Is it expected that if multiple VMs on the same subnet that use default outbound access(no public IP, no load balancer, no NAT gateway) would stop functioning if one of the VMs on the subnet is configured with an explicit outbound access method?
https://docs.microsoft.com/en-us/azure/virtual-network/ip-services/default-outbound-access
I have 2 scenarios where I am seeing that VMs are unable to access the Internet through the default outbound access. In both scenarios, there is no NAT gateway configuration for the subnet they are on.
Scenario 1: there are 3 VMs that used default outbound access, they were primarily accessed over Azure VPN so they do not need public IPs or load balancers. They were working just fine with all 3 having Internet access using a Microsoft controlled IP. One of the VMs is now running a small workload that needs public access so it was given a public IP address. As soon as that VM was given a public it now had an explicit outbound access method according to the link above. But the other 2 VMs immediately lost Internet access, they cannot reach outbound to the Internet for anything despite no changes having occurred directly on them.
Scenario 2: there is a DMZ subnet with web servers that use a public load balancer, there are 2 servers currently in it that work fine. The load balancer pool is going to be expanded by adding 2 more web servers. The 2 new web servers have been added but they do not have any public Internet access. There are programs, management and security tools that need to be installed on the 2 web servers before they are ready to be deployed and added to the load balancer backend pool. Since they are not in the backend pool of the load balancer they should be able to use default outbound access but this is not working. They can't be added to the backend pool until they are prepared and ready to handle the traffic.
The document on default outbound access, from what I can tell, does not make any statement that defining an explicit outbound access method for one VM should have any effect on another. Except for the NAT gateway resource which applies to an entire subnet, adding a public IP address to one VM has nothing to do with another. If a VM is not in the backend pool of a load balancer, why would it be affected by its existence on the same subnet?
I can see the advantage of implementing NAT gateways to resolve this but this is an added cost that doesn't really seem justifiable if I don't require it.
