tensorboard-controller

README

GSoC 2020 - TENSORBOARD CONTROLLER

Related Closed Issues

• Extend Tensorboard Controller to Support PVCs #5039

• Tensorboard controller creates servers that always mount user-gcp-sa secret #5065

• Tensorboard CR doesn't contain information about the Tensorboard Server being ready or not #5166

Related Pull Requests

• Mount GCP secret only when accessing Google storage #5069

• Add scheduling functionality for Tensorboard servers that use RWO PVCs as log storages #5218

• Add functionality to inform TWA frontend about the status of Tensorboard servers #5259

Prequisites to build and run the controller:

1. GO

2. Docker

3. kustomize

4. kubectl

RUN TENSORBOARD CONTROLLER LOCALLY

Steps:

1. Clone the repository

2. Change directories to components/tensorboard-controller

3. Generate and install manifests and build the controller: make install

4. Run the controller locally: make run

If you want to enable the scheduling functionality for Tensorboard servers that use ReadWriteOnce PVCs as log storages, then set the RWO_PVC_SCHEDULING to true and run: RWO_PVC_SCHEDULING="true" make run

BUILD TENSORBOARD CONTROLLER IMAGE AND DEPLOY TO CLUSTER

1. Clone the repository

2. Change directories to components/tensorboard-controller

3. Generate and install manifests and build the controller: make manifests

4. Build and push the docker image: make docker-build docker-push IMG=YOUR_IMAGE_NAME

5. Deploy the Tensorboard controller: make deploy IMG=YOUR_IMAGE_NAME

If you want to enable the scheduling functionality for Tensorboard servers that use ReadWriteOnce PVCs as log storages, then:

1. Change directories to components/tensorboard-controller/config/manager

2. Modify the manager.yaml file by navigating to the deployment.spec.template.spec field and manually setting the value of the RWO_PVC_SCHEDULING env var to "true" in the manager container.

3. Run: make deploy IMG=YOUR_IMAGE_NAME