I don’t know why your builds are failing, however one tip: I’d recommend adding a timeout-minutes value. Else if something does go wrong (like this), Github will indeed wait … and wait … and wait … which is a waste of time. Especially if you are paying per minute. If a deploy to two instances takes longer than X minutes (e.g 15 minutes) I think it’s pretty safe to assume it’s gone wrong and won’t work at all.