About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
A vMotion that finishes without error only proves that the VM's memory and disk ownership moved. It says nothing about whether the destination host can carry the VM's traffic. When a VM goes dark the moment the task completes, the cause is almost always in one of three places: the VLAN on the destination port group, the uplink the VM is placed on, or the physical switch's record of which port the VM's MAC address lives behind.
Work through them in that order. The first two are visible in the vSphere Client. The third needs a look at the physical switch's MAC address table.
What the error means
There is no error message. The symptom is that a continuous ping to the VM works before the migration and fails as soon as it completes. Broadcom describes it as VMs that "lose network connectivity immediately following a vMotion migration or vSphere HA failover event" (KB 318873).
Three mechanisms produce it:
- Different VLAN on the destination. The destination host's port group for the VM network carries a different VLAN ID, or the physical port does not allow that VLAN. Broadcom documents a case where the vMotion task completes but connectivity to a vCenter Server VM is lost because the destination host has the wrong VLAN ID on the VM network (KB 421993).
- Wrong uplink. If the teaming policy can place the VM on an uplink that is not trunked for the VM's VLAN, the traffic is dropped. Broadcom documents a cluster where uplinks 3 and 4 carried only storage traffic, yet all four uplinks were Active on the VM port groups. After a vMotion the load-balancing policy could hash the VM's MAC address onto a storage uplink, creating "a Layer 2 traffic blackhole" (KB 434979).
- Stale MAC table. After a move, the physical switch port on the destination host must learn the VM's MAC address. Per KB 343401, vSphere virtual switches do not send GARP. They send RARP packets to update the upstream switches' MAC-to-port mapping, and with the Notify Switches setting at Yes, ESXi sends them in scenarios including failover and vMotion. If the physical network drops or ignores those packets, or updates its table late, traffic keeps going to the source host (KB 426004).
Likely causes
| Cause | What you see | How to confirm |
|---|---|---|
| Port group VLAN ID differs on the destination host | Fails only when the VM lands on that host; works when it moves back | Compare the VLAN ID of the VM's port group on source and destination (vSphere Client) |
| VLAN not allowed on the destination host's physical trunk | Same-host VMs on the same port group can talk; nothing crosses the physical network | Read the allowed-VLAN list on the physical switch ports for the destination host's uplinks |
| Teaming places the VM on an uplink not trunked for VM VLANs | Intermittent: disconnecting and reconnecting the vNIC sometimes fixes it | Teaming and failover on the port group; esxtop network view shows the uplink in use |
| Physical switch ignores or delays RARP | Ping fails until the VM sends traffic itself or the vNIC is reconnected | pktcap-uw shows RARP leaving the uplink, but the switch MAC table still points at the old port |
| Notify Switches set to No | Same as above | Teaming and failover setting on the port group |
| Switch port without PortFast or Edge Port | Brief or longer blocking after topology changes | Spanning tree state of the destination host's switch ports |
Prove the cause before changing anything
These commands are taken from the Broadcom pages cited and have not been run.
1. Confirm the migration really finished. On the destination host, search the VM's log for the completion entry:
grep "MigrateSetStateFinished" /vmfs/volumes/<datastore>/<vm_name>/vmware.log
KB 318873 gives the expected entry as MigrateSetStateFinished: type=2 new state=MIGRATE_FROM_VMX_FINISHED. If it is absent, you are not looking at this fault. See vmotion-fails-compatibility-error-evc-vmkernel.
2. Compare the VLAN IDs. In the vSphere Client, open the VM's port group and read its VLAN ID. Do the same for the port group the VM used on the source host. For a distributed switch the port group is one object, so the difference is in the physical ports behind each host. For standard switches, compare the port group on each host under Configure > Virtual switches. A good result is the same VLAN ID everywhere. A bad result is a different or missing ID on the destination.
3. Find which uplink the VM is on. On the destination host run esxtop and press n. KB 318873 uses this view to identify the uplink a VM is using. Compare that uplink with the ones that are trunked for the VM's VLAN. A bad result is a VM whose traffic leaves on an uplink that carries only storage or management VLANs.
4. Find the switchport and capture. Find the VM's switchport:
net-stats -l | grep "<VM_Name>"
Then capture at the vNIC and at the uplink, as KB 318873 documents:
pktcap-uw --switchport <Switchport_ID> --capture VnicTx,VnicRx --mac <VM_MAC_Address> --ng -o /tmp/vm_vnic_arp.pcapng
pktcap-uw --uplink <vmnic_name> --capture UplinkSndKernel --mac <VM_MAC_Address> --ng -o /tmp/uplink_vmnic_rarp.pcapng
Broadcom does not print sample output for these captures, so read the result as follows. If the uplink capture shows RARP frames for the VM's MAC leaving the host, ESXi did its part and the fault is on the physical side. If nothing leaves, check Notify Switches and the uplink choice from step 3.
5. Read the physical switch. Ask the network team to show the MAC address table entry for the VM's MAC before and after a vMotion. KB 318873 says to monitor MAC learning in this way: the entry should move to the destination port. A bad result is an entry that stays on the source host's port, or no entry at all. Also check the allowed-VLAN list on the destination uplink ports.
