Closed moonek closed 2 years ago
Sounds like a bug to me in NSX. CPV should just be request ip for the service, and NSX-T manager should be allocating the IP based on availability, from your description, NSX-T allocates IP and it failed to clean up all of them
Let me ask my NSX-T foks
@lubronzhan This is an important issue for us. Thanks for checking further.
Hi @moonek I talked to them, they said you should talk to NSX-T GSS, the support guy. Collect bundle and file issue to NSX-T
It is unclear whether CPV can solve it, so I will close it. I will reopen if there is any update in the future.
What happened?
When using NSX-T based LoadBalancer, I encountered a problem where the number of IPs allocated to the IP Pool and the number of VIPs actually used did not match. This results in the following issues that I checked through the NSX-T Web Console and REST API.
We experienced this issue twice a year in the staging environment.
I wonder if it is a cloud provider issue or an NSX-T issue, and I want to know a solution or workaround.
What did you expect to happen?
I hope that the IP assigned to the IP Pool matches the number of k8s services(w/ type: LoadBalancer) created.
How can we reproduce it (as minimally and precisely as possible)?
In the staging environment, k8s services(w/ type: LoadBalancer) were frequently created/deleted.
Kubernetes version
v1.20.0
Cloud provider or hardware configuration
v1.20.0
Others
NSX-T version: 3.1.2.1.0.17975795 vSphere version: 7.0.2.00200