iSCSI target colocate migration causes SQL service failover during witness rejoin in vSAN

Products

• VMware vSAN
• VMware vSAN 8.x

Issue/Introduction

• Intermittent iSCSI target disconnections occur in vSAN stretched cluster environments, specifically during events where the vSAN Witness host rejoins the cluster.
  
  
• This behavior results in brief ownership transitions that can trigger application-level failovers for iSCSI-dependent services like Microsoft SQL Server.
  

Environment

vSAN 8.0

vSAN Stretched Cluster configuration

 

Cause

If the iSCSI initiator does not have network routing to the hosts on both the preferred and non-preferred sites, it will be unable to log back in when the target owner resides on the alternative site.

This behavior is by design. In a vSAN stretched cluster, the iSCSI target owner selection algorithm uses an "eventual consistency" model for site affinity.

During cluster membership churn (such as a Witness rejoin), a new owner may be elected on either site.

Resolution

To ensure continuous availability of vSAN iSCSI targets, you must configure the environment to support owner transitions between sites.

Follow these steps to ensure proper configuration:

1. Verify that all iSCSI initiators (client machines) have routable network access to the iSCSI network on all hosts in the vSAN cluster, across both sites.
2. Ensure that the client machine can reach the iSCSI target IP regardless of which host currently owns the target object.
   
   
3. Increase the iSCSI initiator login timeouts on client nodes to recommended values in the below KBs to accommodate brief transitions during cluster re-election events.
   Multi-Paths IO(MPIO) Setting Recommendation for VMware vSAN iSCSI Target Service
   Using SQL Server Failover Clustering with vSAN iSCSI Target Service: Guidelines for supported configurations