Virtual Machine hangs during vSphere Replication with error: No connection to VR Server: Disk size mismatch

Products

• VMware vSphere ESXi
• VMware Live Recovery

Issue/Introduction

The primary product impacted is VMware ESXi utilizing the vSphere Replication component.

Symptoms:

• The virtual machine randomly enters a hung state.
• The last event reported in the vmware.log of the impacted virtual machine is destroying the virtual scsi device.
  
  2026-06-26T17:42:50.766Z In(05) vcpu-0 - Destroying virtual dev for scsi0:0 vscsi=13##############
2026-06-26T17:42:50.766Z In(05) vcpu-0 - VMMon_VSCSIStopVports: No such target on adapter
2026-06-26T17:42:50.880Z In(05) vcpu-0 - Destroying virtual dev for scsi0:1 vscsi=13##############
2026-06-26T17:42:50.880Z In(05) vcpu-0 - VMMon_VSCSIStopVports: No such target on adapter
  
  
• Review of the /var/run/log/vmkernel.log file reveals that the task of destroying virtual scsi device is stuck as there is a full sync termination task in queue
  
  2026-06-26T17:42:50.766Z In(182) vmkernel: cpu43:3038679)VSCSI: 8589: handle 13##############(GID:8774)(vscsi0:0):Destroying Device for world 3038140 (pendCom 0)
2026-06-26T17:42:50.880Z In(182) vmkernel: cpu43:3038679)VSCSI: 8589: handle 13##############(GID:8775)(vscsi0:1):Destroying Device for world 3038140 (pendCom 0)
2026-06-26T17:42:50.890Z In(182) vmkernel: cpu43:3038679)Hbr: 1503: Terminating sync task (diskID=RDID-c7######-####-####-####-############) (resumeIndex=406992896)
  
  
• Replication status displays the following error: No connection to VR Server: Disk size mismatch.
• The issue often manifests during VM lifecycle operations such as snapshot removal, power-off, or guest OS restarts while a full-sync operation is in progress.

Environment

• vSphere Replication 9.0, 9.0.0.1, 9.0.1, 9.0.2, 9.0.2.1, 9.0.2.2, 9.0.2.3

• VMware Live Recovery 9.0.4, 9.0.5, 9.0.5.1

• VCF Protection and Recovery 9.1, 9.1.0.0100, 9.1.0.0200

• VMware ESXi 8.x

Cause

The hung state is caused by a race condition between the checksum throttling mechanism and the full-sync cancellation logic within the hbr_filter.

When a full-sync is active, hbr_filter uses a read-ahead mechanism (defaulting to 8192 regions) to optimize checksumming. If a VM lifecycle operation (like a restart or snapshot consolidation) triggers a full-sync cancellation while the throttling mechanism is active, the resource management logic can become deadlocked. The "Disk size mismatch" error is a side effect of the blocked hbr_filter detach process, where the primary site disk count begins to decrease while the secondary site still expects the original disk count during session re-establishment.

Resolution

A permanent fix for this race condition is expected to be released in version 9.1.2 of Protection and Recovery.

Workaround

To mitigate the issue, disable the read-ahead checksumming mechanism on all source and target ESXi hosts. This prevents the checksum throttling sub-system from entering the code path where the race condition occurs.

1. Log in to the ESXi host via SSH.

2. Execute the following command to set the advanced configuration parameter to 0:

   esxcli system settings advanced set -o /HBR/ChecksumRegionMaxReadAhead -i 0

3. Verify the setting has been applied:

   esxcli system settings advanced list -o /HBR/ChecksumRegionMaxReadAhead

Note: A host restart is not required for this change to take effect.