gcp.md

Driverless AI Deployment Template for Google Cloud Run

This template extends the implementation of Local Rest Scorer here and scores data using Driverless AI Mojo pipeline with Java Runtime.

The docker image that is built by this project can be pushed to gcr.io and used for scoring Driverless AI Mojos in Google Cloud Run. The user needs to provide Driverless AI license key and model's pipeline.mojo file as input for scoring. The versions of software used to create the template like mojo runtime are listed here.

Building

Since there is a direct dependency on the separate, above mentioned project local-rest-scorer it is best to build this project from the root directory.

Make sure you are in the working directory mlops-dai-runtimes. Typing pwd in the terminal shell should have a similar output to /my/path/to/mlops-dai-runtimes.

• Run the following command:

  ./gradlew build

  and the docker image required for Google Cloud Run will be in the directory gcp-cloud-run/build:

  /path/to/mlops-dai-runtimes/gcp-cloud-run/build/jib-image.tar
  

• Load the resulting jib-image.tar file to docker:

  docker load < /path/to/mlops-dai-runtimes/gcp-cloud-run/build/jib-image.tar
  

• Follow the steps explained here in Google Documentation: https://cloud.google.com/run/docs/building/containers, to push the container to gcr.io

Deploying

To deploy the container follow the steps in Google Documentation here: https://cloud.google.com/run/docs/deploying

There is one requirement for the container. You MUST include the following environment variables:

• MOJO_GCS_PATH = gs://path/to/pipeline.mojo
• LICENSE_GCS_PATH = gs://path/to/driverless/ai/license.sig

Scoring

On a successful deployment to Google Cloud Run, you will be provided an endpoint that can be scored against.

The api for scoring is the same as the Local Rest Scorer

To access api information: https://<google provided endpoint>/swagger-ui/index.html

To score the model: https://<google provided endpoint>/model/score

Sample curl request:

curl -X POST -H "Content-Type: application/json" \
     -d '{"includeFieldsInOutput":null,"noFieldNamesInOutput":null,"idField":null, \ 
          "fields":["LIMIT_BAL","SEX","EDUCATION","MARRIAGE","AGE","PAY_1","PAY_2", \
                    "PAY_3","PAY_4","PAY_5","PAY_6","BILL_AMT1","BILL_AMT2","BILL_AMT3", \
                    "BILL_AMT4","BILL_AMT5","BILL_AMT6","PAY_AMT1","PAY_AMT2","PAY_AMT3", \
                    "PAY_AMT4","PAY_AMT5","PAY_AMT6"], \
          "rows":[["0","text","text","text","0","0","0","0","0","0","0","0","0","0","0","0","0","0","0","0","0","0","0"]]}' \
     https://<google provided endpoint>/model/score