Use new QEMU environment for regression tests

[framps]

Current regression setup should be upgraded to use this setup.

[UnconnectedBedna]

I would suggest systemd-nspawn instead. Only one apt package needed (and systemd ofc): sudo apt install systemd-container

Then this gives you "chroot access" (assuming you run as root):

if systemctl is-active -q systemctl-resolved.service; then
  systemd-nspawn --resolv-conf=replace-stub -D /mnt bin/bash
else
  systemd-nspawn --resolv-conf=copy-host -D /mnt bin/bash
fi

This is if you need internet access, otherwise sudo systemd-nspawn -D /mnt bin/bash is enough.

https://wiki.debian.org/RaspberryPi/qemu-user-static (Option 2.2)

Or maybe I am misunderstanding what you are implementing, my German is a bit rusty. xD
But with this, you can do whatever you want in a "chroot environment", for example use apt, rebuild initramfs or whatever...

Also, if you want to mount an image, you can just loop it with the -P (probe) option, no need for calculation of offsets.

[framps]

@UnconnectedBedna Thank you very much for your tips on this issue 👍

I'm not very familiar with chroot and systemd-nspawn. I understand I can create a backup in a spawned environment. But how can I test the special Raspberry boot process works on a restored system? Let's say the UUIDs in /etc/fstab and cmdline.txt don't match the UUIDs of the restored device. Will this be detected and the boot fail if a restored image will be used? Or will missing files in /boot cause the boot of the restored image to fail? As far as I understand there will be no Raspberry boot process executed and those errors will not be detected. Please correct me if I'm wrong.

[UnconnectedBedna]

You are correct, this is not "booting" the system, you are just changing "root" to another filesystem. You still use the same kernel you booted from. I don't think it will detect those missconfigurations. Emphasis on think, it actually might, I don't know...
Edit
The more I think about it, it might actually complain, maybe not about cmdline.txt but for fstab. But I would still just do it manually in bash instead.
end edit
This is to make changes you can't do with a simple editing of a config file, like installing with apt, or rebuilding initramsf for example.
But maybe there is some kind of application that runs a test for configurations, I don't know...

I would just do it in bash on a mounted loop device, no need to emulate/chroot/nspawn anything.
For example:

# loop img file and mount somewhere, I will assume /mnt here and that it is looped at loop0 (loop0p1 for boot and loop0p2 for /)

BACKUP_BOOT_PARTUUID=$(lsblk -no partuuid /dev/loop0p1)
BACKUP_ROOT_PARTUUID=$(lsblk -no partuuid /dev/loop0p2)

if echo "$BACKUP_BOOT_PARTUUID" | grep -q -f /mnt/etc/cmdline.txt; then
  echo 'config ok'
else
  <do whatever needed to be done to fix the config>
fi

Or: grep -q $BACKUP_BOOT_PARTUUID /mnt/etc/cmdline.txt, I just find the above line easier to read

And then something similar for fstab, like:

if echo "$BACKUP_ROOT_PARTUUID / " | grep -q -f /mnt/etc/fstab; then
  yadayadayada etc

and

if echo "$BACKUP_BOOT_PARTUUID /boot" | grep -q -f /mnt/etc/fstab; then
  yadayadayada etc

[framps]

I agree it's possible to check for any UUID mismatches - even for PARTUUID or LABEL mismatches. But what about any missing files in /boot? I frankly that's why I decided the restore should be tested as close as possible with a normal boot to make sure the restored image boots the backup successfully. As of now I don't know of any way other than to use QEMU.

[UnconnectedBedna]

How would that even be possible?
I'm all for redundancy, but that seems extremely unlikely.
A rsync would confirm it in a second or two.
Or just use diff -qr directory1 directory2.
https://www.tecmint.com/compare-find-difference-between-two-directories-in-linux/

My point is if there are tools present, why create a bunch of dependencies for stuff not needed.

[framps]

My point is if there are tools present, why create a bunch of dependencies for stuff not needed.

I see your point. But I frankly trust a regression test which executes the Raspberry boot sequence more than to check any known possible boot problems.

I agree it's much more difficult to setup a QEMU environment which executes the boot sequence than to check for known boot problems.

[UnconnectedBedna]

Again, my point is not the difficulty, it's the application requiring to install a bunch of dependencies when there are already tools that can do the same thing.

A diff for example checks if there is even one single data bit of difference in the files. It can even tell you what differences there are.
You don't trust git? the tool the linux kernel uses to make sure things are correct? xD

Besides, an emulated boot is very far different from a live boot anyway.

[framps]

Hm ... you suggest to compare the backup with the restored image bit by bit?

[UnconnectedBedna]

Only the /boot directory (and subdirectories), that is what diff -qr will do. :)
If you only want /boot and not subdirs, remove the -r option.
https://stackoverflow.com/questions/4997693/given-two-directory-trees-how-can-i-find-out-which-files-differ-by-content

[framps]

OK. You convinced me a pure compare of the backup and the restored image should be enough to check the restored image will boot. But what if there is anything missing in the backup because the backup misses some files?

[UnconnectedBedna]

diff will tell you ALL differences, ie missing files too.

Example:
Text File1.txt are the same in both directories
Text File2.txt is different
Text File3.txt only exists in folder1

$ tree
.
├── folder1
│   ├── Text File1.txt
│   ├── Text File2.txt
│   └── Text File3.txt
└── folder2
    ├── Text File1.txt
    └── Text File2.txt

3 directories, 5 files

$ cat folder1/Text\ File2.txt
content 2 and some more

$ cat folder2/Text\ File2.txt
content 2

$ diff -q folder1 folder2
Files 'folder1/Text File2.txt' and 'folder2/Text File2.txt' differ
Only in folder1: 'Text File3.txt'

If you want to see the actual differences with the diff command, simply remove -q

$ diff folder1 folder2
diff "folder1/Text File2.txt" "folder2/Text File2.txt"
1c1
< content 2 and some more
---
> content 2
Only in folder1: 'Text File3.txt'

And lastly if there are no differences it will return zero value, ie you can do:

if [[ -z $(diff folder1 folder2) ]]; then
  echo "no diff"
else
  diff folder1 folder2
fi