passkeys, Web Auth GUI
• Cluster mode — S3-based leader election, distributed catalog management
• Multi-cloud — AWS S3, Azure Blob, GCS, MinIO, filesystem

Companion projects: boilstream-extension for DuckDB/WASM clients | @boilstream/consumer for SSE consumption

Interfaces

Interface	Port	Description
Postgres	5432	BI tools (Power BI, DBeaver, Grafana, psql). Type compliance report
FlightRPC	50051	High-performance Arrow ingestion via Airport extension
FlightSQL	50250	Arrow SQL for ADBC drivers and FlightSQL clients
HTTP/2 Arrow	443	Arrow POST from browsers or HTTP clients
Kafka	9092	Kafka wire protocol with Schema Registry and Confluent binary Avro
SSE	443	Real-time Arrow IPC push to browsers and services

Start

See GitHub releases for the latest version.

# Download — pick your platform from:
#   darwin-aarch64   Apple Silicon Mac
#   linux-aarch64    Linux ARM64 — single Neoverse N1-safe build, runs on
#                    AWS Graviton 2/3/4, Hetzner cax*, Oracle Ampere A1,
#                    Apple Silicon under Docker.
#   linux-x64        Linux x86_64
#   windows-x64      Windows
# Replace {VERSION} with the latest release (see GitHub releases above, e.g. 0.10.27)
curl -L -o boilstream https://www.boilstream.com/binaries/darwin-aarch64/boilstream-{VERSION}
curl -L -o boilstream-admin https://www.boilstream.com/binaries/darwin-aarch64/boilstream-admin-{VERSION}
chmod +x boilstream boilstream-admin

SERVER_IP_ADDRESS=1.2.3.4 ./boilstream

# Docker — single ARM64 image runs on AWS Graviton, Hetzner cax*, Oracle
# Ampere A1, and Apple Silicon (or use x64-linux for x86_64):
docker run -v ./config.yaml:/app/config.yaml \
   -p 443:443 -p 5432:5432 -p 50051:50051 -p 50250:50250 \
   -e SERVER_IP_ADDRESS=1.2.3.4 boilinginsights/boilstream:aarch64-linux-{VERSION}

Use the accompanying docker-compose.yml to start Grafana and MinIO

Streaming DuckLakes

DuckLakes with __stream suffix enable real-time streaming. Tables become ingestion topics.

CREATE TABLE my_data__stream.main.events (user_id VARCHAR, event_type VARCHAR, ts TIMESTAMP, payload JSON);

-- Streaming view: continuous row-by-row filter
CREATE STREAMING VIEW clicks AS SELECT * FROM events WHERE event_type = 'click';

-- Materialized view: windowed aggregation
CREATE MATERIALIZED VIEW events_per_min AS
  SELECT event_type, COUNT(*) AS cnt FROM events
  WITH (window_type='tumbling', window_size='1 minute', timestamp_column='ts');

-- Full-text search: enable indexing, then query
ALTER TABLE my_data__stream.main.events SET (tantivy_enabled = true, tantivy_text_fields = 'payload');
SELECT * FROM multilake_search('my_data__stream', 'events__tantivy_idx', 'error timeout');

Ingest with DuckDB

When INSERT returns, data is guaranteed on S3.

INSTALL airport FROM community;
LOAD airport;
ATTACH 'my_data__stream' (TYPE AIRPORT, location 'grpc://localhost:50051/');

INSERT INTO my_data__stream.main.events
   SELECT 'user_' || i::VARCHAR, CASE WHEN i % 3 = 0 THEN 'click' ELSE 'view' END,
          NOW(), '{"page": "home"}'
   FROM generate_series(1, 20000) AS t(i);

Kubernetes

A production-ready Helm chart ships in charts/boilstream — StatefulSet, per-pod Envoy Gateway routes (TLS-passthrough SNI + per-pod L4 TCPRoute), cert-manager TLS, S3-coordinated cluster mode, and optional pod-to-pod mTLS.

helm install boilstream oci://ghcr.io/boilingdata/charts/boilstream \
  --version 0.3.27 \
  -n boilstream --create-namespace \
  -f my-values.yaml

Reference overlays: values-eks-example.yaml (AWS EKS + IRSA + S3) and values-hetzner-example.yaml (CloudFleet + Hetzner ARM64 + Hetzner Object Storage). Full walkthrough: docs.boilstream.com/guide/kubernetes.

Requirements

• 8GB+ RAM recommended
• macOS (arm64) or Linux (x64, arm64)
• Docker optional (for Grafana, MinIO)

Changelog

See CHANGELOG.md for version history and release notes.

Documentation

Full docs: docs.boilstream.com | Contact: boilstream.com