Update for Kubeflow 1.4 release #7
Conversation
ca-scribner
commented
Sep 10, 2021
ca-scribner
commented
- updates charm to kubeflow 1.4
- adds pytest operator/refactors CI
| #!/usr/bin/env bash | ||
| set -eux | ||
| MODEL=$1 | ||
| echo "MODEL=$MODEL" | ||
| wget https://raw.githubusercontent.com/google/ml-metadata/master/ml_metadata/proto/metadata_store.proto | ||
| wget https://raw.githubusercontent.com/google/ml-metadata/master/ml_metadata/proto/metadata_store_service.proto | ||
| wget -O- https://github.com/fullstorydev/grpcurl/releases/download/v1.8.0/grpcurl_1.8.0_linux_x86_64.tar.gz | tar -xzv | ||
| mkdir -p ml_metadata/proto/ | ||
| mv metadata_store.proto ml_metadata/proto/ | ||
| SERVICE=$(kubectl get services/mlmd -n $MODEL -oyaml | yq e .spec.clusterIP -) | ||
| ./grpcurl -v --proto=metadata_store_service.proto --plaintext $SERVICE:8080 ml_metadata.MetadataStoreService/GetArtifacts |
There was a problem hiding this comment.
Should I python-ify this? The kubectl call could be the python k8s sdk, wgets could be in python, then grpcurl could be a subprocess.run (or maybe connect some other way)
Alternatively, we could use the mlmd python api and just call to it using the native ways in python
There was a problem hiding this comment.
(or should we table this for future and prioritize the 1.4 release first)
There was a problem hiding this comment.
Ideally, this seems like it should be a pytest fixture that does all the setup, and returns a MetadataStore (judging by my reading of the docs) object . I think it's also fine to punt on that for now and focus on 1.4.
| #!/usr/bin/env bash | ||
| set -eux | ||
| MODEL=$1 | ||
| echo "MODEL=$MODEL" | ||
| wget https://raw.githubusercontent.com/google/ml-metadata/master/ml_metadata/proto/metadata_store.proto | ||
| wget https://raw.githubusercontent.com/google/ml-metadata/master/ml_metadata/proto/metadata_store_service.proto | ||
| wget -O- https://github.com/fullstorydev/grpcurl/releases/download/v1.8.0/grpcurl_1.8.0_linux_x86_64.tar.gz | tar -xzv | ||
| mkdir -p ml_metadata/proto/ | ||
| mv metadata_store.proto ml_metadata/proto/ | ||
| SERVICE=$(kubectl get services/mlmd -n $MODEL -oyaml | yq e .spec.clusterIP -) | ||
| ./grpcurl -v --proto=metadata_store_service.proto --plaintext $SERVICE:8080 ml_metadata.MetadataStoreService/GetArtifacts |
There was a problem hiding this comment.
Ideally, this seems like it should be a pytest fixture that does all the setup, and returns a MetadataStore (judging by my reading of the docs) object . I think it's also fine to punt on that for now and focus on 1.4.
