[RAY ODH] KR2: Ray in ODH

MichaelClifford / moc-ray-demo

A demonstration of using the Ray distributed computing platform with Jupyter on Open Data Hub.

Apache License 2.0

0 stars 0 forks source link

[RAY ODH] KR2: Ray in ODH #14

Open MichaelClifford opened 2 years ago

MichaelClifford commented 2 years ago

Kustomize overlays: these will be adapted from the prototype Ray+ODH integration that is currently deployed on the OPF Smaug cluster. This will take the form of one or more pull-requests against the ODH project. Success upon merging.

Acceptance Criteria:

[ ] This issue is done when the PR to integrate Ray as default option for ODH is merged

MichaelClifford commented 2 years ago

There is an ongoing PR by Harshad to ODH were they are adding the existing Ray deployment files to odh-manifests: https://github.com/opendatahub-io/odh-manifests/pull/573/files

For this KR we should follow and contribute to ODH from that PR.

cc @erikerlandson

MichaelClifford commented 2 years ago

Started working on my own fork of odh-manifests and created a new feature branch from the PR above. I will use my fork's branch to keep track of my work on this issue. This branch also represents the deployment used for Ray on our teams DS cluster.

Made two updates today:

fixed ray.init() not working issue: https://github.com/opendatahub-io/odh-manifests/commit/f1ac5353eddef02ee549d5f0d66dd7b3a3172f5c
updated configmaps for GPU: https://github.com/opendatahub-io/odh-manifests/commit/a6b3ed74301c597862d2a73c9910bd380291081c

MichaelClifford commented 2 years ago

ray-workflows do not work in "client mode", that is on a cluster, at the moment. Here is the issue where the ray project is working on making this feature available. https://github.com/ray-project/ray/pull/24789

Will skip this Ray feature for now.