cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
Announcements

307
Views
0
Helpful
2
Replies
nileshgore25
Beginner

Endpoint database after fabric upgrade

Hi experts,
We will be doing ACI upgrade from 3.2(5e) to 4.2(7f)

our ACI fabric have around 12k endpoints learned in EPGs associated with L2 BDs (unicast routing disabled, ARP flooding enabled, gateway not on ACI) as well as other EPGs associated with L3 BDs (unicast routing enabled, ARP flooding disabled, limitIpLearnToSubnets enabled, gateway on ACI)
few of the BDs are IPv6 only while others are IPv4 and IPv6.

I have following queries regarding the endpoint database
1) Will the APIC restore the endpoint information that it had before upgrade ?
2) what care should be taken after upgrade or what process should be followed, so that the traffic is restored soon after upgrade.
I understand that the ACI endpoint will be build based on data-plane traffic. But anything that should be done for issues related to endpoint database perspective ?

2 REPLIES 2
Robert Burns
Cisco Employee

The EP DB is not maintained by switches across reboots.  They would need to re-learn the endpoints either from the dataplane or from COOP (control plan).  Hopefully both your external trunk connection (to your GW device) is dual homed to 2 different leaf Switches, as well as the majority of your endpoints dual attached to leafs.  When you proceed with your upgrade you should divide the leafs into two maintenance groups (typically odds/evens), such that when you run the upgrade, you should always maintain connectivity to devices endpoints and the GW.  The only endpoints that might experience a disruption are those single-homed. 

A good practice for pre/post upgrades is to leverage an App called "State Change Checker".  This will allow you to compare a before & after view of the fabric and quickly identify anything "missing" following the process (routes, adjacencies, EP entries etc).  You can find more info on this App here: https://dcappcenter.cisco.com/statechangechecker.html#:~:text=State%20Checker%20(also%20known%20as,What%20changed%20in%20my%20fabric%20%3F 

Robert

Thanks for confirming about the endpoint DB not maintained across reboot.

We have planned upgraded of switches in multiple odd even groups spread across one week per fabric.

 

I did considered state checker app, but I found it very slow (unresponsive after waiting for long time).

I will be using python scripts to check link state, bgp state, faults state before and after the upgrade

Content for Community-Ad