pkg/asset/internal: Add --cloud-provider to bootstrap manifests #464
Conversation
|
Can one of the admins verify this patch? |
k8s-ci-robot
added
the
cncf-cla: yes
|
rktbot run tests |
There was a problem hiding this comment.
We should add the --cloud-provider flags to the temp control-plane -- but we need to keep --configure-cloud-routes left to false. We assume a CNI plugin is going to handle the routing (not the cloud provider).
One reason for this is that using the AWS provider, for example, you are limited to 50 nodes because that's the max # of routes it can create (last I checked).
By assuming a CNI plugin to handle routes it makes the system more generic across providers.
pkg/asset/internal/templates.go
Outdated
| @@ -439,7 +440,7 @@ spec: | |||
| - --allocate-node-cidrs=true | |||
| - --cloud-provider={{ .CloudProvider }} | |||
| - --cluster-cidr={{ .PodCIDR }} | |||
| - --configure-cloud-routes=false | |||
| - --configure-cloud-routes=true | |||
There was a problem hiding this comment.
We assume the use of flannel (or another CNI plugin) - so we do not want the cloud-provider to also be trying to configure routes.
* Add --cloud-provider to bootstrap apiserver and controller-manager
|
@aaronlevy I've scoped back this change just to set --cloud-provider on bootstrap manifests. It will no longer solve #463 however. |
|
cc @aaronlevy |
This was a solution to #463, not necessarily the right solution. I'm not familiar with the original reasons we setconfigure-cloud-routes=trueto false (upstream default is true) or its impact on other providers.* Set --configure-cloud-routes true. Upstream description: "Should CIDRs allocated by allocate-node-cidrs be configured on the cloud provider. (default true)"