OpenMetadata — Local Docker Compose Stack

Spin up a full OpenMetadata instance locally with one command — PostgreSQL + Elasticsearch + Airflow + the OM server, all in Docker Compose. No remote database, no remote Airflow, no cloud access needed. The same compose is meant to run later on EKS by only swapping environment variables (see docs/ENVIRONMENT.md).

Prerequisites

• Docker with the docker compose v2 plugin. Nothing else (no Java/Python/Node).
• ~6–8 GB RAM free for Docker (two JVMs + Airflow + Postgres).

Quickstart

Option 1 — with make

cp .env.example .env       # local-test config (safe defaults, no real secrets)
make up                    # pulls images, starts everything, waits until healthy

make down stop · make logs-server logs · make ps status · make clean wipe all data.

Option 2 — plain commands (no make)

cp .env.example .env
docker compose up -d --wait

docker compose down                              # stop (keep data)
docker compose logs -f openmetadata-server       # follow server logs
docker compose ps                                # status
docker compose down -v                           # stop AND delete all data

Then open http://localhost:8585 and log in with admin@open-metadata.org / admin. The Airflow UI is at http://localhost:8080 (admin / admin).

First start takes a few minutes (image pulls + one-shot DB migration). --wait blocks until the server is healthy. If a port is already in use, change the *_PORT values in .env.

What runs

Container	Image	Port	Role
openmetadata_server	openmetadata/server:1.10.0	8585 / 8586	REST API + Web UI
openmetadata_postgresql	postgres:16.6	5432	Metadata DB (OM + Airflow)
openmetadata_elasticsearch	elasticsearch:8.11.4	9200	Search / discovery
openmetadata_migrate	openmetadata/server:1.10.0	—	One-shot DB migration, then exits
openmetadata_ingestion	openmetadata/ingestion:1.10.0	8080	Airflow 2.10.5 (ingestion)

Architecture details: docs/ARCHITECTURE.md. Every connection setting is an environment variable whose name mirrors the DevOps deployment — full reference and what to change for EKS: docs/ENVIRONMENT.md.

Troubleshooting

dependency failed to start: container openmetadata_postgresql is unhealthy (postgres logs show FATAL: database "openmetadata_db" does not exist).

The Postgres data volume is stale: PostgreSQL only runs init/postgres-init.sh on a fresh, empty data dir, so a leftover/partial volume from an earlier run skips DB creation. Reset it:

docker compose down -v        # ⚠ deletes the local DB volume (local test data only)
docker compose up -d --wait

down -v wipes the local volumes; on this stack that's only throwaway test data, so it's safe.

Versions — why OpenMetadata 1.10.0 (not 1.13)

The deployment uses Airflow 2.10.3 and PostgreSQL 16.6. OpenMetadata bundles its own Airflow inside the ingestion image, and that version is fixed per OM release:

OpenMetadata	Airflow
1.6.0	2.9.1
1.7.0 – 1.10.0	2.10.5
1.11 – 1.12	3.1.x
1.13.0	3.2.1

OM 1.13 ships Airflow 3.2.1 — a different major version with breaking changes — which would not match the deployed 2.10.3. So we pin OM 1.10.0, the newest release still on Airflow 2.10.5 (the same minor as 2.10.3), to keep local behaviour as close to production as possible. PostgreSQL is pinned to 16.6 exactly. (If you ever need exactly Airflow 2.10.3, build a custom ingestion image FROM apache/airflow:2.10.3 + openmetadata-managed-apis — see AGENTS.md.)

Going to EKS / production

The stack is portable by design: keep the variable names, change their values to the managed services (Aurora Postgres, the real Airflow, managed search) and drop the local postgresql / elasticsearch / ingestion containers. See the "TO FILL FOR EKS" block in .env.example and docs/ENVIRONMENT.md.