VM cannot be contacted following vMotion

[cawn]

Hello to you all.  When I migreter a vmware guest from one host to another host, there are sometimes that I can not contact the Guest afterwards. I am told that this may have something with arp to do, on the physical switches. With the ARP table is not updated. How can this bee? Thanks.

[iw123]

Hi

After you have moved the VM can you ping out from the guest?

Is your network configuration consitent across all hosts? Do you have notify switches enabled in your vswitch settings?

thanks

[cawn]

Hello and thanks for the replay. The notify switches is set to yes on alle my hosts. And the network configuration is consitent across all hosts. I can ping the guest from another guest on the same network, but not from a another guest on another network.

[iw123]

Hi,

Can your Guest ping it's default gateway?

[rickardnobel]

Carsten Widtfeldt Nielsen wrote:

I can not contact the Guest afterwards. I am told that this may have something with arp to do, on the physical switches. With the ARP table is not updated.

That is likely a confusion. ARP is the mapping between layer 3 IP addresses and layer 2 MAC addresses and is typically only held at the end devices, that is for example a PC or Server. Since vMotion does not change the MAC address there is nothing that can go wrong with ARP.

However, it might be that your physical switch Port-to-MAC mapping is not being updated correctely. Do you have access to your physical switches?

Can you run commands like:

show mac

Depending on your switch type the command line syntax might be different but it is usually similar to this.

[cawn]

Hello - I did not try to ping the gateway. But I coud ping the guest from another guest on the same network. When i disabled the network, and then enabled the network again - the server got online again.

[cawn]

Hello, and thansk for the replay.  The physical switches are HP procurve, and i do have access to them, and can run commands on them.

[rickardnobel]

Carsten Widtfeldt Nielsen wrote:

The physical switches are HP procurve, and i do have access to them, and can run commands on them.

Which Procurve model?

Could you first check the MAC address of a VM, then do a :

show mac-address-table (if I remember the syntax correct)

Verify that the MAC address is mapped to the port going to the current ESXi. Run a ping -t against the VM from some external PC. Do a vMotion.

Check again with the show mac command and verify that the MAC address has been remapped to a port going to the new ESXi host.

If you have a lot of MAC addresses displayed you could (on some Procurves) run:

show mac | include PART-OF-MAC

[cawn]

Hello - I did som testing whit a guest. And after 8 maigrations the guest failed - I could not pint the guest.

We coud see that the HP switch did not update the MAC adress tabel - the HP switch stil thinks that the guest are in the same port as before the guest got maigraited.

From the guest we could not ping the gatweay og the guest.

2 HP Procurve switches i trunk.

[iw123]

Is this behavior affecting all your virtual machines? Is there anything in common, e.g. are the affected machines on the same vlan/portgroup, or does this happen regardless of what vlan your guest is on?

[rickardnobel]

Carsten Widtfeldt Nielsen wrote:

2 HP Procurve switches i trunk.

You are using HP Trunk? Could you describe some more how your vSwitch configuration is done:

How many vmnics per vSwitch?

Which NIC Teaming Policy do you have? (This is very important together with HP Trunk mode.)

[cawn]

Hello - the behaivor is only affecting the guest that are being migrated, and this is not every time - this i random. The guest are not on he same vlan ore portgroup.

[cawn]

We are not using trunks from the host to Ho switch. But between the to HP switches there is a trunk. On my vSwitch2 - where the HP VLAN are loacted - there are 4 nic's. Tow nic pr. HP switch - 2 HP switch is i an trunk - trunk between the 2 HP switches.

Loade Balancing: Route based on the originating virtual port ID

Network failover Detction: link status only

Notiry Switch: yes

Failback: yes

[iw123]

Do you have portfast set on your physical switches?

[rickardnobel]

Ian wrote:

Do you have portfast set on your physical switches?

That specific command is not available on HP switches, but the ports should default to "edge ports" and I do not think Spanning Tree is any issue here.