Azure strengthens reliability before failures become outages

For IT teams running business-critical workloads in Azure, reliability is not just about backup and failover design. Microsoft is highlighting how Azure itself is improving resilience at the platform layer with Azure Failure Prediction and Detection (AFPD), a service designed to identify and mitigate hardware issues before they interrupt virtual machines.

AFPD became fully operational in 2024 and now plays a growing role in reducing downtime risk across Azure Compute and select Storage scenarios.

What’s new with Azure AFPD

Microsoft says AFPD has improved prevention of hardware-related VM interruptions by 36% across the Azure fleet. The system combines machine learning, telemetry, and automated remediation to catch likely failures early.

Key capabilities include:

• Predict and detect: ML models and hardware detection rules monitor node behavior, SSD health, and other component signals.
• Mitigate automatically: When a node is flagged, Azure marks it unavailable for new workloads and may live-migrate eligible VMs to healthy hardware.
• Remediate faster: Azure removes the affected node from production, adds diagnostic context for technicians, and uses staged spare parts to speed repairs.
• Improve repair velocity: Microsoft reports an 80% increase in monthly node repair rate.

Azure also surfaces AFPD events through several customer-facing channels:

• Resource Health for per-VM health details and recommended actions
• Resource Graph Explorer for querying issues across large environments
• Event Grid for automation and integrations
• Scheduled Events for advance VM notifications via IMDS

Why this matters for administrators

Although Microsoft uses SAP as a high-profile example, AFPD is not limited to SAP. It also supports General Purpose, Specialized Compute, HPC/AI, and select Storage workloads.

For Azure administrators, the main takeaway is that platform reliability is becoming more proactive. Instead of discovering a hardware issue only after a reboot or outage, teams may receive an alert early enough to automate response actions or avoid disruption entirely.

This is especially important for workloads with strict uptime requirements, including ERP systems, line-of-business apps, databases, and third-party services hosted on Azure.

Recommended next steps

Microsoft recommends that organizations review how they consume AFPD notifications:

• Verify Resource Health is monitored for critical VMs
• Use Resource Graph to query health events across subscriptions
• Integrate Event Grid with Functions, webhooks, or monitoring workflows
• Ensure automation can process Scheduled Events, including notifications up to seven days in advance
• For SAP HA clusters, validate Pacemaker integration with Azure Scheduled Events

Bottom line

Azure AFPD shows how Microsoft is shifting reliability from reactive recovery to predictive prevention. For IT teams running mission-critical workloads, the most important action now is not enabling AFPD itself, but making sure your monitoring and automation can respond to the signals it already provides.