Internal Developer Platform  ·  View 06 of 21  ·  People and journeys

Journey — Debug a Service the Platform Generated

The journey that prices the abstraction, because every hidden concept is met here for the first time.

Editable source SVG draw.io All views
On-Call Engineer 03:14, six months in Goal — Find out why a service I did not write is failing, on a runtime I have never had to look at Trigger — An SLO burn alert on a component scaffolded by someone who has left the team Done when — Cause understood or rolled back, inside the 30-minute escalation window 1 · Paged alert fires 2 · Orient ◆ moment of truth 3 · Look inside it leaks ◆ moment of truth 4 · Decide 5 · Recover ≤ 5 min What they do Opens the alert link Reads component page Needs the pod itself Chooses rollback Confirms recovery Where Pager Component page Escape hatch · kubectl Platform CLI Dashboards What the platform does Routes to owning rota Deploys · deps · docs Grants scoped read Rollback, pipeline-free Records the event How it feels Confident Coping Lost Where it hurts Owner group is stale Never seen the substrate User Journey — On-Call Engineer Debugs a Service the Platform Generated This is the journey that prices the abstraction question. Every concept the golden path hides is a concept somebody meets for the first time at 03:00, which is why the escape hatch is a designed path rather than a leak. v 1.0 · owner Platform Architecture · date 2026-09

The trough

  • The moment the abstraction leaks and the engineer needs the pod, not the model of a pod — at 03:00, on a substrate the golden path has spent a year hiding.
  • The escape hatch is therefore a designed, entitlement-scoped path rather than a leak to be apologised for (ADR-07).
  • A stale owning group makes this worse than the technical problem: the second dip is not knowing who else to wake.

What this demands

  • Rollback must not depend on the pipeline that produced the deploy, because the failure may be the pipeline (ADR-13).
  • Ownership resolved against live identity groups, with an empty group treated as an incident (ADR-10).
  • The component page must answer health, deploys, dependencies, runbook and on-call in one step, or the first five minutes are spent navigating.

Open

  • How much substrate the golden path should hide is the one question this set leaves genuinely open (Question 4 in the requirement, ADR-07).
  • The answer changes with the seniority mix of 900 engineers and with who is actually on the rota.
◀  Journey — Drive a Change Across 1,400 Components All views Layered Architecture  ▶
Solution Architecture v1.0 · Microsoft Azure · Platform Architecture · 2026-09 · 21 views · 16 architecture decision records06-journey-oncall-golden-path