ValueError: Incorrect path or url, URLs must start with `http://` or `https://`,
Hi, got this errors in the machine-learning container when running the 'tag object' job. Any idea what a reason can be?
ValueError: Incorrect path or url, URLs must start with
http:// or
https://, and upload/thumbs/5c86787b-cae9-4382-b7ca-cc0075d77b96/7cfcd751-2dda-4ead-a18a-93ec16ee388d.jpeg is not a valid path
12 Replies
@jrasm91 is this a bug caused by the new path handling?
we saw this once but haven't been able to reproduce it
@theautomation Can you make sure the mount can be accessed from the machine learning container?
Hm I get a permission denied when cd into the upload folder inside the machine-learning container 🤔
Looks like a user/file permissions issue
Yup, gonna dive into this, thanks so far!
try to bring down and bring the stack up
Fixed it by adding the GID an UID env of the mountpoint owner
hopefully it is not breaking anything else when using a non-root user
Hm now when I search for a random word in the searchbar I get an only when I am using a non-root user (GID UID) for the machine-learning container
Looks like my UID doesn't have the right permissions to /.cache
Is this a known issue ?

I think we encountered this
If you're running under a non root guid you'll need to create a volume for that folder with read write permissions for the user your are running the container with.
Thanks! I have made a EmptyDir volume (kubernetes ephemeral storage I am running immich in Kubernetes) and mount that into the container /.cache and that indeed work. So there is a and a folder? Must both be persistent storage or can it be ephemeral/temporary?
Not 100% sure which one is used where but both are non-essential and would just mean redownloading ml models again probably