KB: Kubernetes how to bypass Tolerations/Taints/Affinity
Note, by design it is "Scheduler" pods responsibility to place the pod in the correct Node. But if you manually specify the nodeName selector you are bypassing the "Scheduler" pod and that means you skipping the Tains/Tolerations and any affinity. Basically skipping the "Scheduler Logic" all together. if you manually specify the nodeName field in a Kubernetes Pod spec, you are bypassing : Taints and tolerations Node affinity Scheduler logic entirely Here's what happens: When you set .spec.nodeName in a Pod or Deployment template, Kubernetes does not use the scheduler to determine where the pod should run. The pod is directly assigned to that node . As a result: Tolerations are ignored – the pod is placed even if it doesn’t tolerate the node’s taints. Affinity rules are ignored – because the scheduler isn’t consulted at all. NodeSelector and affinity constraints are not evaluated – they are bypassed . What can go ...