Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Support production level volume types for model data #128

Closed
lianhao opened this issue Jun 25, 2024 · 4 comments · Fixed by #306
Closed

Support production level volume types for model data #128

lianhao opened this issue Jun 25, 2024 · 4 comments · Fixed by #306
Assignees
Labels
feature New feature or request helm manifest
Milestone

Comments

@lianhao
Copy link
Collaborator

lianhao commented Jun 25, 2024

Currently, our k8s installation (either helm-charts or manifest) have hostPath volume type to store the model data to save model download time. This is NOT ready for production environment, and is NOT suitable for scale outing with multiple nodes.

In production environment, users often use persistent volume(PV/PVC) to store those kind of data. We should give users the options to choose which volume type they want to use.

For PVs, it should have annotation "pv.beta.kubernetes.io/gid: " so non root user in pod can have write access to the volume mounted from this PV. See https://kubernetes.io/docs/tasks/configure-pod-container/configure-persistent-volume-storage/#access-control as reference.

@lianhao lianhao added the feature New feature or request label Jun 25, 2024
@endomorphosis
Copy link

I agree with this, and this is one of the reasons why previous to OPEA I built huggingface model manager, that pulls models from ipfs / s3 / huggingface, and evicts models as needed.

@daisy-ycguo
Copy link
Collaborator

@lianhao can this issue be closed?

@lianhao
Copy link
Collaborator Author

lianhao commented Jul 25, 2024

nope, we didn't have time to resolve this in M3. should target for next release

@moting9
Copy link

moting9 commented Aug 7, 2024

hostpath is easy for getting started, maybe we can have a "sample" dir in the repo, and add diverse nfs, s3 and so on backend pv/pvc to store model files for user reference.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
feature New feature or request helm manifest
Projects
None yet
Development

Successfully merging a pull request may close this issue.

5 participants