noderunner
Run arbitrary Docker containers on ephemeral single-node Dataproc clusters with GCS buckets
mounted as local filesystems via gcsfuse. No Spark, no YARN — just docker run on hardware
you control, accessible to anyone with Dataproc permissions.
Why it exists
In Terra/Broad environments, researchers often have Dataproc permissions but not raw Compute
Engine permissions. noderunner exploits this: it creates a single-node Dataproc cluster, mounts
GCS buckets via gcsfuse, runs you
This is a companion discussion topic for the original entry at github.com/gileshall/noderunner/noderunner-run