6. Test the vMotion network itself (only if the migration was slow or failed). vmkping -I vmkN <destination-vmk-IP> -s <size> -d -c 3, where -s is the MTU minus 28 (1472 for MTU 1500, 8972 for 9000), with -S vmotion added if the vmk uses the vMotion TCP/IP stack (KB 344313). The KB shows a good result as replies with 0% packet loss and a bad result as no replies and 100% packet loss. This tests the vMotion path, not the VM's network.
Example (illustrative): a VM app01 on esx01 is moved by DRS to esx02. The task succeeds and the ping to 10.0.20.15 fails. The vmware.log on esx02 shows MIGRATE_FROM_VMX_FINISHED. esxtop shows the VM on vmnic3, an uplink used only for the storage VLAN. The port group has all four uplinks set to Active. The cause is the teaming policy, not the migration.
Fix
VLAN ID or physical trunk mismatch
Set the destination port group's VLAN ID to match the working source, and have the network team add the VLAN to the allowed list on the destination host's physical ports (KB 421993; KB 318873).
Risk: changing a port group VLAN ID on a host that already carries running VMs interrupts those VMs. Make the change in a maintenance window, or migrate other VMs off that host first. To back out, restore the previous VLAN ID.
Uplink teaming
For a distributed switch, in the vSphere Client open the affected switch, right-click the VM port group, choose Edit Settings, open Teaming and failover, and move the storage-dedicated uplinks to Unused. Repeat for every VM port group that includes them (KB 434979).
Disruptive step: a teaming change can move running VMs between uplinks and cause a short network interruption. Check first that the uplinks left Active are trunked for every VLAN the port group serves. To back out, restore the previous uplink order.
Notify Switches and RARP
In the port group's Teaming and failover settings, set Notify switches to Yes (KB 318873). Ask the network team to confirm that Dynamic ARP Inspection or Storm Control is not dropping or rate-limiting GARP and RARP, that the VLAN is trunked on both the source and destination uplinks, and that the ESXi-facing ports use PortFast (Cisco) or Edge Port (KB 318873).
Broadcom's KB 426004 lists a stop-gap for delayed MAC learning: a continuous ping from inside the guest to its default gateway, so that the physical switch learns the VM's new location. Treat this as a workaround while the switch is fixed. The same KB says to confirm the switch ports carry the correct VLANs and are not in a discarding or blocking state for the VM's MAC.
Confirm the fix
- Start a continuous ping to the VM from a machine outside the host.
- vMotion the VM to the previously failing host and back.
- The ping should keep answering through both moves, with at most a missed reply or two. KB 318873 says the switch MAC table entry should move to the destination port; check that on the switch.
- Test every host in the cluster, not only the one that failed. A VLAN or trunk mismatch is usually specific to one host.
What to watch afterwards
- Add a post-vMotion ping test to the checklist for new hosts, and for any change to a physical switch port or a port group.
- Keep the allowed-VLAN lists on all host-facing switch ports identical, so any host can take any VM.
- Do not leave storage-only uplinks Active on VM port groups.
- If failures return after switch maintenance, check the Notify Switches setting first and then the switch's ARP and storm-control configuration.
Read next
Related articles you may want next.
- TroubleshootingvDS Says a VLAN Is Unsupported on an Uplink the Port Group Does Not UseInterpret vSphere 8 VLAN/MTU health-check warnings against per-port-group uplink selection, without dismissing real trunk or failover faults.Read article
- VMwarevMotion fails: telling CPU compatibility, device dependencies and the vMotion network apartvMotion failing after adding a host or changing networks? Classify the error, test the vmkernel path with vmkping, find device blockers and plan EVC safely.Read article
- VMwareVM Network Adapter Disconnected After a Reboot: Read Its Device StateA VM's network adapter comes up disconnected after a reboot or HA restart. Read Connect at power on, the port group binding and the VM log before changing it.Read article
Risks and limits
- Version scope is as each page states it. KB 318873 lists ESXi 8.0 Update 3 and VCF 5.2; KB 434979 lists any ESXi and vCenter version; KB 344313 lists vSphere 7.x and 8.x.
- Broadcom publishes no sample output for the two
pktcap-uwcaptures. The reading given above is guidance, not a vendor-printed result. - No ESXi command shows the physical switch's trunk list. That check has to be made on the switch (or through CDP or LLDP information in the vSphere Client).
- KB 426004 applies to vMotion and cross-vCenter migration on ESX 8.x. Guest-tagged (VGT) VLANs are a separate case and are not covered here.
NetLeverage runs its own hosting on VMware vSphere, vCenter and ESXi. Talk to an engineer.
Sources
- A virtual machine loses network connectivity after vMotion (KB 318873) — Broadcom, updated 11 November 2026, accessed 17 November 2026
- VMs lose network connectivity after vMotion/migration due to incorrect uplink teaming configuration (KB 434979) — Broadcom, updated 25 September 2026, accessed 17 November 2026
- vCenter Becomes Inaccessible After vMotion Due to Incorrect VLAN Configuration (KB 421993) — Broadcom, updated 8 January 2026, accessed 17 November 2026
- Network connectivity loss to a VM after vMotion or cross-vCenter migration (KB 426004) — Broadcom, updated 9 October 2026, accessed 17 November 2026
- IP to MAC mapping, GARP, RARP and Notify Switch setting for Virtual Machine Connectivity (KB 343401) — Broadcom, updated 13 July 2026, accessed 17 November 2026
- Testing VMkernel network connectivity with the vmkping command (KB 344313) — Broadcom, updated 1 October 2026, accessed 17 November 2026
This article was originally published on .