Skip to content
This repository has been archived by the owner on Sep 18, 2024. It is now read-only.

DLTS: Get NNI manager IP / port from endpoints #2305

Merged
merged 8 commits into from
Apr 30, 2020

Conversation

Gerhut
Copy link
Member

@Gerhut Gerhut commented Apr 14, 2020

Previously, DLTS training service is using Kubernetes internal network to make to connections between different workers, which is not supported by all deployments.

This PR is switching it into job endpoints, which is officially supported by DLTS. Therefore, a manager job should:

  1. Expose container port of RestServer to public as a job endpoint
  2. Wait the exposed job endpoint be ready, and than get the hostname and host port (which is different from the container port) of the job endpoint
  3. Start trial job with the hostname and host port as the address of RestServer.

@Gerhut Gerhut changed the title Get NNI manager IP / port from endpoints DLTS: Get NNI manager IP / port from endpoints Apr 14, 2020
@Gerhut Gerhut requested a review from QuanluZhang April 20, 2020 15:59
@SparkSnail
Copy link
Contributor

looks fine to me.

@QuanluZhang QuanluZhang merged commit fba2dd5 into microsoft:master Apr 30, 2020
@Gerhut Gerhut deleted the dlts/fix-cross-workers branch May 6, 2020 02:33
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants