ESXi hosts experience All Paths Down (APD) during Cisco UCS Fabric Interconnect updates

Products

• VMware vSphere ESXi
• VMware vSphere ESXi 8.0

Issue/Introduction

When performing infrastructure updates on Cisco UCS Fabric Interconnects (FI), VMware ESXi hosts may experience brief (6–10 second) "All Paths Down" (APD) events. While this behavior is often associated with the FI reboot process, it is frequently exacerbated by misconfigurations in host-side storage multipathing. This article details the primary cause—the lack of fabric evacuation—and provides necessary host-side tuning to prevent APD events during Cisco UCS maintenance.

 

 

Environment

• VMware vSphere ESXi 7.x, 8.x
• Cisco UCS Manager (All supported versions)
• Cisco MDS Fibre Channel Switches
• Storage: All Fibre Channel (FC) and NFS configurations

Cause

• By default, if a Fabric Interconnect is rebooted without evacuation, the physical links (vNICs) on the blades drop before I/O can be failed over to the redundant fabric. An APD event occurs when:

  1. Fabric Evacuation is not used: Traffic is not gracefully migrated, causing a sudden link-down event.
  2. Lack of Path Redundancy: The host does not have active, working storage paths on the redundant fabric. The absence of a secondary path prevents the host from riding through the maintenance window, causing the device to enter an APD state rather than simply losing path redundancy.

 

Resolution

• To prevent APD events during UCS maintenance, ensure that traffic is gracefully failed over to the redundant Fabric Interconnect before the target Interconnect reboots.
  
  
• Contact Cisco for advice regarding the Evacuate setting and functionality.