Virtual machines using Multus macvlan experience network disconnects and ping failures when on different ESXi hosts
Article ID: 432502
Updated On:
Products
Issue/Introduction
You may experience network disconnections and ping failures for virtual machines utilizing a secondary VMXNET3 vNIC on a storage PortGroup configured for macvlan. This issue specifically occurs when the communicating VMs reside on different ESXi hosts. If you migrate the VMs to the same ESXi host using vMotion, network communication is temporarily restored.
When verifying the issue within the guest OS of the affected VM, running an ARP check (e.g., arp -a) will show incomplete ARP entries for the target IP addresses:
? (##.##.##.##) at <incomplete> on odf-shimEnvironment
- Red Hat OpenShift (utilizing Multus and macvlan)
Cause
This issue is caused by the physical network infrastructure dropping the ARP broadcast traffic.
Packet captures at the ESXi host uplinks (UplinkSndKernel and UplinkRcvKernel) confirm that ARP broadcasts successfully leave the source ESXi host's physical uplink but are never received by the destination ESXi host's physical uplink. Because the packets leave the VMware environment but fail to arrive at the destination host, the packet drop is occurring within the physical network tier (e.g., the physical switch is not properly forwarding macvlan ARP broadcasts between the hosts).
Resolution
To resolve this issue, you must correct the configuration on your physical network hardware.
- Engage your physical network vendor or internal network administrator to review the physical switch configuration.
- Identify the physical switch ports that are connected to the affected ESXi host uplinks.
- Update the physical switch port configurations to ensure they properly allow and forward macvlan ARP broadcast traffic across the network.
Additional Information
Feedback
