vMotion fails with 'Broken Pipe' or 'Limit Exceeded' due to reduced SndBufSize

Products

• VMware vSphere ESXi

Issue/Introduction

When attempting to vMotion a virtual machine from an ESXi host, the migration fails during the device checkpoint phase with the following errors:

• vSphere Client: 
  
       Failed waiting for data. Error 195887167. Connection closed by remote host, possibly due to timeout


• /var/run/log/hostd.log:
  
       VMotionInitiateSrc: Start message: Failed to write checkpoint data (offset 64762, size 262144): Limit exceeded.
  
  
• /var/run/log/vmkernel.log:
  
       WARNING: VMotion: 5262: <vMotion_ID>S: failed: Unable to send remote call to write checkpoint data  (262144 bytes @65116) to remote host <remote-host_IP>: Limit exceeded


• vmx Logs:

In(05) vcpu-0 - [vob.migrate.net.cb.wait.set.lowat.failed] Migration [a74e54d:<vMotion task id>] failed to set socket lowat: Limit exceeded.
In(05) vcpu-0 - Migrate: Caching migration error message list:
In(05) vcpu-0 - [msg.checkpoint.migration.writefail] Failed to write checkpoint data (offset 64762, size 262144): Limit exceeded.
In(05) vcpu-0 - Msg_Post: Error
In(05) vcpu-0 - [msg.checkpoint.migration.writefail] Failed to write checkpoint data (offset 64762, size 262144): Limit exceeded.

Cause

The advanced configuration parameter /Migrate/SndBufSize (vMotion network send buffer) has been set to a value lower than the size of the data chunks vMotion transmits.

Resolution

To restore vMotion functionality, the affected ESXi host must have the custom buffer limit reverted to the system default value of 562540.

1. Enable SSH on the affected ESXi host and log in as root.

2. Verify the current value of the buffer size:
   
       esxcli system settings advanced list -o /Migrate/SndBufSize

3. Revert the value to the system default:
   
       esxcli system settings advanced set -o /Migrate/SndBufSize -i 562540 
   
   

Note: This change is dynamic and applies immediately to all VMs on the host. No ESXi reboot or Virtual Machine power cycle is required.