search

Azure / AKS

Azure Kubernetes Service
https://azure.github.io/AKS/
1.97k stars 306 forks source link

[Feature] Support for excluding AKS node pools from Azure LoadBalancer #4167

Open kszyntar opened 7 months ago

kszyntar commented 7 months ago

Currently there is no supported option to exclude AKS node pools from Azure LoadBalancer. As a result, regardless of how node pools are shaped, all nodes are added to the Load balancer. Depending on externalTrafficPolicy on the cluster, this either results in extra traffic hops or failed health checks.

Describe the solution you'd like A clean way to remove a given AKS node pool from Azure LoadBalancer - either by explicit support in AKS API or with standard k8s label exclude-from-external-load-balancers It is acceptable if this support comes with certain limitations - e.g. has a NAT Gateway for outbound as a prerequisite.

Describe alternatives you've considered Standard method for doing so would be with labeling a node with node.kubernetes.io/exclude-from-external-load-balancers=true - but as mentioned here it is not supported with AKS (implying - label works only with self-managed k8s). There is no alternative way to accomplish this currently.

Additional context Such issue has already been mentioned: https://github.com/Azure/AKS/issues/3545 but resulted only in a documentation update.

treyhendon commented 7 months ago

This is something that our organization is extremely interested as well.

kevinharing commented 5 months ago

I noticed this new feature. Is it possible to implement/utilize this now in AKS?

https://azure.microsoft.com/en-us/updates/public-preview-azure-load-balancer-now-supports-admin-state/

mailbox171 commented 2 months ago

Hello, this is VITAL. This property cannot be skipped.

mailbox171 commented 2 months ago

Hello, I want to reiterate my plea. This feature cannot wait any longer.

I have large node pools with autoscaling and SPOT VMs. The situation is very .. dynamic, and I get occasional connection reset, which in the grpc world are real DISASTERS.

I tried externalTrafficPolicy: Local, but it has real bugs. Any update to the situation, topology, .. it breaks, and the workaraound is update the service to eTP: Cluster

Please help!!!