🔒 Paid
Load Balancer Imbalance
A wave of returning, authenticated users is landing on the product, and the fleet behind the load balancer is healthy in aggregate — but two of the four instances are visibly carrying far more request volume, connections, and latency than the other two. Every customer-facing request must keep succeeding, and the business cannot tolerate forcibly dropping in-progress sessions as a routine response. Diagnose why load is concentrating on a subset of otherwise-healthy instances and bring the fleet back to a durable, evenly-shared state.
- Difficulty
- Senior
- Expected Time
- ~16 min
- Category
- API & Traffic
Skills
- Diagnosing session-affinity-driven traffic skew
- Distinguishing a routing/pinning problem from a capacity problem
- Recognizing when adding instances cannot relieve an existing pin
- Evaluating a forced-reset mitigation against a structural routing fix
System Architecture
- Clients
- Load Balancer
- Instance A
- Instance B
- Instance C
- Instance D
This is a Pro scenario. Upgrade to Pro to unlock it.
Upgrade to ProBack to Incidents