Mike,
We have a topology similar to yours and had a similar issue with our 7k's after upgrading from IOS version 5.1 to 5.3. DHCP ran fine for several months and then out of nowhere we had sporadic DHCP issues across most of our Vlans. This caused a major disruption for over 2000 clients.
We could see all of the DHCP discovers arriving on the DHCP server whilst being relayed from the Nexus 7k's. We could see all of the offers being sent by the DHCP server, received on the Nexus 5k's and then forwarded to the 7k's. On the 7K's we could see these packets were not all being forwarded successfully.
We couldn't figure out why this was happening but after much digging and support from TAC we found that DHCP did not have its own class in COPP on the 7k's. We ended up updating the COPP configuration (ran copp strict) again to bring the new profiles of COPP to the 5.3 version. This did not get updated as ISSU does not upgrade the configuration with the new COPP automatically. This caused our packets to be not classified under a specific class. This meant they went under the radar from a forwarding behavior (in class default) which was our scavenger class. Updating the COPP configuration did the trick for us.
Now..... The trigger for our woes ended up being a host on our network that was sending a high volume of DHCP broadcast.
Due to these broadcast the COPP which is there to protect the CPU is active and dropping packets, only allowing a certain packet rate through. As COPP cannot differentiate between "good and bad" packets it was dropping between both sent to the CPU. Hence this cause some host to work but the majority not and sporadic issues across multiple Vlans with DHCP because of the excessive traffic being sent out by the source end host.
I hope this helps if not at least you got a good read and understanding of the misery we went through with this issue.
Good luck,
William
