[WIP] Add Dockerfile for modelmesh-minio-examples image

[tedhtchang]

Motivation

Provide instructions to build a MinIO model examples image

Resolves #297 #302

---

***** UPDATE Jan 24, 2023 *****

The files in this PR are entirely moved to this new repository:
https://github.com/kserve/modelmesh-minio-examples/

---

[kserve-oss-bot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: tedhtchang
To complete the pull request process, please assign animeshsingh after the PR has been reviewed.
You can assign the PR to them by writing /assign @animeshsingh in a comment when ready.

The full list of commands accepted by this bot can be found here.

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[tedhtchang]

/cc @ckadner

[ckadner]

Hi @tedhtchang -- thank you for your PR. Do you know where all those large model files originate from? Since they are so large, I think it would be great if we don't have to keep a copy of those files in the repo, instead, maybe they could be copied/downloaded from their original source at the time we build the image? Or whenever we need them for testing. And delete them afterwards.

[rafvasq]

Thanks @tedhtchang this is really helpful.

Probably a question for @njhill and @ckadner, I'm wondering whether fvt would be the correct directory to host both Dockerfiles. Maybe it'd be a new quickstart directory for the examples-only Dockerfile while the FVT-related image stays here?

[njhill]

Yes thanks @tedhtchang. I think we should add just the Dockerfile(s) for now while we figure out where to host the larger model files, maybe we could do that in a separate PR so that we keep this branch with all the files until we figure that out.

I agree with @rafvasq that the existing fvt dir probably isn't the best place for this to live. Probably it makes sense to have a new top-level dir called something like mino_build, since I can't really see anywhere else that it would fit very well.

[ckadner]

Hi Ted,

I moved the files to into a new minio_examples folder at the project root level, for now.

Although I don’t think we should add this to the modelmesh-serving repo — a 600 MB addition is just too large.

I think we create a new repo soon modelmesh-minio-examples with it's own CI/CD and figure out a way to offload the model files either to COS buckets with public read-access (or even credentialed access for MM maintainers only) or we use another mechanism like Git LFS for Large File Storage as @njhill has suggested

[ckadner]

For today, I will just build it off of your branch, push it to DockerHub, deploy it on an OCP cluster and test it.

[ckadner]

as Nick pointed out, doing a chown after the COPY layers causes the final Docker image to be about 86 MB larger.

$ docker images | grep "modelmesh-minio-examples"

kserve/modelmesh-minio-examples   chown_before_cp      15eb3d6f159e   45 minutes ago   438MB
kserve/modelmesh-minio-examples   chown_after_cp       ad16135035ff   2 hours ago      524MB
kserve/modelmesh-minio-examples   cp-chown             7579fe5ccdea   57 minutes ago   438MB
quay.io/cloudservices/minio       RELEASE.2021-06-17   1377bf686561   11 months ago    351MB
kserve/modelmesh-minio-examples   latest               c773b3dc52e5   11 months ago    382MB
kserve/modelmesh-minio-examples   v0.9.0               c773b3dc52e5   11 months ago    382MB

I reverted to COPY --chown 1000:0 ... in commit 33c2f17

[ckadner]

https://hub.docker.com/r/kserve/modelmesh-minio-examples/tags
https://hub.docker.com/r/kserve/modelmesh-minio-dev-examples/tags

[ckadner]

@rafvasq -- I pushed a new modelmesh-minio-dev-examples:latest image with the FVT models you added

https://hub.docker.com/r/kserve/modelmesh-minio-dev-examples/tags

[ckadner]

I created a new repository https://github.com/kserve/modelmesh-minio-examples/ from the commits in this PR with a few modifications to

• remove the biggest unused models:
  • fvt/onnx/onnx-mobilenet (14 MB)
  • fvt/onnx/onnx-pytorch-converted (4.5 MB)
  • fvt/onnx/onnx-test-model (38 MB)
  • fvt/tensorflow/tfmnist (78 MB)
  • fvt/tensorflow/tfmnistnew (39 MB)
• removing any duplicate models in favor of doing COPYs in the Dockerfile:
  • fvt/keras (6.5 MB)
  • fvt/pytorch/pytorch-mar-dup (4.6 MB)
  • fvt/tensorflow/keras-mnist (6.5 MB)
  • fvt/tensorflow/keras-mnistnew (6.5 MB)
  • fvt/tensorflow/mnist-dup.savedmodel (39 MB)
  • fvt/tensorflow/mnist.savedmodel (39 MB)
  • tensorflow/mnist.savedmodel (39 MB)

Shaving off about 315 MB from the size of the repo -- before any of those files were ever committed since Git would keep a history of the files and the repo would stay larger even if we deleted those file subsequently.

[ckadner]

@tedhtchang -- thank you again for this PR. It was really helpful to get the new https://github.com/kserve/modelmesh-minio-examples/ repo seeded.

If you don't mind keeping your fork and branch alive for the coming weeks so we can go back to it in case some of my trimming down was over eager ;-)

[tedhtchang]

@tedhtchang -- thank you again for this PR. It was really helpful to get the new https://github.com/kserve/modelmesh-minio-examples/ repo seeded.

Good good. Thanks.

If you don't mind keeping your fork and branch alive for the coming weeks so we can go back to it in case some of my trimming down was over eager ;-)

I don't mind.