Exercises: design before execution
1. Draw the graph
Draw the three analysis stages. Label every arrow with the files it carries. Which jobs can run concurrently? Why must selection wait for all candidate fits?
2. Name the dependencies
A fit reads its model and interval from an INI file, raw data from HDF5, and executes fit_correlator.py. Which of these should appear in its rule's input? What could go wrong if only the INI were declared?
3. Missing inputs and stale outputs
Assume the workflow has finished and all output files exist. An incorrect fit rule lists only the INI in input, although Python also opens the HDF5 file through that INI. The rule code, parameters and INI remain unchanged.
- You replace the HDF5 file with new data; its modification time is newer than the outputs. Why can the dry run incorrectly report that there is nothing to do?
- Which inputs should you add to the rule? Which stages should rerun when the shared raw data change?
- In a separate scenario, the inputs remain unchanged. You remove one candidate output and explicitly request it as a target. Does its fit recreate only that missing file or all four of its outputs?
Also explain why 18 configurations do not mean 18 independent fit jobs in this snapshot.
Compare with the solutions.