Amazon EKS is built for high reliability. ReplicaSets, liveness probes, and the cluster autoscaler are all designed to keep workloads running when a node disappears. That safety net is real, and it is also the reason most teams have never checked whether it holds for their own services.

Node loss on EKS is routine. Spot instance reclamation, version upgrades, and autoscaling events remove nodes every week, and your cluster absorbs them quietly until the day it doesn't. Install the Gremlin Agent on your worker nodes to prepare for that day. With Gremlin, you can:

• Prove your pods reschedule cleanly. Run a shutdown experiment to remove a node while you monitor your services. You'll find out whether pod disruption budgets were ever set, how long it takes Kubernetes to schedule replacements onto nodes with capacity, and how long your service runs degraded before recovering.
• Test the upgrades you can't opt out of. Surge upgrades cycle nodes faster than Kubernetes can migrate slow-starting containers. Applying CPU and memory pressure during a rolling replacement shows you whether your cluster has the headroom it needs before AWS schedules the next version bump.
• Validate redundancy across availability zones. Blackhole experiments isolate a zone so you can confirm the remaining zones carry your traffic. Multi-AZ node groups look correct in Terraform, but only a test proves your topology constraints distribute pods the way you expect.