GitLab

Support aks of kuberflow training service
Support nnictl set nniManagerIp

* add one more trial job status, EARLY_STOPPED

* fix datastore/nnimanager/mockeddatastore. test/webui/metrics_reader not done. USER_TO_CANCEL

* fix bug

* modifications based on Deshui's comments

* fix bug

* fix bug in remote mode

Add nniManager Ip in nnictl, pai TrainingService and kubeflow TrainingService.
If users set nniManagerIp, pai and kubeflow will use this ip instead of using getIPV4() function.
Web UI will also use this nniManagerIp.

* Adjust sleep position for sdk_test.py

* Exit dispather process if receive Terminate command

* Add comment for sleep change in sdk_test.py

* Use different output folder for ps and worker

* Add cuda_visible_devices env var if gpuNum is 0

[Kubeflow training service] Update kubeflow exp job config schema to support distributed training (#387)

* Support distributed training on tf-operator, for worker and ps

* Update validation rule for kubeflow config

* small code refactor adjustment for private methods

* Use different output folder for ps and worker

* Kubeflow TrainingService support, v1 (#373)

1. Create new Training Service: kubeflow trainning service, use 'kubectl' and kubeflow tfjobs CRD to submit and manage jobs
2. Update nni python SDK to support new kubeflow platform
3. Update nni python SDK's get_sequende_id() implementation, read NNI_TRIAL_SEQ_ID env variable, instead of reading .nni/sequence_id file
4. This version only supports Tensorflow operator. Will add more operators' support in future versions