-
Notifications
You must be signed in to change notification settings - Fork 266
disable kubelet readonly access from outside cluster #227
Comments
Agree we should do this but: @aaronlevy said that this should be bound to localhost |
heapster / prometheus might still want access to this -- which would mean wherever they run would need access. At minimum this needs to be available on localhost because the pod-checkpointer uses this. |
@aaronlevy Can you link to the docs/resources on how we secure this port and get the certs to heapster, etc? |
This is the doc line from kubelet docs https://kubernetes.io/docs/admin/kubelet/
|
We have a new release pending for tectonic-prometheus-operator and the question is whether we switch to the secure port + certs or not. |
So I talked to @s-urbaniak about this as well. When it seemed an easy target we did implement this: https://github.com/coreos-inc/tectonic-prometheus-operator/pull/34 As the discussion stalled there, by now it seems unlikely that this is feasible for the next release, which is already pretty packed with heavy changes. @s-urbaniak, @alexsomesan, and I would suggest to defer this change to the next release to have more confidence in making the required changes. (Discussion is not always clearly distinguishing between completely disabling the read-only port vs just blocking it from outside of the cluster. I'm referring to the former above – the latter is no problem for us an what @s-urbaniak is working on AFAIK.) |
Currently masters and workers share a pretty open security group. Furthermore workers expose ingress traffic at critical k8s ports like 10250 and 10255. This fixes it by removing the common cluster default security group and specifying separate ingress/egress rules reflecting settings from the current tectonic installer. It also assigns only one security group for masters and workers. Fixes coreos#248, coreos#243, coreos#227
* modules/aws: tighten security groups Currently masters and workers share a pretty open security group. Furthermore workers expose ingress traffic at critical k8s ports like 10250 and 10255. This fixes it by removing the common cluster default security group and specifying separate ingress/egress rules reflecting settings from the current tectonic installer. It also assigns only one security group for masters and workers. Fixes #248, #243, #227 * Documentation/generic-platform: change flannel port to 4789 ... because that one is configured and recommended since it is the IANA based one. Tools like tcpdump then decode vxlan packets natively. The old port (8472) is retained as the default port in the kernel for backwards compatibility purposes only, see [1]. Other projects also switched to the new IANA assigned port. [1] http://lxr.free-electrons.com/source/drivers/net/vxlan.c#L43
resolved by #264 |
Port 10255 should only be accessible from other cluster master/worker nodes.
The text was updated successfully, but these errors were encountered: