Survey the machine, and say what an install would do to it

The first two pieces of the installer, and deliberately the two that change
nothing: what this machine is, and what would happen to it.

The survey is the floor under every later choice -- a component can only
offer a container shape if Docker answers for this user, or a host shape on
a systemd machine with what that shape needs. It reports what it could not
establish as unknown rather than guessing: an unprivileged probe of port 25
means "I may not bind this", not "something is listening", and reporting
the first as the second is a lie an operator would act on.

The plan is one list, and it will have three readers: --dry-run prints it,
the interface will show it before anything happens, and apply will walk it.
What you are shown is what runs.

Tested on a fresh Debian 13 in qemu, which has neither Docker nor Node and
so exercises every unavailable shape: 23 checks, including that nothing was
created by any of it. The lab that machine runs in is in e2e/vm.
This commit is contained in:
2026-09-22 17:56:20 -07:00
commit f97fd12556
13 changed files with 2046 additions and 0 deletions
Executable
+127
View File
@@ -0,0 +1,127 @@
#!/bin/bash
# SPDX-FileCopyrightText: 2026 Coffey Labs
# SPDX-License-Identifier: AGPL-3.0-or-later
#
# Build the throwaway machine the installer is tested on.
#
# The installer writes units, creates users, takes 25 and 443 and can install
# a web server. None of that belongs on a workstation, and none of it can be
# proved in a container either -- systemd, users and ports are the thing under
# test. So: a Debian cloud image in qemu, seeded with cloud-init, with a copy
# of the disk kept the moment it is up. Every run starts from that copy, so a
# run can break the machine as thoroughly as it likes.
#
# e2e/vm/up.sh build it and keep a clean copy (idempotent)
# e2e/vm/reset.sh back to the clean copy, a few seconds
# e2e/vm/run.sh build the installer, copy it in, run a case inside
# e2e/vm/down.sh stop it and remove its disks
#
# Plain qemu, run as you, rather than libvirt: no root, no storage pool, no
# permissions to argue with, and the logs are readable. Networking is qemu's
# user-mode NAT with ssh forwarded to 127.0.0.1:2222, which gives the guest
# the internet it needs to pull images without putting it on the LAN.
#
# Needs: qemu-system-x86_64, /dev/kvm, xorriso, ssh, curl.
set -euo pipefail
LAB="${LAB:-$HOME/.cache/inbuxa-lab}"
MEM="${MEM:-4096}"
VCPUS="${VCPUS:-2}"
DISK_GB="${DISK_GB:-20}"
SSH_PORT="${SSH_PORT:-2222}"
BASE_URL="https://cloud.debian.org/images/cloud/trixie/latest/debian-13-genericcloud-amd64.qcow2"
BASE="$LAB/debian-13-base.qcow2"
DISK="$LAB/lab.qcow2"
CLEAN="$LAB/lab-clean.qcow2"
SEED="$LAB/seed.iso"
LOG="$LAB/console.log"
PIDFILE="$LAB/qemu.pid"
KEY="${KEY:-$HOME/.ssh/id_ed25519.pub}"
say() { echo "==> $*"; }
. "$(dirname "$0")/lib.sh"
[ -r "$KEY" ] || { echo "no public key at $KEY (set KEY=)" >&2; exit 1; }
[ -w /dev/kvm ] || { echo "no writable /dev/kvm -- is this user in the kvm group?" >&2; exit 1; }
if vm_running; then
say "already up on 127.0.0.1:$SSH_PORT -- reset.sh rewinds it, down.sh removes it"
exit 0
fi
mkdir -p "$LAB"
if [ ! -f "$BASE" ]; then
say "fetching the base image (once)"
curl -fL --progress-bar -o "$BASE.part" "$BASE_URL"
mv "$BASE.part" "$BASE"
fi
# cloud-init: one unprivileged user with our key and passwordless sudo, and
# almost nothing else. The installer is what is under test, so the machine
# should be as plain as a new VPS -- if the installer needs a package, the
# installer should say so rather than the lab quietly providing it.
WORK="$(mktemp -d)"
trap 'rm -rf "$WORK"' EXIT
cat > "$WORK/meta-data" <<EOF
instance-id: inbuxa-lab
local-hostname: inbuxa-lab
EOF
cat > "$WORK/user-data" <<EOF
#cloud-config
users:
- name: lab
sudo: ALL=(ALL) NOPASSWD:ALL
shell: /bin/bash
ssh_authorized_keys:
- $(cat "$KEY")
package_update: true
packages:
- curl
- ca-certificates
- openssl
- python3
growpart:
mode: auto
devices: ['/']
resize_rootfs: true
write_files:
- path: /etc/inbuxa-lab
content: |
This machine exists to be broken by e2e/vm/run.sh. Nothing on it is kept.
EOF
say "building the seed"
xorriso -as mkisofs -quiet -o "$SEED" -V CIDATA -J -r "$WORK/user-data" "$WORK/meta-data"
# An overlay on the base, so the download is written once and every rebuild is
# instant. The base itself is never modified.
say "building the disk"
rm -f "$DISK"
qemu-img create -q -f qcow2 -F qcow2 -b "$BASE" "$DISK" "${DISK_GB}G"
say "starting the machine"
vm_start
say "waiting for ssh on 127.0.0.1:$SSH_PORT"
for _ in $(seq 1 90); do
if vm_ssh true 2>/dev/null; then break; fi
sleep 2
done
vm_ssh true 2>/dev/null || { echo "no ssh after three minutes; see $LOG" >&2; exit 1; }
say "waiting for cloud-init to finish"
vm_ssh 'sudo cloud-init status --wait >/dev/null 2>&1 || true'
say "keeping a clean copy of the disk"
vm_stop
cp --reflink=auto "$DISK" "$CLEAN"
vm_start
for _ in $(seq 1 90); do
if vm_ssh true 2>/dev/null; then break; fi
sleep 2
done
say "up: ssh [email protected] -p $SSH_PORT (console log: $LOG)"
vm_ssh 'echo " guest:" $(. /etc/os-release && echo $PRETTY_NAME) "| kernel" $(uname -r) "| disk" $(df -h / | awk "NR==2{print \$2}")'