Spine/Leaf Policy Persistence On Reboot with APICs Unavailable

[tonybourke]

Hi All, I have a question about the failure scenario as follows: 

• APICs become unavailable (failure, network issue, whathaveyou)
• One or more leafs/spines reboots

I think it's well understood that the fabric will continue to forward in the event that the APICs aren't reachable, but what would happen if the leafs/spines reboot during that time? Would the policy be persistent on reboot? Or would they fail to even gain TEP addresses? 

[6askorobogatov]

Leaf will start up with existing config.  Even after the software the upgrade, leaf still keep existing config. 

Only, If you do clean reboot and APIC is not online, leaf will have no configuration. 

[tonybourke]

Thanks. Do you know of any documentation that would support that? Going through the APIC docs, I can't find any indication and I don't recall anything form any Cisco Live sessions I've watched. 

You don't have to point me to the exact doc if you don't have it, but if you remember seeing it anywhere and perhaps have an idea of what document you saw it in I'd be grateful. 

Cheers

[6askorobogatov · Accepted solution]

No, I have not seen one. I actually reload leaf with APIC disconnected in our lab. Also, we have TAC case when after the code update, leaf load config with some attributes that created faults with the new code. My impression, before that was, that after APIC upgrade,  when leaf got upgraded the new config will be pushed from the APIC. It is not the case, to force leaf to get new config from the APIC,  clean reboot is required.

[tonybourke]

Cool, many thanks!