crash after edit vg in disk setup

Hello David,

Unfortunately, I can't get a hand on the crash report that you are mentioning.

I tried to reproduce your test case in a VM (with disk sizes much lower) but didn't manage to make it to crash.

If you can reproduce the crash again, could you please attach the corresponding log files? The content of /var/log/installer along with any relevant /var/crash/*.crash file would be good.

Thanks!

Found the logs on the USB stick.

Hello David,

I just managed to reproduce the crash but only when selecting the following option in the guided storage screen:

  [X] Encrypt the LVM group with LUKS

Could you please confirm if you had this option selected when the crash occurred for you?

Thanks!
Olivier

Hi Olivier,

Interesting; to answer your question yes and no :)

During the installation the installer crashed three times:
- First attempt, I made numerous changes to the disk layout (and on the screen before, I configured my network settings). I was creating multiple partitions, different mount points, different filesystems (should I use XFS, Btrfs or Ext4....), and trying to get to my 'desired' state. Then it suddenly crashed. One interesting note: it didn't write a crash report.
I don't know anymore whether I needed to restart the machine, or whether the machine rebooted itself, or whether the installer restarted. As I had 'played' around with so many options, I didn't really took notice.
- Second attempt. Again, I configured network, and then started with the disk layout. As I kindof knew what I wanted (but it wasn't clear how I could achieve it - I wanted a root-partition that was not encrypted, and some LVM partitions that were), I filled in all information 'in one' go, and I tried to edit the PV (to try to get to the state I wanted), and then it crashed again. This time with a crash report.
This time the installer restarted.
As I now knew there was a bug somewhere, I wanted to replay it in a minimal way.
- Third attempt. Everything default (most significant change: no static network config, but just DHCP), until the disk page. There the least amount of steps to crash. This time, no LVM encryption.

The crash report is from the third attempt, without LVM encryption. But the second attempt, I definitely selected LVM encryptionm a. When I checked the crash report of the second attempt, I saw this is in the steps taken (and somewhere in the report I see, password = redacted). As the third attempt was less steps (and, as there is binary data in the report, I cannot verify that the binary data does not contain my entered password), I uploaded this one.

So, apparently, the trigger/state change that is caused by selecting LVM encrpytion was persisted during the restart of the installer.

(In the end, 4th attempt: just /boot, /boot/efi, / partitions, left most of the space unpartitioned, and created partition + LUKS + LVM structure manually afterwards)

Regards,
David

Thank you for your insights, David. That is very helpful!

There are two processes involved in a Subiquity installation:
1. the server process which stores the data and actually performs the install
2. the client process which shows the UI and communicates with the server.

The crash that happens when editing the encrypted VG is a crash on the client side. When that happens, a new client process is started (and it starts over from the first screen) but the server does not automatically discard the data entered so far. I believe this inconsistency is what caused the presence of an encrypted LVM even though you chose not to on your third attempt.

I just submitted a patch to fix the crash when editing the encrypted VG: https://github.com/canonical/subiquity/pull/1532

As for the first crash that you encountered, sadly this is a bit of a mystery for me. If there is any log that you can share from this run (possibly in a different bug report), we will have a look.

Best regards,
Olivier

Hi Olivier,

Ah, nice that you found the cause. Regarding the first crash: I don't think I'll be able to reproduce it, and there is no log (basically, the second/third attempts were the reproductions).

Regards,
David