+3
−5
Loading
With the current dataloaders, WorkloadData.start_date can end up being after sim_config.start. Most of the dataloaders set timestep_start and start_date to whatever the first job in the dataset is. This issue is most apparent when the dataloader does any filtering based on start/end so start_date will be whichever job started in the interval. Eventually I think we should refactor the dataloaders to return absolute unix timestamps to avoid all these relative vs non-relative timestep complications. But for now I'm just going to shift the telemetry_start back to match the sim_config.start if needed. This does mean that telemetry_start can now be a negative value, but that should work fine.