Release Notes for Kaptain 1.2.0

View release-specific information for Kaptain

Release Notes

1.2.0 - Released 16 August 2021

This document describes the new features, caveats, and resolved issues in D2iQ Kaptain.

New features

  • New Kaptain Dashboard for ML workload monitoring
  • Automatic installation of Grafana dashboards for Kaptain monitoring
  • Added Tensorboard Controller for easy deployment of Tensorboards
  • Added Volume Manager to simplify volumes creation and deletion from the UI
  • New Kaptain SDK 0.3.0:
    • New Resources API for fine-grained resource specification
    • Explicit feedback in case a workload is unschedulable because of insufficient resources
    • New Model.build() API for building custom Docker images
    • Content-based conditional image builds to speedup the model development
    • Canary model deployments
    • Configurable SDK logging to suppress debug-level messages by default
    • Improved feedback on failed workload scheduling
    • Auto-generated API documentation

Improvements

  • Konvoy 1.8.2 support
  • Kubernetes 1.21 support

Software updates

  • Kubeflow 1.3.0
  • Argo Workflows 2.12.9
  • Katib 0.11.0
  • KFServing 0.5.1
  • Kubeflow Pipelines 1.5.0
  • KNative Serving 0.18.3
  • JupyterLab 3.0.16

Limitations

  • When running on Konvoy 1.8.2 with GPU, the default base addons version stable-1.20-4.1.0 doesn’t report GPU metrics to Prometheus, and thus the Kaptain Dashboard is unable to display them. Consider using a newer version of the base addons, or the previous version: stable-1.20-4.0.0 in your Konvoy cluster.yaml:
    addons:
      - configRepository: https://github.com/mesosphere/kubernetes-base-addons
        configVersion: stable-1.20-4.0.0
    

Breaking changes

  • Automatic Profile creation on the first login is disabled by default for security purposes. Consult the documentation on onboarding new users for details.