Partner outage impact attribution
When an external partner fails, responders now see exactly which citizen services are affected and by how much, in one view.
Code available on request
Context
Irembo's platform depends on many external partner integrations. When one failed, the team could see the failing integration but not which citizen-facing services it was breaking, or how badly.
What I did
- Designed a request-level attribute that records which citizen service triggered each partner call. It rides on the telemetry already in place, so no new data pipeline was needed.
- Calls that can't be attributed are labelled as such, so the coverage gap can be measured instead of hidden.
- Separated citizen service requests from internal processes.
- Designed the incident dashboard so every dependent service is visible at once and compared with its normal behaviour, with no filters to click through.
Key decisions
Attribution at runtime over a manually maintained mapping table
A manual table goes out of date quickly.
Attribution at runtime over static code scanning
Scanning shows which calls are possible, not which ones production actually makes.
Result
In production. During a partner outage, responders open one dashboard and see every affected service and how far it has moved from normal.