Stateful status of experiments #716
Description
Component
Run
Desired use case or feature
- Currently, it is not always clear which experiments/workloads completed successfully.
- The proposal is to add persistent state information to the results dir
- This could also simplify synchronization -- between benchmark replicas, between monitoring and benchmarking
Proposed solution
- Add to the benchmark wrappers code to save a status file
- Rename status file, as benchmark progresses (rename should be an atomic operation), e.g., --
status.READY,status.START,status.COMPLETE,status.SUCCESS,status.FAIL - Check for completion indicators before changing status (e.g., look for expected string in logs)
- Add timing info inside file
- Add UI to allow list of results with status
- Extract start/end timing from the status
Alternatives
No response
Additional context or screenshots
No response