What would happen if the endpoint host move on the same ACI leaf

[Brijendra Pratap Singh]

What would happen if the endpoint host move on the same ACI leaf switch from one port to another since the COOP database at spine did not track the information about the port of the leaf switch from which the EP is learned?

[RedNectar]

Hi @Brijendra Pratap Singh ,

The answer depends on the type of endpoint.

If the endpoint is a physical server, then the move would involve loss of Layer 1 connectivity when the cable was moved.

• The leaf switch would immediately time out the entry (MAC and IP) for the endpoint
• The leaf switch would update the Spine Proxy

When the cable was plugged into the new port

• The leaf switch would do nothing until a frame was received from the endpoint, at which time it would
  • Update it's MAC address table
  • and update the Spine Proxy of the new MAC
• When the first ARP request was seen from the endpoint (this is likely to be the same first frame), the leaf switch would
  • Update it's IP endpoint table
  • and update the Spine Proxy of the new IP
• [The above step would be the same if an IP packet addressed to the default gw MAC arrived before an ARP - VERY UNLIKELY if the host has lost connectivity during the move]

IF the endpoint is a VM, or attached via a L2 switch then the leaf will not know that the endpoint has moved UNTIL it sees a frame from the endpoint on the new port.  In the case of a VM move, this frame is most likely to be a RARP or GARP broadcast frame created by the vswitch that the VM moved to.  This frame will be sent with the source MAC address of the VM

In any case, when the leaf switch sees the first frame on the new port, it will update is MAC address table - probably also report this as a new AMC to the Proxy (not sure about that, but it doesn't matter because it is the same leaf, and that's all the Proxy cares about).  And ditto for the fist ARP or IP packet addressed to the gw MAC - the leaf would update its local tables and no-one cares if the Spine Proxy table is updated or not because the endpoint is on the same switch.

Note: I haven't had a chance to lab this yet - I may have missed some subtleties in my description.

[Brijendra Pratap Singh]

Thanks a lot RedNectar for your briefing. The issue here is the endpoint is in VM. After the Vmotion the EP is being learned on the same switch but on the different port and the LST timer for the EP is 900 seconds. What we have observed is after this move the EP is not able to communicate to rest of the network for 15 mins (900 Seconds) which indicates that it is waiting for the timer to be elapsed & since COOP database doesn't track the interface details of the leaf like through which interface at leaf that particular EP was learned so it really doesn't care if the same EP is being learned on the same leaf over different interface. I am searching for an option to resolve this issue & what I can think of is moving the physical interface (where the EP appears after Vmotion) to a different leaf so that once the EP is moved it will get learned on the local leaf who will further update the coop and this will trigger the Old leaf to make the bounce entry for that particular EP. 

Is there any other way to resolve the problem (other than clearing the EP entry from the leaf switch side)?

Is my approach looks fine?

Regards,

Brijendra

[RedNectar · Accepted solution]

Hi @Brijendra Pratap Singh ,

The issue describe shouldn't happen.  And if it is happening, I suspect it is a problem on the leaf switch rather than anything to do with the COOP database.

Anyway, I tried to reproduce your problem, but couldn't.  You can see (and hear) my ramblings as I tried to reproduce your problem below

One thing you might want to check is to see if Endpoint LeLearning has been turned off. If I get time I'll explore that later, but right now, life calls...

[Brijendra Pratap Singh]

Thanks @RedNectar for your detailed video and response. It helped a lot. Let me revisit the issue and if possible I will try to share the details as well otherwise will keep you posted on my further analysis.