-
Notifications
You must be signed in to change notification settings - Fork 135
Description
The checkpointing system is pretty rudimentary at the moment, it simply checks if a function in workflow.task_list has been run (by name) and skips the function if it has. This is not optimal as it means a lot of repeated simulations will take place. E.g., if 60 forward simulations are launched and all but one finishes, restarting the workflow will cause all 60 jobs to be re-run.
Additionally, there is no clear way to interact with the state file. At the moment it involves directly editing the text file, but that is a bit clunky and prone to error. It would be great to have a command line tool to check, edit and clear a statefile. The state file should likely also be a hidden file to prevent casual users from accidentally deleting or editing it manually.
Tasks to be completed are then:
- Improve task-level checkpointing (this can be done by checking solver logs, or generated waveforms in the 'traces' dir.)
- Create a
seisflows statecommand line option to manipulate the state file - Generate the entire state file at the setup stage, as opposed to building it throughout the first iteration
- Make the state file a hidden file