Pipelines and job within kaos rely on Pachyderm, which uses the following descriptions for distinguishing different states.
The following states are valid when deploying a training job in kaos.
Job initialized with inputs (i.e. source and data bundle)
Job running based on process (i.e. train vs. build)
Job in merging state prior to completion
Job successfully completed
Job failed - check logs
Job killed by user
The following states are valid when deploying a serve pipeline in kaos.
Serve pipeline awaiting inputs
Serve pipeline is initialized with inputs (i.e. source bundle)
Serve pipeline running with valid endpoint
Serve pipeline is restarting with new inputs
Serve pipeline failed (likely due to network or resource problems)