Comment 2 for bug 1414098

It's probably true that in some cases the seed just isn't contactable - but because of the way the command is all bundled together and stdout/stderr are thrown away, we can't distinguish that case from other cases.

https://review.openstack.org/#/c/149819/ is a partial fix for this. Just looking at the test failures on that particular review, we've already discovered one other source of error: we provide a list of --exclude filenames to tar; tar tries to stat those files, and errors if any of them don't exist. So now we know that we need to be more selective about what we --exclude, or possibly that we need to use wildcards instead of literal filenames to make tar less fussy about the files.