Miguel Lucía
C#·.NET 10·ASP.NET Core·PostgreSQL·EF Core·Docker
- Role
- Solo — design, backend and deployment
- Problem
- I run several services on a single host and had no way to know whether they were up, or when that stopped being true. Prometheus with Grafana is two more services to operate, and its pull model assumes it can reach the machine — which is precisely what fails when the machine dies.
- Decision
- My own ingestion engine: push-first, sized for one operator. The domain never reads, never writes, never sleeps and never asks what time it is — the instant is always a parameter — and guard tests fail the build if any file touches the ambient clock. The hot path writes through binary COPY instead of EF Core, and retention drops whole partitions rather than deleting rows.
- Outcome
- Ingestion, rollups and the alert engine run in production on the host where the Discord bot on this page lives, and the alerts deliver to Discord for real, not in a test. The dashboard is not built. No rule has yet fired on a real condition — the only test was provoked — so the thresholds are set by judgment, not calibrated against how the host actually behaves.
Last push: Sep 7, 2026