Hi, I'm running Kubeflow (V0.5) on AWS, using Pipelines I want to use the Tensorboard button in the pipelines metadata result.

I saved my pipeline result into a private S3 bucket with the following code:

metadata = {
    'outputs' : [{
      'type': 'tensorboard',
      'source': 's3://'+ bucket_name + '/' + destination,
    }]}

with open('/mlpipeline-ui-metadata.json', 'w') as f:
        json.dump(metadata, f)`

But when I try opening the Artifacts results of that operation I only get the circle loading bar and nothing happens. If I look at the Network Request i find a POST REQUEST with the following response :

{"outputs": [{"type": "tensorboard", "source": "s3://BUCKET_NAME/FOLDER1/FOLDER2/1564392073"}]}

I tried using GCS instead to save the file and in that case, the Tensorboard button appears and I can start the TB instance. (It fails because I don't have any GC creds (MountVolume.SetUp failed for volume "gcp-credentials" : secrets "user-gcp-sa" not found).

How should I save my summary writers to be able to start the TB instance from the pipeline artifacts on AWS/ S3?

Thank your for your help,

EDIT :

I found that issue kubeflow/pipelines#337 in the pipeline project of Kubeflow. Can someone confirm that Tensorboard through Pipelines on AWS stored in S3 is currently not supported ?