ADR-0021: Contributor Compose and dev containers here, infrastructure in stackr¶

Status: Accepted Date: 2026-09-28 Deciders: Alex Nodeland

Context¶

Working on reflexr needs PostgreSQL and the reference app. Evaluating and operating it also needs the observability and LLM infrastructure: an OpenTelemetry Collector, Grafana with Prometheus, Tempo, Loki and Pyroscope, self-hosted Langfuse, and a LiteLLM proxy. reflexr and evalr need the same infrastructure, and so will applications built on them. The existing dev container installs only the Python environment.

Decision¶

• This repository keeps what developing reflexr needs: a compose.yaml with PostgreSQL by default and the reference app under an app profile, and a dev container built on it that sets the test environment.
• The shared infrastructure lives in stackr (ADR-0023), in one place for every library and application. When it is running, the dev container joins its network and sends telemetry to its Collector.
• Dashboards stay here, next to the metric registry they are tested against, and are published with each release for stackr to provision by version.
• CI validates the Compose file and dashboard JSON without starting containers.

Amendment (2026-09-29): CI builds and starts oncall¶

Validating the Compose files without starting containers cannot tell whether the reference app's image still builds; artifactr's stopped building unnoticed (artifactr#52). CI's Compose job now builds oncall's image, starts it on the Compose PostgreSQL, waits for its health check, and requests its root page. It still starts nothing else.

Options considered¶

Consequences¶

• Easier: the infrastructure is maintained once and used the same way everywhere.
• Harder: seeing traces while developing needs stackr's stack running as well.

Action items¶

1. Implement RFC-0002 phase B4 (ADR-0037, ADR-0038).