Just reporting this here because I can't make a support case for it (production equipment, and just wanted to get it up and running ASAP).
I have a BGP signaled MPLS network across a few sites with CCR routers. This morning I had a port flapping on one of the CCRs, so I decided to upgrade the firmware on the affected router from 6.20 through to 6.28 to rule that out as a cause.
When the router rebooted, my OSPF sessions were dropping every 10-30 seconds and resetting. After trying a large amount of things I figured out what it was.
After simplifying the network down to two CCRs, both with 6.28, I have a static IP on an interface facing each other. I then have OSPF running to advertise their loopback addresses. That works fine, and they can receive and ping each others loopback.
Then I enabled BGP onto of the OSPF sessions by using the loopback address of each for the peer address. Each peer was added with l2vpn as the address type in BGP, not ipv4.
This works fine and the connection establishes.
Then you go into the VPLS menu and create a BGP VPLS. As soon as you enable this setting you can see in the debug logs that the VPLS gets installed, then instantly the OSPF session will drop, meaning the loopback routes are removed, the BGP session will disconnect, and the VPLS will be removed.
The OSPF will then reconnect a few seconds later, and as soon as the BGP timer expires and restarts the BGP session, the BGP VPLS will connect, and the OSPF will die again.... Repeat x100000.
It seems the combo of BGP signaled VPLS on a OSPF base layer is stuffed in .28.
Exact same config and rolled back to 6.20 and it works perfectly (however my original port flapping problem may reappear now).