305 enable post processing on each rank

[spossann]

Solves the following issue(s):

Closes #305

Test with mpirun -n 2 python src/struphy/post_processing/tests/test_pproc.py.

[Copilot]

Pull request overview

This PR targets issue #305 by adding an option to run post-processing across MPI ranks (instead of only rank 0), and adds an MPI test that compares serial vs parallel post-processing outputs.

Changes:

• Add a parallel_pproc flag to Simulation.pproc() / PostProcessor and introduce rank-aware loops + reductions in post-processing.
• Add an MPI regression test comparing results from serial vs parallel post-processing.
• Adjust simulation startup logging and modify initial weight-reset behavior in the Vlasov–Ampère model.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 7 comments.

File	Description
src/struphy/simulation/sim.py	Adds parallel_pproc option to pproc() and updates run-start logging.
src/struphy/post_processing/post_processing_tools.py	Implements parallel post-processing path (per-rank processing + MPI reductions) and rank-0-only output writes.
src/struphy/post_processing/tests/test_pproc.py	Adds an MPI test comparing serial vs parallel post-processing results (currently includes interactive plotting).
src/struphy/models/vlasov_ampere_one_species.py	Comments out particle weight reset after initial Poisson solve.

Comments suppressed due to low confidence (9)

src/struphy/post_processing/post_processing_tools.py:674

• This print() logs the entire reduced temp_val array; this is very noisy and can severely impact performance for large grids. Prefer logger.debug and only log metadata (e.g., shape).

                            print(f"{self.rank = } after\n {temp_val = }")

src/struphy/post_processing/post_processing_tools.py:677

• This print() logs the full contents of temp_val[j] during MPI reduction. For realistic grid sizes this will overwhelm stdout and slow post-processing substantially. Prefer logger.debug and log only shape/summary.

                                print(f"{self.rank = } before \n{temp_val[j].shape = }\n{temp_val[j] = }")

src/struphy/post_processing/post_processing_tools.py:692

• This print() logs the full reduced component array temp_val[j], which is extremely verbose and can degrade performance. Prefer logger.debug with shape/summary only.

                                print(f"{self.rank = } after\n {temp_val[j] = }")

src/struphy/post_processing/post_processing_tools.py:958

• Unconditional print() in library code will spam stdout for MPI jobs and makes output hard to parse. Prefer logger.debug/info (and avoid interpolating expensive values) so users can control verbosity.

        print(f"{self.rank} starting post-processing of distribution functions for {path_kinetic_species} ...")

src/struphy/post_processing/post_processing_tools.py:1010

• This unconditional print() inside the per-slice loop will create a lot of noisy output for multi-rank runs. Prefer logger.debug or remove it.

                print(f"{rank = } ----------------------------")

src/struphy/post_processing/post_processing_tools.py:1023

• Avoid computing and printing xp.sum(...) in the hot post-processing path: it can be expensive (and forces materialization of arrays) and spams stdout. If progress logging is needed, use logger.debug without expensive summaries.

            print(f"{self.rank =} with {xp.sum(data) =} and {xp.sum(data_df) =}")

src/struphy/post_processing/post_processing_tools.py:1053

• This second print() with xp.sum(...) repeats expensive reductions/logging and will spam stdout. Prefer logger.debug without forcing array reductions.

            print(f"{self.rank =} with {xp.sum(data) =} and {xp.sum(data_df) =}")

src/struphy/post_processing/post_processing_tools.py:1055

• Unconditional print() in post-processing adds noisy output in MPI jobs. Prefer logger.debug/info so verbosity is configurable.

            print(f"{self.rank =} done.")

src/struphy/post_processing/tests/test_pproc.py:8

• Tests should not depend on an interactive Matplotlib backend; importing pyplot without forcing a non-interactive backend can fail/hang in headless CI. Force the Agg backend before importing pyplot.

from matplotlib import pyplot as plt

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.