Every component is green, therefore the service works

A team aggregates component health checks into availability while accepted work never reaches its terminal state.

Fictional educational scenario. It does not describe a particular organisation or incident.

Situation

A dashboard shows a green API, database and broker. The health endpoint returns 200, so the service availability report says 99.99%. A stopped consumer is not completing accepted work, however, and users remain in an in-progress state.

Where the anti-pattern appears

Component diagnostics are aggregated and renamed as a service measure. Nobody has established whether the promise ends at accepting a command or includes reaching a terminal state within an agreed time. The green report answers a technical question correctly and a business question incorrectly.

Warning signs

process availability is calculated as an average of component availability;
an HTTP 200 ends measurement even though work continues asynchronously;
queue depth is measured, but not oldest-work age or completion rate;
alarms do not distinguish acceptance, completion and permanent failure;
there is no safe end-to-end probe or stopped-consumer test.

Consequences

Operators see healthy infrastructure during a real service interruption. They scale the wrong component and detection takes longer. An SLA report can remain green while users receive no outcome. Uncontrolled replay after recovery can create duplicate effects. These are possible consequences of the model, not a reported incident.

A better decision

Separate durable acceptance from completion and name the terminal states.
Keep liveness and readiness for diagnostics, then add progress measures: work age, completion rate and permanent failures.
Define an outcome SLI at the user boundary and agree its time window.
Run a controlled synthetic journey with harmless data and verify its terminal state.
Test stopping and restarting a consumer, including duplicate prevention and accounting for every accepted item.

Boundary

An end-to-end test does not replace component monitoring. It reveals that the promise is broken; layered signals help find the cause. Both levels are needed.

A green component is a clue. A green process outcome needs evidence.

Foundation: business-process availability · Practice