We had a prod incident on April 15, 2026 where one DB host of a 3-member replica set became unresponsive, there was an election, but for some reason the Mongo Ruby driver didn't kick that DB host from the cluster config; we had to manually restart all webservers to get the correct hosts (i.e. the 2 surviving members, without the unresponsive one.)

This should be handled better in the Ruby driver itself. I haven't tried to reproduce it yet, but this is not the first time I've seen this behavior in the driver, there is definitely some edge case of unresponsive DB hosts that it is not considering.