Single Compute or Node Failure
When a single compute or node fails, one of the cdl-ep pods goes down and tries to schedule on the other session nodes. If all the other session nodes already have a cdl-ep, pod scheduling will not take place since there is node anti-affinity defined for cdl-ep pods. So, with single node failure, the remaining cdl-ep pods handle the traffic and might get loaded with the many requests. After recovering from the node failure, the cdl-ep pod, which was in the pending state earlier, starts scheduling on the recovered node to process the requests.