🔒 Paid
Deployment Regression
A new version was deployed behind a partial rollout and is now showing elevated errors, latency, and memory usage — but only for the portion of traffic currently routed to it; the old version is unaffected so far. The new version also shipped other, non-regressed improvements that matter to the business, and it shares a downstream dependency's connection pool with the old version. Diagnose which part of the new version is actually responsible for the regression and bring customer impact back to a durable, sustainable state without assuming a full rollback is the only acceptable move.
- Difficulty
- Senior
- Expected Time
- ~18 min
- Category
- Deployments & Reliability
Skills
- Distinguishing rollback from traffic isolation and feature disabling
- Recognizing when scaling absorbs a regression's cost without fixing it
- Reasoning about shared-dependency blast radius across versions
- Evaluating comparably safe alternatives to a full rollback
System Architecture
- Traffic
- Load Balancer
- Old Version
- New Version
- Shared Dependency
This is a Pro scenario. Upgrade to Pro to unlock it.
Upgrade to ProBack to Incidents