Grat.ARP on Nexus 7000

[HUBERT RESCH]

Hi we observed that if one of our Cluster is switching over( secondary is taking over the cluster IP adress) and sending Grat.ARP to announce the new MAC adress for the Cluster IP address, the Nexus does not update the ARP table based on a Grat.ARP.

I am not absolutely sure it seems if the Grat.ARP hits the HSRP active Switch than it works, if it hits HSRP standby switch it does not work.

Under vPC we have arp synchronize konfigured.

The only way how we can bring it to workmis to clear the arp table.

Any idea?

Thx

Hubert

[rsimoni]

Hi Hubert,

I think I saw something similar.

Which NX OS version is your N7K running?

About the cluster, is it a VM on ESXi by any chance? Which OS?

How do you exactly switch your cluster over?

Are you sure the cluster really sends GARP? I am asking because both Windows and VMware made some changes in that area lately and as a matter of fact GARPs are not always sent out.

Riccardo

[HUBERT RESCH]

Hi Riccardo, we are running NXOS 5.2, and we already veridied that the clustermembers are sending GARP when overtaking the active IP.

KR

Hubert

[rsimoni]

Hi Hubert,

ok I see; but what about the cluster? Is it somehow related to vmware? which OS are we talking about?

I insist on this point as usually server administrators are not too much into networking and they might give misleading info. I.E. the vmware heartbeat that control the IP floating mechanism (which happens in case of cluster failovers) relies on ARP probes and not on GARP. The 2 are pretty similar but a N7K ignores it if it is not destined to the local IP (which it is true for the active hsrp member only). It flags it as invalid packet and drops it... you can verify taking the following before and after the switchover from both n7k

show ip arp statistics vlan 

often you see increasing value if you have many hosts in the vlan... so the outputs of this command is not definitive in all cases.

Riccardo

[HUBERT RESCH]

Riccardo its a Netscaler Loadbalancer Cluster

[rsimoni]

can you capture such GARP sent upon switchover to confirm what is being sent by the cluster?

beside the command suggested above try also to gather the following from both nexus

sh ip arp internal event-history event | i

we need to look for events at the time of the switchover

R

[rsimoni]

also show log can help us.. do you see any unusual message?

[terry.willis1]

Hi Hubert,

Was there any resolution on this issue?  We are experiencing a similar issue with a different load balancer vendor.

Thanks,