Skip to content

Latest commit

 

History

History
85 lines (53 loc) · 3.85 KB

readme.md

File metadata and controls

85 lines (53 loc) · 3.85 KB

dask gateway cluster

Overall goals and benefits

To be able to launch and configure dask cluster workers for parallell processing for ML training and large data loads from within python with minimal effort.

installation

server

conda install -c conda-forge dask-gateway-server

kan be done on kubernetes, still working on making that work properly

client (match version of the server, now lates is 0.7.1)

conda install -c conda-forge dask-gateway

kubernetes control

uses the kube config file dask-cluster.cfg make kubernetes find it by: export KUBECONFIG=dask-cluster.cfg (path this smarter so you can move around in your folders)

(note: helm reads this KUBECONFIG too) (note2: not pushing this cfg to git for passwords / secrets. ask for it)

then you can do:

kubectl --namespace dask-gateway get deployment NAME READY UP-TO-DATE AVAILABLE AGE gateway-dask-gateway 1/1 1 1 57d scheduler-proxy-dask-gateway 1/1 1 1 57d web-proxy-dask-gateway 1/1 1 1 57d

kubectl --namespace dask-gateway get pods

kubectl --namespace dask-gateway log gateway-dask-gateway-65f5485c8b-s7bmg

helm

helm env

helm --namespace dask-gateway list NAME NAMESPACE REVISION UPDATED STATUS CHART APP VERSION dask-gateway dask-gateway 4 2020-04-17 10:44:43.9856887 +0200 CEST deployed dask-gateway-0.7.1 0.7.1

helm uninstall [NAME]

helm install

(need to have the helm repo for dask-gateway added for this to work)

helm repo list

RELEASE=dask-gateway NAMESPACE=dask-gateway VERSION=0.7.1

helm upgrade --install --namespace $NAMESPACE --version $VERSION --values values.yaml $RELEASE dask-gateway/dask-gateway Release "dask-gateway" has been upgraded. Happy Helming!

check and get the ips. need external ip to talk

kubectl get service --namespace dask-gateway if its shown as pending, you have a problem.

what is working and what is not at the moment

the deployment looks ok on the kude cluster but does not get an external ip. I suspect that it need to be configure in the values.yaml file with the proper external ip and port. Devops team?

When it has an external ip, on the previous version 0.6.1, I had problem launching new workers / gateways? on the kubernetes cluster, but it worked great locally. Maybe the problem is gone now with version 0.7.1? or still needs to be solved.

Python setup code: from dask_gateway import Gateway
gateway = Gateway(address='172.30.102.126') # ip from kubectl --namespace dask-gateway get services EXTERNAL-IP gateway.list_clusters()
[]

gateway.new_cluster()
GatewayCluster<dask-gateway.952d83a481f6491a8d8e69cff2322413, status=running>

gateway.list_clusters()
[ClusterReport<name=dask-gateway.952d83a481f6491a8d8e69cff2322413, status=RUNNING>]