We are observing an issue where a configured aws ec2 cloud instance configuration randomly switches its type from Windows to Unix. The rest of configuration is unaffected, but this leads to agent being started afterwards to get stuck looking for ssh which, obviously, will never happen.
There is nothing in the logs that suggests what might be the cause, but this seems to correlate with ether jenkins version upgrade or global configuration change events. Popping into global configuration, changing instance type back to windows and re-adding password (and deleting the agent(s) stuck at the time when this is noticed) gets everything back to normal into the horrors of winRM land.
This has been ongoing with the past 3-4 LTS versions and with at least last 2 ec2 plugin versions.
Magic like somebody rolling back local configurations on disk or similar have been excluded.
(I know this is not particularly useful issue report, but raising for the record and just in case if somebody else finds the title with the same issue)