- kubernetes 1.19+
- kubeflow 1.3+
git clone https://github.com/jimthompson5802/dask-kubeflow
# install as editable package
pip install -e dask_kubeflow
- GIT Extension Pack
- PyLint install (conda)
Key class: KubeflowCluster()
with these methods:
__init__()
: instantiates the DASK cluster. Starts scheduler, workers and enables the DASK Scheduler UI.scale()
: scales worker tasks up/downclose()
: shuts down the clusterworker_count
: returns tuple containing(requested_number_of_worker, ready_worker_count)
scheduer_service_address
: returns string to be used by DASKClient
in connecting to DASK Scheduler.wait_for_workers()
: method to wait for number of ready workers to equal requested number of workers.
- Proof-of-concept of using k8s api to manipulate DASK k8s resources in
kubeflow
enabled cluster. - All methods described in the High-level Design Outline implemented.
- Support for
kubeflow
specific constructs, such asistio
VirtualService
andEnvoyFilter
. - Example Jupyter Notebook