Problems in the stable gate appeared to pointing to problems in ceilometer because of these logs but the exception was misleading because:

[11:17pm] cdent: in case you're curious, the oslo.messaging code is fundamentally broken:
[11:17pm] ddieterly joined the chat room.
[11:17pm] cdent: If there is an exception in __init__ of BrokerConnection called at https://github.com/openstack/oslo.messaging/blob/4cb33ecc94ce24497a76f2b3981832c6319f157b/oslo/messaging/_drivers/impl_rabbit.py#L538
[11:18pm] cdent: that is _not_ an IOError it will lead to the misleading exception we're seeing
[11:18pm] cdent: thankfully that code is completely different in modern oslo.messaging
[11:18pm] jogo: cdent: oh stable branches
[11:18pm] jogo: aren't they fun
[11:19pm] cdent: totz

The presumption is that some kind of exception was happening we don't know what, probably completely unrelated, was causing the trouble. An update of boto made things run through. Why? Who knows.