Does anyone have any idea what the source of the intermittent Travis failures for Hackage is? The HighLevelTest program sometimes fails to connect to the server -- you can see an example at the bottom of https://travis-ci.org/haskell/hackage-server/builds/18423099 with the (almost) equivalent code passing at https://travis-ci.org/haskell/hackage-server/builds/18970590 -- but there's no indication that the server had problems starting (for instance, because the test port isn't available, or something like that).