R. P. Taylor

Before: ``` TASK [kubespray-defaults : Gather ansible_default_ipv4 from cluster-prod-k8s-node-f54] *************************************************************************************** ok: [cluster-prod-k8s-master-1 -> cluster-prod-k8s-node-f54] Thursday 15 September 2022 23:45:24 +0000 (0:00:02.927) 0:08:35.094 **** TASK [kubespray-defaults : Gather ansible_default_ipv4 from cluster-prod-k8s-node-f55]...

@cristicalin for now the main improvement is cleanup/simplification and adding `unique` to avoid running multiple times on the same nodes. This can make it maybe just a couple seconds faster,...

I used this test.yml play to evaluate the impact of the change. This is a sort of benchmark that represents 2 out of the 12 times that the cluster.yml play...

@cristicalin or @oomichi could you please take a look?

Interesting ... I don't understand how it could be doing anything with the equivalent of `when: false` on every task but I'll take a closer look. Oh of course, the...

I'm not sure if Kubespray is the best place to do that, especially if you get into dependencies on the NVIDIA drivers and which drivers and setup should be used...

> I could reproduce #9019 and am working on a fix. It's mind-boggling. That looks serious, would be good to have a fix in 2.20 otherwise clusters could still have...

@floryut what about including https://github.com/kubernetes-sigs/kubespray/pull/9026 , https://github.com/kubernetes-sigs/kubespray/pull/9109?

/remove-lifecycle stale

IMHO Helm is a good choice, relatively lightweight and simple while being sufficiently flexible.