VMSS Scale-In and Application Gateway Connection Draining – How to Gracefully Handle Requests?

Question

VMSS Scale-In and Application Gateway Connection Draining – How to Gracefully Handle Requests?

pipi 25

We are currently using VMSS along with Application Gateway to handle user web requests. Our question is regarding scale-in behavior: even with connection draining enabled, it seems ineffective because the VMSS instance is removed immediately during scale-in, causing ongoing requests to be abruptly terminated. Connection draining does not have a chance to take effect.

What is the recommended way to configure auto scale-out/in for VMSS so that Application Gateway’s connection draining can be effectively utilized? Or is there a better approach to ensure graceful shutdown of instances during scale-in?

1 answer

Your answer

Answer 1

Hi pipi,

Thanks for reaching out to the Q/A. VM Scale Set (VMSS) does not wait for Application Gateway’s connection draining to complete during scale-in. Although connection draining stops new incoming requests, VMSS terminates the instance immediately, resulting in active requests being dropped before completion.

This happens due to Application Gateway marks the backend instance as draining and stops sending new requests and VMSS does not wait for draining to finish and deletes the VM before existing sessions are completed.

Below are the recommendations

1.Enable VM Termination Notification:

This allows the VM to receive a shutdown signal (typically 30 seconds’ notice), so the application can stop accepting new requests and complete in-flight transactions.

Reference: https://learn.microsoft.com/en-us/azure/virtual-machine-scale-sets/virtual-machine-scale-sets-terminate-notification

2.Use VMSS Instance Protection:

This Prevents critical instances from being deleted during scale-in until it is safe to remove them.

Reference: https://learn.microsoft.com/en-us/azure/virtual-machine-scale-sets/virtual-machine-scale-sets-instance-protection

3.Verify Application Gateway Connection Draining Configuration:

Ensures new requests are stopped while allowing existing connections to finish.

Reference: https://learn.microsoft.com/en-us/azure/application-gateway/configuration-http-settings?tabs=backendhttpsettings

Additional Best Practices:

Review and adjust scale-in policies (Newest VM, Oldest VM, or Least CPU usage) based on workload behavior
Configure appropriate autoscale thresholds and cooldown periods to avoid aggressive scale-in
Monitor scale events using Azure Monitor to validate expected behavior
Consider leveraging Availability Zones to improve resiliency during scale operations

Reference: https://learn.microsoft.com/en-us/azure/virtual-machine-scale-sets/virtual-machine-scale-sets-scale-in-policy

Hope this helps! Please let me know if you have any queries.

pipi 25 Reputation points

2025-11-12T00:47:52.89+00:00

hi Jilakara,

Thank you for your response and recommendations. However, the application hosted on our VMSS handles certain requests that can take more than 5 minutes to complete. We are looking for alternative solutions that allow a more graceful shutdown of VMSS instances with a longer draining period.

Aside from using the built-in VMSS autoscale settings, are there other ways to trigger scale-in and scale-out operations based on metrics? For example, can we implement a manual or custom-triggered scale-in process where a 15-minute buffer period is applied before the instance is actually removed?
Jilakara Hemalatha 5,970 Reputation points Microsoft External Staff Moderator

2025-11-12T15:45:21.4766667+00:00

You’re right. When your application handles requests that can exceed a few minutes, the default VMSS scale-in behavior does not provide sufficient time for those requests to complete. Currently, VM Scale Sets don’t support extending the built-in scale-in delay beyond the short termination notification window (up to 15 minutes), which only signals impending termination but does not pause instance removal.

For a longer and more controlled drain time, consider these approaches:

Use Azure Automation Runbooks, Logic Apps, or Azure Functions to design custom scale-in workflows. These can be triggered by your own metrics or events, then include a buffer delay before calling the VMSS API to delete the VM. This gives your application time to gracefully complete ongoing requests.

Use Azure Monitor autoscale rules to trigger your automation workflows instead of direct scale-in. The workflow can add delay or a manual approval step before scaling, avoiding aggressive instance removal.

Enable VM Termination Notifications with shutdown scripts on your VMs. This ensures your application stops accepting new requests and completes active transactions before shutdown.

Reference: https://learn.microsoft.com/en-us/azure/azure-monitor/autoscale/autoscale-overview

https://learn.microsoft.com/en-us/azure/automation/automation-runbook-execution

https://learn.microsoft.com/en-us/azure/virtual-machine-scale-sets/virtual-machine-scale-sets-terminate-notification

Hope this helps Please let me know if you have any queries.

Share via

VMSS Scale-In and Application Gateway Connection Draining – How to Gracefully Handle Requests?

1 answer

Your answer