This only affect the qemu and virtualbox image. I'm not sure how you
enable trim support in VMware and the documentation is spare.

Tested with: packer build -only=qemu local.json
698M	Arch-Linux-x86_64-libvirt-2020-08-01.box <- before
425M	Arch-Linux-x86_64-libvirt-2020-08-01.box <- after

Fix #106

In my tests, this slows down the builds from 4m to 10m per build BUT allows use build
on public non-kvm-enabled builders which gives us the ability to run random MRs on CI.
Additionally, we can now run many builds in parallel.

nobody should use them, because it's insecure.

Keeping artifacts brings us no benefits and the produced artifacts
should never be used, because they consist of vagrant specific
settings.

For building QEMU/Virtualbox images, checkout the cloud.json file

Increase expire_in to 2d so that we can be sure that there are always fresh VM images available as artifacts

They are built nightly so we'll have a backlog of two VM images available to download at all times.

Going forward, it makes sense to host these raw images as well as there are some use cases for that.
For instance, what if you don't want to use vagrant but instead just use the raw images in qemu directly?

This makes sure there are no race conditions on our workers or during deployment.
We can only use one hypervisor at a time.

We decided to use more specific tags here depending on the type of runner for more runner separation.

Justus Rossmeier authored May 29, 2020 and Christian Rebischke committed May 29, 2020

Vagrant has experimental support for adding SATA disks. Although this
box does not support that yet (at least for the virtualbox provider,
since no AHCI storage controller with the name "SATA Controller" is
included), it is possible to add another disk. This causes a boot
failure in some circumstances because although the root is set to a UUID
in the kernel cmdline, systemd's fsck still looks for /dev/sda as stated
in /etc/fstab.