The issue here is really that /tmp is getting cleaned after the above runs.
Then, the test harness is going looking for the file, but its long gone.
I'm not sure if its specifically a bug or not to rely on runcmd running after
/tmp cleaning has run. but it does seem like /tmp is being cleaned
'mounted' events are handled synchronously in mountall, and the
'filesystem' isn't emitted until they're all done. Prelim tests seem to
confirm that this is working correctly (and if it wasn't that would be a
very bad bug, that I think we would have caught before now).
In a test instance from