SherlockShemol
commented
2 days ago two optional solution:
- make FederatedLearningJob CRD modifiable, and once the config is modified, we delete and create the specific worker with new config.
- make FederatedLearningJob CRD unmodifiable.The reason is above.
Mr. Tang Ming suggests that, unlike inference tasks, Federated Learning should be treated as a job that should not be modified once it is running. Modifying it essentially means restarting the job, so users should delete the existing job and create a new FederatedLearningJob after making any changes. Therefore, users should be directly prohibited from modifying the CRD.
Could you please provide your feedback? @tangming1996 @hsj576 @MooreZheng