0011 — Self-hosted Grafana LGTM monitoring

Status: reconstructed — needs confirmation

Context

The platform needs metrics, logs, and (eventually) traces without a SaaS observability bill, on the same single VPS.

Decision

monitoring/docker-compose.yml runs Grafana 12, Prometheus 3, Loki 3, Tempo 2, Grafana Alloy, node-exporter, and cAdvisor — all memory-capped for the 1-vCPU box, none with published ports. Grafana is reached only through Caddy at monitor.aimgold.org. Alloy tails every container's logs into Loki; Prometheus scrapes the NestJS apps' prom-client metrics at /v1/metrics (libs/common/src/metrics/). Dashboards and datasources are file-provisioned from git (monitoring/grafana/).

Consequences

• Monitoring config is code-reviewed and redeployable (.github/workflows/deploy-monitoring.yml); dashboard edits made in the Grafana UI are not persisted to git.
• Metric retention 30 d / 10 GB, logs 14 d, traces 7 d — bounded by VPS disk.
• Tempo receives OTLP but the deployed apps don't ship traces yet.