I'd wanted to debug this but it looks like I'm not going to have time. The test makes a snapsot in 5.6 and then does a full cluster restart to upgrade to master. Then it does GET /_snapshot/repo/_all and looks for the snapshot and doesn't find it.