Load balancer marks both backends down
lb-app at 10.1.1.10 times out for everyone since last night.
- Get vm-client reaching lb-app on 443
- Get vm-web1 and vm-web2 both passing the health probe
- Keep UFW running on both backendsbroken
- Keep SSH from vm-client to the backends blockedbroken
Background
A hardening script ran on the web servers, vm-web1 and vm-web2, last night. Since then lb-app at 10.1.1.10 times out for everyone, vm-client included. The network team has checked nsg-web: it allows the clients and the AzureLoadBalancer tag on 443. Both backends have to be healthy again, and UFW stays on.
- 1See it fail
Prove the problem yourself before you change anything. Reports are often wrong.
VM commands forVMs: Linuxssh vm-clientWhat's this?
Opens a shell on vm-client, so the next commands run on that VM. In this lab it's a shortcut. In real Azure you'd connect through something like Azure Bastion, or SSH to an address you can reach.
nc -zv 10.1.1.10 443What's this?
netcat, a standard Linux tool. Tries to open a TCP connection and reports whether it worked.
-zonly check whether the port answers. Send no data-vprint the result in words10.1.1.10where to connect443the port to try
Ticks off when the test runs on vm-client.
- 2Find where it stops
- 3Name the cause
- 4Fix it and test again
- 5Explain it
Sources on Microsoft Learn (5)
?
- Resource group, the outer dashed frame
- Virtual network, with its address space top right
- Subnet, with range and usable IPs (5 reserved)
- Network security group. A small shield on a VM's NIC means the NIC has one too
- Route table on the subnet
- VM Appliance Private endpoint
- Peering Connected Initiated or access off
- Goal working Goal broken, X marks where it stops
- Private DNS zone, dotted lines go to linked VNets
- Orange pulse: what your last command changed. Drag to pan, Ctrl and scroll to zoom.
vm-client → lb-app-fe:443should work: not working