On GKE you want to simulate a single microservice suddenly crashing to validate system behavior. Which method should you use?

Istio fault injection is the most precise method for simulating a microservice crash. It allows you to inject specific faults, like delays or aborts (HTTP 500 errors), into requests for a particular service without affecting other services or the underlying infrastructure. Tainting a node and using anti-affinity would evict pods, not simulate a crash within a single microservice. Destroying a cluster node is too disruptive and impacts all services on that node, not just one microservice. Steering traffic away with Istio traffic management is a mitigation strategy, not a fault simulation.