is multipass reliable? (qemu image corruption issue/rant)

[andresmmujica]

I've been using multipass without major issue for a few months, I deployed an Ubuntu VM for my work and been setting it up for a few weeks. Today, for no apparent reason, the image is corrupted. How come?

I was using the vm when it froze for a few seconds, and after that whatever that has happened completely borked the VM. I managed to fix the errors with qemu-img check -r leaks/all but still the VM although it boots, it gives input/output error with everything. I have the suspicion it was related to a brew upgrade I was running on my mac, but it didn't really upgrade anything related as far as I could tell.

The thing is that now I have to fully setup my VM with everything I had on it, and tomorrow I'll need it first thing in the morning... damn... I'm so pissed... After browsing the issues there are hundreds of similar situations...

I'm afraid multipass is not reliable as I have thought.

I'm re-deploying my vm in lima.

[ricab]

Hi @andresmmujica, I am sorry that you ran into this situation. Given your observation of the VM hanging for a few seconds during a brew upgrade, all I can think of is some shared library getting updated while in use by the VM process. Hopefully only things shipped with Multipass would be used, but I guess there is a chance that something went in front.

VM corruption is typically caused by a sudden disconnect between the QEMU process (the one running the VM) and the image file. For example, when the process gets killed abruptly. There are multiple possible reasons for that, like an OOM killer, the filesystem being yanked out, macOS deciding to reboot without giving Multipass enough time to finish, running out of disk space, or a crash in QEMU.

On top of that, we suspect our current procedure for VM suspension could be causing some corruptions. That is something that we hope to devote some time to very soon.

One suggestion to prevent loosing your setup to something like this would be taking snapshots of your VM (multipass snapshot) or even keeping a sane clone on the side (multipass clone) just in case.