Richard's debugging suggested it was BIRD not responding to valid BFD packets transmitted from our Juniper router.
Re: "valid". In looking at the packet capture again, I'm not sure how the discriminator values are supposed to look. So maybe that is related.
The problem for diagnosis is reproduceability. Since a session reset fixed the problem, we can't reproduce the problem, and that makes diagnosis impossible without additional data.
To be clear, I don't expect any further action from MICE, as the problem is resolved.
I sent a write-up to the bird-users mailing list. We'll see if
anyone there has any ideas.
-- Richard