Experimenting with GKE Multi-cluster Inference Gateway and TPU Failover

GKE managed DRANET and Multi-cluster Inference Gateway are new enough that real-world failover patterns remain under-documented.

This experiment confronts the classic problem: what happens when a regional inference workload dies, but latency-sensitive users expect seamless continuity.

The author deliberately tests a cross-region TPU deployment with Gemma 3, not synthetic microbenchmarks, exposing the operational gap between claiming high availability and actually wiring it up with VPCs, fleets, and custom resource claims.