01 · The challenge
One expired credential. A cascade of stale data.
The dashboard combines enterprise order feeds with warehouse pick activity so the floor can see wave progress and zone performance in real time.
An expired integration credential stopped the upstream exports. A coupled trigger then let files pile up, and when they replayed they created duplicate staging rows and stale operational state. The floor was looking at numbers it couldn’t trust.
02 · What I did
What I owned and built.
- Traced the failure across API credentials, integration connections, storage events, pipelines, SQL staging, and scheduled processing.
- Preserved the affected data, isolated the valid current dataset, and removed duplicate staging records in controlled batches.
- Restored clean current-day processing without deploying application code, and validated downstream wave creation.
- Documented prevention measures: independent triggers, idempotent ingestion, file lifecycle controls, and credential-expiration alerts.
03 · Architecture
How it fits together.
Logic Apps · Key Vault credentials
Storage-event triggers
Duplicates isolated & removed in batches
SignalR
04 · Key decisions
The judgment behind the build.
Why preserve before cleaning?
In an incident, the evidence is the data. Keeping a copy of everything affected meant every cleanup step could be checked and reversed.
Why controlled batches?
Deleting duplicates in one pass against a live operational table risks locks and mistakes. Batches kept the floor running and every step verifiable.
Why recover without a code deploy?
Shipping code mid-incident adds new risk. Recovering through data and configuration got operations back fastest and safest.
Why look past the expired credential?
The credential started it, but the coupled trigger and non-idempotent ingestion turned a pause into corruption. Those are the real fixes.
05 · Impact
What changed because of the work.
- Restored trustworthy, current-day warehouse visibility for floor operations.
- Identified architectural failure modes beyond the initial expired credential.
06 · What’s next
Where it goes from here.
Prioritize durable safeguards, then validate restart and replay behavior under abnormal source volumes.
This case study is a sanitized reconstruction. Organization names, proprietary code, data, and internal schemas are intentionally left out.