Resilient on-premises AI workloads on Kubernetes with hyperconverged infrastructure