During FC fabric upgrade, vmfs datastores unexpectedly become inaccessible

Products

• VMware vSphere ESXi

Issue/Introduction

- As part of fabric upgrade, one side of the fabric is taken down, upgrade is completed, and then that side of the fabric is brought back up. 

- The process is repeated for the second side of the fabric. 

- Expected behavior at ESXi level, is that at any time datastores would lose no more than half their paths.

- Some datastores unexpectedly become inaccessible on ESXi hosts during the upgrade process. 

Environment

VMware vSphere ESXi (all versions)

Cause

This will arise if:

1) prior to upgrade an ESXi host has paths to a LUN only over one side of the fabric. In that case, when that side of the fabric goes down, the ESXi host will lose its only paths to that LUN

2) following completion of upgrade to the first side of the fabric, paths that went down do not recover as expected. In that case, when the second side of the fabric, ESXi will lose all paths to LUNs. 

Resolution

1) Prior to upgrade, validate that all devices have the expected number of active paths.

2) Prior to proceeding with upgrade of the second side of the fabric, validate that all expected paths are active at vSphere level.