Since systemd-docker reparents the Docker process to the systemd
cgroups, it makes sense to implement resource controls there.

Fixes issue #39.

Prevents audit log duplication.

And add EQUALITY rules, or we won't be able to delete attributes.

This reverts commit 2e4e6375.

This reverts commit 0a9b16eb.

Start with DiskWillFillIn4Hours, which isn't really relevant in a
small testing environment.

The solution is still a bit hackish, and it involves a
cron job that runs every 15 minutes and checks for new
additions to the replds@acme repository.

Run Elasticsearch in a container

See merge request !11

Keep getting Docker debian gpg key from a URL, due to the bug
described in https://github.com/ansible/ansible/pull/48888,
until the fixes hit the stable Ansible branches.

When a systemd unit was defined by multiple services (as it may happen
for system-wide services that are split by inventory group, for
instance), we would disable it in the base/cleanup stage as it was
wrongly added to float_disabled_systemd_units.

Since there isn't necessarily a 1:1 mapping between float services and
systemd units, we have to be smart when computing which services to
disable.

Drop the 'nodelay' clause from the 'perip' rate limiter.

It seems that the Ansible apt_key module was downloading the keys
every time, even if the key ID was already present on the
system. Apparently this prevents it from doing so.