We have recently fixed a problem involving SNAT ports exhaustion. The hotfix was simple and we needed only to allocate more ports to the backend pool.
Along with that fix, we changed the vNet to allow PaaS applications to go through the Azure Private Network. So we expected to see a drop in the used connections.
What we couldn't figure out is why even after changing the settings the number of SNAT ports in use kept the same even though the problem is gone.
We ssh into the node (AKS) and ran a simple netstat command and it reported very few open and time wait connections.
Below is an example of the VM using the most ports (10.240.0.128).
Azure Metrics:
Netstat Results:
