New ML volume
Hey everyone, updating to 1.48 and I am having an issue I am not completely sure what to do with. I am getting the attached error referencing the new model-cache volume. Do I need to define a new location (such as /mnt/immich-cache:/cache) for this volume? Thank you!

27 Replies
You will need to include the volume at the end of the file, as shown here

Thanks, that did the trick. Sorry I am probably ahead of the docs on this one. Do I need to re-run any of the ML jobs after the upgrade or will that happen automatically?
No you don't need to re run any ML job now
it is a drop-in replacement for the previous implementation
new features coming soon
Thanks and I do see that volume on the compose example now. I just plain missed it. Thank you and the other devs for your work on this project.
No problem! Enjoy! š
One more thing, before you answered I had tried to run the object detection and got this error. Is this to be expected?

No, good catch, probably a bug
Should I grab the output from the immich-server container for you?
yes please
So we have to add that to the volume at the end and make changes to the machine learning section?
Correct, it is outline in the release note
GitHub
Release v1.48.0 Ā· immich-app/immich
February 2023 (version V1.48.0)
š“ Breaking Changes and Important Notes for this release š“
The server and the mobile app must be on the same version to function correctly.
There are changes to the ...
I found the error
[Nest] 1 - 02/21/2023, 8:13:59 PM ERROR [ExceptionsHandler] column si.id does not exist
QueryFailedError: column si.id does not exist
at PostgresQueryRunner.query (/usr/src/app/node_modules/typeorm/driver/postgres/PostgresQueryRunner.js:211:19)
at runMicrotasks (<anonymous>)
at processTicksAndRejections (node:internal/process/task_queues:96:5)
at async SelectQueryBuilder.loadRawResults (/usr/src/app/node_modules/typeorm/query-builder/SelectQueryBuilder.js:2132:25)
at async SelectQueryBuilder.executeEntitiesAndRawResults (/usr/src/app/node_modules/typeorm/query-builder/SelectQueryBuilder.js:1992:26)
at async SelectQueryBuilder.getRawAndEntities (/usr/src/app/node_modules/typeorm/query-builder/SelectQueryBuilder.js:667:29)
at async SelectQueryBuilder.getMany (/usr/src/app/node_modules/typeorm/query-builder/SelectQueryBuilder.js:733:25)
at async AssetRepository.getAssetWithNoSmartInfo (/usr/src/app/dist/apps/immich/apps/immich/src/api-v1/asset/asset-repository.js:47:16)
at async JobService.run (/usr/src/app/dist/apps/immich/apps/immich/src/api-v1/job/job.service.js:72:23)
at async JobController.sendJobCommand (/usr/src/app/dist/apps/immich/apps/immich/src/api-v1/job/job.controller.js:32:20)
This is what I am seeing, each time object detection is clicked.
All the other jobs seem to run with no issues (the "Missing" button -- I did not try clicking "All")
yeah
got it fixed
š
Awesome. I noticed I do not see my other user in the stats page, but she has not updated the app on her phone yet. I'll try clearing cache again though.
good catch
another bug
š
Her photos are there, just seems like the counts are combined now....maybe
yeah we change some queries from string to a more streamline api processes, so something might have missed
I can take a look to fix it
thank you
will push the fix over the hotfix release soon
Good deal. I am not a dev so this is about all I can do to help out š Thanks again!!
Hi, I have an issue with the new ML container also. I have a folder mounted in for cache but I get errors about flask not being installed.
I believe its because I'm running as non-root user (1000:1000) but the pip install in the new ML docker container is installing dependencies as root
There is a report issue with a user running K8S. I am unsure what is happening there as the installation is baked into the image
@_jx2 I get the same errors with plain docker run commands as well. I don't think this is strictly related to k8s. I just opened a parallel issue in case the k8s one turns out to just be k8s related
I think I know what is the problem is
I am building a test image and I will let you guys know once it done and you can test
Yeeep, I see your review and have updated both my BUG and the K8s BUG linking the MR
TYTY for being so quick with this @Alex Tran
yeah I can't sleep with these bugs
Can you guys try running this image
ghcr.io/immich-app/immich-machine-learning:pr-1830
?
let me know if it resolves the permission issue
Looks good!
Sorry for the late response
I have put
ghcr.io/immich-app/immich-machine-learning:pr-1830
in my docker compose and everything starts up nicely too