The booking portal of Nordlicht Reisen GmbH, a fictional travel agency, has been delivered daily for months, and nobody looks at the delivery chain itself. On Monday the main branch sat in the red for four hours because a test was unstable and the notification landed in a mailbox that nobody reads. On Tuesday, pull request builds waited half an hour for a free agent. You are to monitor the delivery chain so that such things stand out before anyone asks: Who is called when, and what does that person look at?
The exam calls this: configure monitoring with Azure Monitor and Logs, Application, VM, Container, Storage and Network Insights, GitHub insights and alerts for GitHub Actions and Azure Pipelines.
Prior knowledge: Metrics, diagnostic settings, Log Analytics, alert rules and action groups as well as the insights for VM, storage and network are covered in AZ-104, lessons 5.1 to 5.4. Application Insights with OpenTelemetry, KQL, health checks and alerts for a .NET application are shown in K3, module 3.4; lead time and recovery time as metrics in K3, lesson 3.7.7 and AZ-400, lesson 1.3. This lesson applies that to the delivery chain: pipelines, runners, agents and the path to production.
What does the delivery chain monitor?
The delivery chain has signals of its own that are not in any application:
| Signal | Question | Source |
|---|---|---|
| Pipeline health | How often does it fail, how long does it take, which task fails most often? | pipeline reports (Azure DevOps), Actions metrics (GitHub) |
| Wait time | How long does a job stand in the queue? | Actions performance metrics, agent pool view |
| Agents and runners | Are they alive, are they busy? | VM insights for self-hosted machines, pool pages |
| Approvals | Is an approval waiting too long? | notification "Run stage waiting for approval" |
| Changes to the platform | Who changed permissions, connections, PATs? | audit log and audit stream |
| State after delivery | Is the new version running healthy? | Azure Monitor, Application Insights, alerts as a gate |
A rule of thumb: Measure every signal at the place that knows it anyway, and forward only what gets someone to act.