Projects · 2026

IncidentFlow

An event-driven platform that turns repeated production errors into incidents, then correlates, notifies and records them.

  • .NET 8
  • .NET Aspire
  • MassTransit
  • RabbitMQ
  • Redis
  • PostgreSQL
  • YARP
.NET services behind one gateway
7
signal types accepted
5
detection window in Redis
5 min

What it does

Production systems emit errors, logs, metrics, deployments and health checks. IncidentFlow takes all five in through one gateway, answers immediately, and does the thinking in the background. A detection worker watches for the same error repeating inside a five-minute window; when it does, an incident is opened. From there, separate workers correlate it with recent deployments, send notifications and keep a timeline, and none of them can hold up the others.

The architecture page walks through every service and the path a repeated error takes, checked against the code.

What I learned

  • Answer first, think later. Ingestion returns 202 Accepted as soon as a signal is stored and published, so an alert storm queues instead of blocking the way in.
  • Let one service own a decision. Only the Incident service decides what an incident is. Everything else reacts to the events it publishes.
  • Check the diagram against the code. My first drawing had a deduplication stage that does not exist and a straight line where the real flow fans out.