Closed thesoundofsilent closed 1 year ago
@thesoundofsilent: Can you expand and clarify what do you mean by "one or more pods are killed by kubernetes system pods like kube-system/calico-node-8xc5f", please?
Any specific details as to the reason or why the case would help. Besides what we support at the level of deployer APIs (that interact with the API Server behind the scenes), it is unclear how we can help with the prevention of such a thing that is happening due to other components running in a given K8s cluster.
Looks like we should add a 'mirrored' deployer property for priorityClassName
. See https://kubernetes.io/docs/concepts/scheduling-eviction/pod-priority-preemption/
When using deployer partitions to launch 50+ or more pod, it is high probability that one or more pods are killed by kubernetes system pods like kube-system/calico-node-8xc5f
Is there anything we can do for this issue? is it possible to suppport priority class properties in SCDF to let me set the priority for the worker pod thus it will not be preeempted? or is there any suggestion?