APIC Failover IP Connectivity question

[AlexNastas74909]

Hello community.

I'm very new to Cisco ACI so sorry if my question will sound stupid, but maybe you can point me into the right direction. 

If I have a 3-APIC cluster, my understanding is that each of them will have a separate management address. 

Now, if I have a 3rd party app that interacts with my ACI fabric via REST API and uses one of those management IPs. 

What happens if that APIC fails ? Is there any way that management IP could get taken over by another APIC that will become active ? So that my 3rd party app would continue to function ? If that's not the right approach, then which one is ?

I have looked everywhere to try and find a clear document explaining how clustering works from the MGMT IP Connectivity standpoint, but haven't found anything. 

Thanks in advance

Alex

[tuanquangnguyen]

You could try sending queries to all APIC in the cluster.
In Python (or in any programming languages, really), multiple sequential try-catch blocks should do the trick. If the application is neither managed by Cisco nor you (hence, 3rd party), maybe the documentation would describe how to connect to the APIC cluster. Else, better raise that to the vendor who created the application.

[AlexNastas74909]

I'm the vendor of the application.
And I would am asking because I would like to know how to handle this case.

[Sergiu.Daniluk]

Hi Alex,

What you can do is use DNS failover, where you register all 3 APIC IP addresses for a single name/DNS record.

In your application you will be using the name instead of the IP. This way, regardless which APIC will be used, the application works.

Regards,

Sergiu

[AlexNastas74909]

Thanks for the replies. 

So my question really was more if my understanding is correct and Cisco does not handle this. 

This means that it needs to be handled on the application side. 

And also was wondering what is the good practice here, how the other developers implement this. 

Thanks

Alex

[Sergiu.Daniluk · Accepted solution]

Hi,

Yes, you are correct. There is no shared VIP for APIC management. You will have to handle the communication with APICs on the application side.  From this perspective, the best way to handle the communication, is to always send the requests to only one APIC. How you achieve that (DNS, natively in the application, load-balancing etc) or to which APIC your app will send the requests, it is all in your hands. In the end, is not important with which APIC your app communicates, since the cluster is constantly synchronizing the configuration database (through the use of sharding).

One thing which you might want to know is if 2 out of 3 APICs are experiencing an issue, the 3rd APIC will go into minority state which is a read-only state, and you will not be able to make any changes to it, neither GUI nor CLI.

Hope that helps.

Regards,

Sergiu

[AlexNastas74909]

Thanks Sergiu,
This is the answer I was looking for !

Regards
Alex