Hello, I am a newcomer to Kubernetes, we are investigating to deploy our computing platform on Kubernetes. Our computing cluster is a master-slave architecture, the master takes charge of managing all the master and slave pods in a low-level semantics, said not by Deployment or ReplicaSet that Kubernetes provides. Also, the master needs to keep state for High-Availability mode. I have two questions:
- If we send a request to Kubernetes master to create a Pod with a specified label, is that possible that in some periods we can’t find that Pod via successfully polling the Kubernetes API with that label?
- If we set the RestartPolicy of the master pod to “OnFailure”, is it possible that in some extream situation or edge exception cases that the old master pod is not deleted but a new one is deployed to run(), so there are are two master pods running at the same time for a short period? I concern about it because, in our design, we must ensure that at any time there could be at most one master pod running to avoid the master racing condition.