An Azure network security service that is used to protect Azure Virtual Network resources.
Sandeep Darak Hi & thx for join me at Q&A platform,
fact that everything works when u remove both UDRs is a pretty useful clue. It strongly suggests the issue is related to routing through Azure Firewall, although it doesn't necessarily prove asymmetric routing yet.
Your overall design is valid, and global VNet peering supports gateway transit, so having the hub and spoke in different Azure regions isn't automatically a problem. The important part is making sure both directions of the connection pass through the same Azure Firewall.
For traffic originating from on-premises, the expected path is:
On-prem VM > VPN Gateway > Hub GatewaySubnet > Azure Firewall > Spoke VM
The return path should be:
Spoke VM > Azure Firewall > Hub VPN Gateway > On-prem VM
If the return traffic bypasses the firewall, its stateful connection tracking won't see both directions of the session, which can cause TCP connections such as HTTPS and RDP to fail.
I'd start by checking Effective routes and Next hop in Azure Network Watcher for the spoke VM. Test the destination 192.168.0.x and verify that the selected next hop is the Azure Firewall private IP, 10.5.0.4. Your 0.0.0.0/0 UDR should handle this traffic when gateway route propagation is disabled, but it's worth confirming what Azure is actually selecting.
As a troubleshooting test, u could replace the spoke's default route with a more specific route:
192.168.0.0/16 > Virtual Appliance > 10.5.0.4
This makes the intended return path explicit without redirecting all other outbound traffic through the firewall. It's a useful way to isolate the hybrid connectivity issue before deciding on the final routing configuration.
Next, check the hub GatewaySubnet route table. Your 10.6.0.0/16 > 10.5.0.4 route is the right idea, but confirm that the effective route for the actual spoke VM address points to the firewall. Also make sure gateway route propagation remains enabled on the GatewaySubnet route table.
Another important check is the firewall's route back to 192.168.0.0/16. Azure Firewall should be able to reach that network through the hub VPN gateway. If the firewall doesn't have a valid route back to the simulated on-premises network, the connection can fail even when the gateway-to-spoke direction is correct.
I'd also enable Azure Firewall Network Rule logs and attempt a TCP connection from the on-prem VM to the spoke VM on port 443 or 3389. If there's no corresponding firewall log entry, the traffic may not be reaching the firewall at all. If u see an Allow entry, check whether the TCP handshake completes using packet captures at the source and destination VMs. This will help distinguish routing problems from firewall policy, guest OS firewall, or application listener issues.
One thing worth mentioning: successful ping doesn't guarantee that TCP traffic is following the same path or that the application ports are reachable. ICMP connectivity alone isn't enough to rule out asymmetric routing, a firewall rule mismatch, or a Windows/Linux host firewall restriction.
I wouldn't remove the UDRs permanently, bc that appears to restore connectivity by bypassing the intended firewall inspection path.
My main suspicion is that the effective routes aren't producing the symmetric path u expect, but I'd verify that before changing the architecture.
Could u share the effective route selected for 192.168.0.x on the spoke VM, along with whether Azure Firewall logs show an Allow or Deny entry for the failed TCP connection? Those two details should narrow this down considerably.
rgds,
Alex
p.s. if my answer is help - pls mark it as an answer