ADR-0007: One container image, Docker Compose first

• Status: accepted
• Date: 2026-09-03
• Deciders: Mahdi Amirabdollahi

Context

The fastest lever on open-source adoption is time-to-first-screenshot. Options range from a single jar to a Helm chart to an ArkMQ operator integration.

Decision

• One container image: the React SPA is built and baked into the Spring Boot jar (frontend Maven profile / Dockerfile stage), served from classpath:/static. The backend owns /api/**.
• Docker Compose first, in deploy/compose/:
  • compose.dev.yaml — Postgres + a real Artemis primary/backup pair + Studio. just up and you see a live topology graph in minutes.
  • compose.prod.yaml — Studio + Postgres only (brokers are external and registered through the UI), values from .env, resource limits, restart policy. A reference, not a turnkey production deploy.
• Helm chart in v1.0.
• ArkMQ operator integration (read broker CRs to auto-register clusters) is v1.1+, explicitly out of scope until the core is proven.

Consequences

• Trivial to evaluate; trivial to run on a VM or small k8s.
• The dev compose pulls the Artemis image and starts a replication pair — heavier than a bare jar, but that is the point (you see the product against a real cluster).
• Production hardening (TLS termination, managed Postgres) is the operator's job, documented, not automated in the MVP.

Alternatives considered

• Helm chart first — raises the bar to try the project; better fit only for k8s-only shops.
• ArkMQ operator integration first — high effort, narrow audience, premature before the core exists.