Virtual Machine Scale Set Jump to 100% CPU Usage

Hey Heather, it sounds like one of your VMSS instances unexpectedly spiked to 100% CPU and then got restarted by the platform (you’re seeing that as a Platform-Initiated Downtime health incident), which led to about 15 minutes of reduced availability. Here’s how you can dig into the root cause and prevent it from happening again:

1. In the Azure portal, go to “Resource health” for that VM instance and also “Service health” for your subscription. Look for any planned or unplanned host maintenance events around 4/23/2026. Platform-initiated restarts often map to host updates or failures.
2. Review the Activity Logs for more context on the health incident and any underlying maintenance events.
3. Turn on or collect an on-demand report from Azure Performance Diagnostics (PerfInsights) on that VMSS instance. This will show which processes were consuming CPU and whether it was sustained vs. a sudden burst.
4. Pull CPU and memory metrics (e.g. Percentage CPU, CPU credits if you’re on a burstable SKU) over the last few days to see if this was a one-off spike or a build-up.
5. Did you deploy new code, patch the operating system, or change any dependencies around that date? Even a seemingly minor update can trigger a CPU-heavy startup job or leak.
6. Check your application or system logs inside the VM to see if there were long-running jobs (e.g. backups, indexing, antivirus scans) that kicked off right before the spike.
7. If you haven’t already, configure autoscale rules based on Average CPU (for example, add instances when CPU > 70% for 10 minutes). That way sudden bursts are absorbed by spinning up more VMs. Right‐size or redistribute load: If one instance is getting hammered, consider using a larger VM SKU or spreading that workload across more instances/services.
8. Enable rolling OS/image upgrades in your VMSS model so platform updates don’t reboot all instances at once. Use health probes to keep your load balancer from sending traffic to unhealthy VMs.
9. Run PerfInsights (continuous or on-demand) on that instance and share the report.
10. Correlate the spike with any maintenance events in Service Health. Review your autoscale and upgrade policy settings to ensure there’s enough buffer during platform operations.

Reference docs:

• PerfInsights for VMSS instances – https://docs.microsoft.com/troubleshoot/azure/virtual-machines/windows/perfinsights-and-scale-set-vm-instances

• Autoscale best practices for VMSS – https://docs.microsoft.com/azure/virtual-machine-scale-sets/virtual-machine-scale-sets-autoscale-portal

• Troubleshoot Azure Resource Health – https://docs.microsoft.com/azure/service-health/resource-health-overview