GrowlerDB

Open-source text search engine over Apache Iceberg. Iceberg stays the system of record; GrowlerDB keeps a fast derived full-text index and returns the matching primary keys, which hydrate back to authoritative rows in the lake.

Get started View on GitHub

---

The model in one minute

• You define an index over an Iceberg source table — which columns to index, the composite key, and an optional tenant_field. A connector keeps the index in sync with the table.
• A search runs against the derived index and returns document coordinates (the composite key) + scores — not documents.
• Your client (or built-in hydration) then fetches the authoritative rows from Iceberg by key (POST /v1/keys:get), governed by the catalog. The lake stays the source of truth.

This is the inverse of a search engine that owns its documents — and it’s why GrowlerDB layers onto a lakehouse instead of duplicating it.

Documentation

Page	What it covers
Getting started	Compose stack → your first search → hydrate → console.
Install & run modes	Build from source; embedded, serve, gateway, control-plane.
Configuration	CLI flags, env vars, and the index-definition YAML (fields, key, tenant).
Reference	Query language, the REST/gRPC API, and the OpenSearch _search adapter.
Migrating from Elasticsearch / OpenSearch	Concepts + the two integration paths.
Deployment	Local Compose and Kubernetes (Helm).
Storage & tiering	Hot vs cold object-storage tiering — when to use it (and when to keep hot).
GA criteria	What’s GA-ready vs. the road to 1.0.

Feature overview

• Search over Iceberg with PK hydration (/v1/search → /v1/keys:get).
• Query language — native structured AST + a Lucene/KQL string parser.
• Distributed — control plane + stateful searcher nodes + a scatter-gather gateway.
• Secure & multi-tenant — OIDC/JWT, API keys, mTLS; control-plane RBAC; non-widenable tenant scoping.
• Observable — OpenTelemetry traces/metrics/logs, Prometheus, bundled Grafana SLI dashboards.
• Console UI + an optional OpenSearch _search adapter.