meshcore-hub/docs/database.md

Database

MeshCore Hub runs on PostgreSQL — the only supported database backend since v0.19. A bundled container ships with the stack for development (it joins the core compose profile via docker-compose.dev.yml, so a standard dev compose up works with zero configuration); managed/external Postgres and multiple schema-isolated instances sharing one cluster are equally supported — and are the norm in production.

Note

SQLite support (the pre-v0.14 default) was removed in v0.19. Deployments still on SQLite must upgrade through v0.19 and run the in-place migration there (the meshcore-hub db migrate-to-postgres command was itself removed in v0.20) — see the v0.19 upgrade guide.

Docker (bundled container)

Postgres ships with the stack. It is a development convenience: the base compose file only enables it for the all and standalone postgres profiles, and docker-compose.dev.yml unions it into core so the documented dev workflow needs no extra flags:

# Default dev stack: bundled Postgres starts alongside the app services
docker compose -f docker-compose.yml -f docker-compose.dev.yml --profile core up -d

Production (docker-compose.prod.yml) deliberately leaves it out of core — the bundled container's postgres network alias would shadow a shared instance of the same name on the same network. Start it explicitly with --profile postgres (or --profile all) if you want the bundled database on a production host.

The container derives POSTGRES_USER / POSTGRES_PASSWORD / POSTGRES_DB from DATABASE_USER / DATABASE_PASSWORD / DATABASE_NAME, and ships a dev-only default password (meshcorehub) so a default up initializes out of the box. Production deployments must override DATABASE_PASSWORD — the container is not published outside the compose network, but the default password is public knowledge (it is in the repo).

The migrate service waits for Postgres to be healthy and runs db upgrade before collector and api start. That dependency is optional (required: false): when the bundled container is not enabled (production with an external database), migrate simply runs against DATABASE_HOST directly.

Production provisioning (role and database)

The bundled container provisions the role and database for you on first start from the DATABASE_* values. For a managed or external Postgres, create them once before pointing Hub at it. This mirrors the init script used in the ipnet-mesh/infrastructure cluster:

-- Run once as a superuser/admin role on the target cluster
CREATE DATABASE meshcorehub;
CREATE ROLE meshcorehub LOGIN PASSWORD 'your-password';
GRANT ALL PRIVILEGES ON DATABASE meshcorehub TO meshcorehub;

The application schema and tables are created automatically by db upgrade (run by the migrate service on startup); the role just needs CREATE privilege on the database. Hub only ever connects as DATABASE_USER — no admin or bootstrap credentials are needed at runtime.

Managed or external Postgres

To point Hub at an already-running Postgres (e.g. a managed cloud instance or a shared cluster on the same Docker network), set the DATABASE_* connection variables. With the production compose override the bundled container is not started at all — nothing extra to skip, and no risk of its postgres alias intercepting your DATABASE_HOST. (If you use the dev override against an external database, add the standalone --profile postgres only if you also want the bundled container running.)

DATABASE_HOST=your-managed-postgres.example.com
DATABASE_PORT=5432
DATABASE_NAME=meshcorehub
DATABASE_USER=meshcorehub
DATABASE_PASSWORD=your-password

For advanced cases (custom driver, extra query params), set a full SQLAlchemy URL instead — it takes precedence over all the component variables:

DATABASE_URL=postgresql+psycopg2://meshcorehub:your-password@host:5432/meshcorehub

Missing connection configuration (no DATABASE_URL and no DATABASE_* components) fails fast at startup with an error naming the missing variables — Hub never silently falls back to a default.

Schema-per-instance (search_path)

Each Hub instance is isolated to its own Postgres schema via the connection's search_path, rather than its own database. This lets several instances (e.g. prod, stg) share one Postgres cluster without colliding — each gets its own tables and its own alembic_version.

Give every instance a distinct DATABASE_SCHEMA:

# Production (.env)
COMPOSE_PROJECT_NAME=hub
DATABASE_SCHEMA=meshcorehub_prod

# Staging (.env, separate directory)
COMPOSE_PROJECT_NAME=hub-beta
DATABASE_SCHEMA=meshcorehub_stg

The schema is created automatically on db upgrade if it does not exist, so no manual CREATE SCHEMA is required. Connect both instances to the same DATABASE_HOST / DATABASE_NAME / DATABASE_USER; only DATABASE_SCHEMA (and COMPOSE_PROJECT_NAME) differ.

Note: This is the database-level isolation for instances sharing a Postgres cluster. For running multiple instances on the same Docker host (separate volumes, Traefik routing), see Multi-Instance Deployments.