Instead of relying on the mess of downloading a tarball release, use the
nicely packaged webapp.

The emails where not delivered to anyway nor do they continue useful
information. Log output is now viewable in journalctl and some hardening
is applied.

This reverts commit bcfe1cbd.

Broken.

    fatal: [matrix.archlinux.org]: FAILED! => {"changed": false, "msg": "value of state must be one of: present, absent, latest, forcereinstall, got: 1.14.0"}

Archivetools syncs from kitchensink_tier1 for some reason, to create the
repository and stuff.

In preparation of the move to gemini make it already run the
archive.archlinux.org frontend and the archive-uploader. So the
migration is done in steps.

This seems more proper as default.target is not a real target that
services should require. multi-user.target is the sensible default
for such services.

Rename the variable so it's more clear it's for the retro website.

Btrfs by default has a pretty stupid algorithm that checks for whether a file is
compressable or not. In practice this algorithm is pretty slow in my tests and
just directly passing the file to zstd is faster. I'm running compress-force=zstd
on all my machines now and it seems safe. Let's roll with this.

This ensures that install_arch behaves the same as common.

This used to also needlessly and confusingly run on non-cloud boxes.

This seems way more correct to me as our DNS config is all over the place
and at least this way, we get correct fallback configs right off the bat.

Frederik Schwan authored Jun 12, 2020 and Sven-Hendrik Haase committed Jun 17, 2020

also use systemd instead of service module