Google Kubernetes Engine#

This guide offers recommended configurations and settings unique to Google Kubernetes Engine (GKE) in Data Science & AI Workbench. These should be used to augment the generic requirements offered on our primary requirements page.

Instance types#

  • Minimum: n2-standard-8

  • Recommended: n2-standard-16 or larger


We have found that the Basic SSD and High Scale SSD performance tiers for Google Cloud Filestore provide competent performance for use with Workbench. The minimum storage sizes are 2.5TiB for the Basic SSD tier and 10TiB for the High Scale SSD Tier. Because Filestore volumes supports the ReadWriteMany access mode, a single volume can serve both the anaconda-storage and anaconda-peristence storage requirements.


  • Do not change the default export options. In particular, squash-mode should remain as NO_ROOT_SQUASH.

  • When granting access to this volume, include not just the GKE cluster itself, but to the Administration server as well. This will enable the server to be used to initialize relevant directories, set permissions, perform backups, and so forth.

  • The ownership of the root directory should be set to <UID>:0, where <UID> is the non-zero UID selected to run the Workbench containers.

  • The permissions should be set to 770 or 775; group writability is required.

GPU Support#

Please see this guide for adding GPU resources to your GKE cluster.