Workshop breaks when a symlink can't be resolved

[mikeraymond7]

Issue

Workshop fails to get GPU information when a device symlink can't be resolved (i.e. a device was physically disconnected)

Relevant context

I went to workshop launch and received the following output:

 error: cannot perform the following tasks:
  - Setup "system" SDK profile (Failed retrieving GPU information: Failed adding device information for "/sys/class/drm/card0/device": Failed reading "/sys/class/drm/card0/device/de
  vice": open /sys/class/drm/card0/device/device: no such file or directory)
  "dev" launch aborted

workshop remove on a completely different workshop resulted in:

error: cannot perform the following tasks:
- Disconnect "dev/copilot:copilot-config" from "dev/system:mount" (Failed retrieving GPU information: Failed adding device information for "/sys/class/drm/card0/device": Failed reading "/sys/class/drm/card0/device/device": open /sys/class/drm/card0/device/device: no such file or directory)

The issue here is that path /sys/class/drm/card0/device/device exists, but is a broken symlink to a DisplayLink device:

lrwxrwxrwx 1 root root 0 Jul  7 10:00 /sys/class/drm/card0/device/device -> ../../pci0000:00/0000:00:0d.0/usb2/2-2/2-2.1

When I reconnected the DisplayLink device, the symlink was resolved and all of my workshop commands worked again. I cannot launch or remove any workshop when I am away from my desk unless I uninstall the evdi module, which seems undesireable.

What it should do

Seeing as the DisplayLink has nothing to do with workshop, this definitely doesn't seem like ideal behavior. I suspect that the GPU search functionality should pass on broken symlinks rather than aborting.

[jonathan-conder]

Thanks for the report. This is a bug in LXD rather than Workshop. If you like I can report it upstream, or you can do it. It should be possible to reproduce with this command:

# lxc info --resources

[mikeraymond7]

Thanks for the report. This is a bug in LXD rather than Workshop. If you like I can report it upstream, or you can do it. It should be possible to reproduce with this command:

lxc info --resources

I can add it there. Thanks. I suppose that closes this then?

[jonathan-conder]

We can probably add a workaround as well, leave it open for now

[jonathan-conder]

I'm also wondering why there's a broken symlink in the first place, might be a kernel bug

[mikeraymond7]

I'm also wondering why there's a broken symlink in the first place, might be a kernel bug

The DisplayLink driver was a pain to install and setup, so I wouldn't be surprised if the driver itself has some funky behavior. Either way, seems like LXD should warn instead of aborting.

[jonathan-conder]

It looks like DisplayLink/evdi#581 might fix it?

evdi: destroy dynamic devices when their USB parent is removed

Since it was already a pain to install, hopefully it's not too much of a pain to apply this patch. If you give me your distro and kernel version I might be able to build it for you.

[jonathan-conder]

Might be easier to downgrade the kernel instead

[mikeraymond7]

Looks like that PR probably will fix it. I'm not honestly too worried about it right now though, I'll just workshop from my desk 😂