At present, I have taken over a cluster. The default deployment mode is three keepers, three sentinels, and three proxies, and DaemonSet Controller is select to control the pod.
At present, I feel that current deployment method is not very reasonable. I have the following ideas. I don’t know if it is reasonable. I hope to get help:
Three keeper pods are deployed using DaemonSet (node label selection, one per node). The problem is not particularly big, but I think two keeper pods should be enough, and there is no need to deploy three
Three sentinels are deployed using DaemonSet (node label selection, one for each node), which is not very reasonable. I feel that Deployment Controller should be used, so that when one pod is down, the other can be pulled up quickly, and no node label is required bound. Does Sentinel really need three at the same time? Is one enough? Like redis sentinel, does this sentinel need more than one to ensure the correct selection of master? Does sentinel itself need a leader role?
Three proxies are deployed using DaemonSet (node label selection, one per node), which is not very reasonable. I feel that it is enough to use Deployment Controller and one replica, and no node label binding is required.
At present, I have taken over a cluster. The default deployment mode is three keepers, three sentinels, and three proxies, and DaemonSet Controller is select to control the pod. At present, I feel that current deployment method is not very reasonable. I have the following ideas. I don’t know if it is reasonable. I hope to get help: