Comment 2 for bug 2016868

I *think* this is something that is happening in 2.9 since we use the same peergrouper logic to be able to support Mongo 4.4.

We also should ultimately get a CI test that tests going into and out of HA, and checks that Mongo is happy, not just that Juju thinks it is happy.

(In this case, Juju kept running because the peergrouper still had 2 nodes in it that could both vote, and the third voter just never responded. Which allowed us to try to go into HA again, but then we couldn't get the right entities involved in the decision.)