HA fails with IBM SVC storage, VMs freeze and Datastore stays connected when inter-site ports are disconnected

Products

• VMware vCenter Server

Issue/Introduction

VMs freeze and Datastore stays connected when trying to failover from Source to Destination hosts with a IBM SVC Storage Backend(Datastore) 

 

Datastore never disconnects or shows as offline

Environment

• vCenter 7.x
• vCenter 8.x

Cause

All Paths Down (APD) is not triggered for LUNs behind IBM SAN Volume Controller (SVC) target even when no paths can service I/Os 
In an ESXi configuration with multiple paths leading to LUNs behind IBM SVC targets, in case of connection loss on active paths and if at the same time the other 
connected paths are not in a state to service I/Os, the ESXi host might not detect this condition as APD even as no paths are actually available to service I/Os. As 
a result, I/Os to the device are not fast failed. 

Resolution

The fix is disabled by default. To enable the fix, set the ESXi config option Scsi.ExtendAPDCondition to 1 in the advanced settings of the host -> click Edit -> you can search for APD in the filter to find this


Also lower the Misc.APDTimeout timer (minimum = 20secs) 

Additional Information

This will not be instantaneous failover but will take a short period of time, However it will not keep the Datastore connected like before and will force a HA failover - this is depending on your setting on the HA, and APD settings in the HA and aslo how many seconds you set on the Misc.APDTimeout timer.