jade-data-repository ·

The Terra Data Repository built by the Jade team as part of the Data Biosphere.

• Support complex, user-specified schemas and different types of data
• Fine-grained access control
• Support share-in-place, copying data is expensive
• Cloud-transparency: support off-the-shelf tool access to data

Documentation

This repository is currently designed to be deployed inside of a Google Cloud Platform project to manage tabular data inside of BigQuery datasets and file data inside of Google Cloud Storage buckets. The project setup has been automated via Terraform.

Terraforming a project

Clone the terraform-jade repo and follow the terrform commands there to set it up.

Note: those Terraform scripts and the deployment script here make an assumption that they can retrieve secrets from a Vault server at certain paths. If you are standing this repo outside of the Broad infrastructure, the current best alternative is to set up a Vault server to supply these secrets until we implement less opinionated way to supply secrets to the deployment scripts.

Setting up access

Now that your cluster is Terraformed, you need to be able to access it with kubectl commands. To do this, go to the Google Cloud Console -> Kubernetes Engine -> Clusters click the Connect button next to you cluster info. Copy the command and execute it on your local system. Now, if you click on Docker -> Kubernetes you should see a check next to the cluster you just created.

Your cluster will be running under the default compute service account for the project. This account is used by Google Kubernetes Engine (GKE) to pull container images clusters by default. It is in the form [PROJECT_NUMBER][email protected], where [PROJECT-NUMBER] is the GCP project number of the project that is running the Google Kubernetes Engine cluster.

Note: this next part is specific to the Broad setup. If you are standing this up externally, you will need an instance of Google Container Registry (GCR) where you can put images to be deployed in GKE.

Give your service account access to dev GCR:

gsutil iam ch serviceAccount:[PROJECT_NUMBER][email protected]:objectViewer gs://artifacts.broad-jade-dev.appspot.com

Using cloud code and skaffold

Once you have deployed to GKE, if you are developing on the API it might be useful to update the API container image without having to go through a full re-deploy of the Kubernetes namespace. CloudCode for IntelliJ makes this simple. First install skaffold:

brew install skaffold

Next, enable the CloudCode plugin for IntelliJ.

Then you should be able to either Deploy to Kubernetes or Develop on Kubernetes from the run configurations menu.

Build and Run Locally

Set up

You must have authenticated with google for application-default credentials:

gcloud auth application-default login

and login with an account that has access to your project. This will save credentials locally. If you are using multiple accounts, you can switch to the correct one using this command:

gcloud config set account <account email>

Then you must specify a google project to use. Run this command:

gcloud config set project <project-name>

To see what you currently have set, use: gcloud config list

When running locally, we are not using the proxy. Therefore, the system doesn't know your user email. Edit the src/main/resources/application.properties file and set the userEmail field. If you are running sam locally, set sam.basePath to https://local.broadinstitute.org:50443.

Run linters and unit tests

If you are making code changes, run: ./gradlew check

Run jade locally

To run jade locally: ./gradlew bootRun

To run jade locally and wait for debugger to attach on port 5005: ./gradlew bootRun --debug-jvm

To have the code hot reload, enable automatic builds in intellij, go to: Preferences -> Build, Execution, Deployment -> Compiler and select Build project automatically

The swagger page is: https://local.broadinstitue.org:8080

Run connected and integration tests

./gradlew testConnected

The integration tests will hit the data repo running in the broad-jade-integration envrionment by default. To use a different data-repo, edit the src/main/resources/application-integration.properties file and specify the URL. Before you run the integration tests, you need to generate the correct pem file by running ./render_configs

To run the tests, use: ./gradlew testIntegration

Swagger Codegen (deprecated)

We are using swagger-codegen to generate code from the swagger (OpenAPI) document. Therefore, in order to build you need to have the codegen tool installed from swagger-codegen.

The gradle compile uses swagger-codegen to generate the model and controller interface code into src/generated/java/bio/terra/models and src/generated/java/bio/terra/controllers respectively. Code in src/generated is not committed to github. It is generated as needed.

Adding an endpoint to the API source (data-repository-openapi.yaml) will generate the endpoint definition in the appropriate controller interface file. Swagger-codegen provides a default implementation of the endpoint that generates a NOT_IMPLEMENTED return. You add the actual implementation of the new interface by editing the Jade controller code in src/main/java/bio/terra/controller. That overrides the default interface implementation.

Clearly, you can make breaking changes to the API and will have to do the appropriate refactoring in the rest of the code base. For simple addition of fields in a structure or new endpoints, the build will continue to run clean.

In the rare case of wanting to have swagger-codegen create a controller class, in a directory other than a git cloned workspace, run: swagger-codegen generate -i path/to/data-repository-openapi.yaml -l spring -c path/to/config.json

Then copy the files you want into the source tree

skaffold

To render your own local skaffold.yaml run the following with your initials

sed -e 's/TEMP/<initials>/g' skaffold.yaml.template > skaffold.yaml

Run a deployment you must set env var IMAGE_TAG

skaffold run

Add new application property

1. Locally, application properties are controlled by the values in the various application.properties files.
   • application.properties contains the base/default values. A new property should be added here first.

   datarepo.gcs.allowReuseExistingBuckets=false
   

   • You can override the default value for connected and integration tests by adding a line to application-connectedtest.properties and application-integrationtest.properties.

   datarepo.gcs.allowReuseExistingBuckets=true
   

2. Now that we use Helm, the properties also need to be added to the base Data Repo charts.
   • Find the the api-deployment.yaml file.
   • Add a new property under the env section. The formatting below might be messed up, and the yaml is very picky about spaces. So, copy/paste from another variable in the section instead of here.

           {{- if .Values.env.datarepoGcsAllowreuseexistingbuckets }}
           - name: DATAREPO_GCS_ALLOWREUSEEXISTINGBUCKETS
             value: {{ .Values.env.datarepoGcsAllowreuseexistingbuckets | quote }}
           {{- end }}
   

   • Find the the values.yaml file.
   • Add a new line under the env section.

     datarepoGcsAllowreuseexistingbuckets:
   

   • Release a new version of the chart. Talk to DevOps to do this.
3. To override properties for specific environments (e.g. integration), modify the environment-specific override Data Repo charts.
   • Find the deployment.yaml for the specific environment.
   • Add a new line under the env section.

   datarepoGcsAllowreuseexistingbuckets: true
   

   • It's a good idea to test out changes on your developer-namespace before making a PR.
   • Changes to integration, temp, or developer-namespace environments are good with regular PR approval (1 thumb for this repository).
   • Changes to dev or prod need more eyes, and perhaps a group discussion to discuss possible side effects or failure modes.

Developer Notes

Proper Handling of InterruptedException

Care must be taken with a handling of InterruptedException. Code running in stairway flights must expect to receive InterruptedException during waiting states when the pod is being shut down. It is important that exception be allowed to propagate up to the stairway layer, so that proper termination and re-queuing of the flight can be performed.

On the other hand, code running outside of the stairway flight, where the exception can become the response to a REST API, should catch the InterruptedException and replace it with a meaningful exception. Otherwise the caller gets a useless error message.

Deployments

The deployments of Terra Data Repository are:

• Production
• Development
• Development - swagger page