Documentation: https://snek5000.readthedocs.io/
Snek5000 is a Python package which allows one to write Fluidsim solvers based for the simulations on the Fortran CFD code Nek5000. There are open-source solvers (in particular snek5000-phill, snek5000-cbox and snek5000-tgv) and it's not difficult to write your own solver based on your Nek5000 cases (as shown in this tutorial).
With a Snek5000-Fluidsim solver, it becomes very easy to
- launch/restart simulations with Python scripts and terminal commands,
- load simulations, read the associated parameters/data and produce nice figures/movies.
Snek5000 can be seen as a workflow manager for Nek5000 or a Python wrapper around Nek5000. It uses Nek5000 on the background and is thus NOT a rewrite of Nek5000!
Snek5000 is powered by nice Python packages such as Snakemake, Fluidsim, Pymech, Matplotlib, Jinja, Pytest, Xarray, etc.
Install it as follows:
export NEK_SOURCE_ROOT="/path/to/Nek5000"
pip install snek5000See here for detailed installation instructions.
The snek5000 Python
API allows
you to launch/restart/load simulations. For example, the periodic hill Nek5000
example can be
launched with our snek5000-phill solver (installable with pip install snek5000-phill) as follow:
from phill.solver import Simul
params = Simul.create_default_params()
# modify parameters as needed, for example
params.output.sub_directory = "examples_snek_phill"
params.short_name_type_run = "readme"
params.oper.nx = 12
params.oper.ny = 10
params.oper.nz = 8
params.nek.general.num_steps = 10
...
# instantiate the object representing the simulation
sim = Simul(params)
# compile and launch the simulation (blocking)
sim.make.exec("run_fg")A simulation directory is created automatically (with this example, something
like
~/Sim_data/examples_snek_phill/phill_readme_12x10x8_V1.x1.x1._2022-10-27_15-21-58).
Then, the simulation object can be recreated from this directory. An easy way
would be to go into this directory, start IPython with the snek-ipy-load
command, and run:
# get/print the simulation parameters from the object
sim.params
# few examples of various read and plots
sim.output.print_stdout.plot_dt()
sim.output.print_stdout.plot_nb_iterations()
sim.output.history_points.plot()
sim.output.history_points.coords
data = sim.output.history_points.load_1point(2)
sim.output.phys_fields.plot_hexa()
sim.output.phys_fields.animate("pressure", interactive=True)
sim.output.phys_fields.animate(
"pressure", dt_frame_in_sec=0.1, equation="y=0.5", save_file="my_great_movie.gif"
)For example, this movie has been produced by a sim.output.phys_fields.animate
call from a snek5000-cbox
simulation:
movie.mp4
Check out the tutorials to learn how to use Snek5000.
Need more reasons to use snek5000?
- Saves you from the trouble in setting up multiple source files (
.box,.par,SIZE) - Uses sensible names and defaults for the parameters
- Avoids typos and human errors thanks to a nice parameter container object
- Records metadata related to the simulation into human and machine readable files (
params_simul.xml,config_simul.yml) - Checks for consistency of parameters
- Automatically sets some parameters as Python properties
-
Out of source build (per run), which can be inspected or executed using the conventional
makenekfor debugging -
Reproducible workflows, not susceptible to changes in environment variables by default
-
Scriptable simulation execution allowing parametric studies
-
Easy to load simulation for performing offline post-processing and restarting the simulation
-
Better than Bash scripting like:
# Build case cd src/phill/ CASE="phill" echo "$CASE.box" | genbox mv -f box.re2 phill.re2 echo "$CASE\n0.01" | genmap FFLAGS="-mcmodel=medium -march=native" CFLAGS="-mcmodel=medium -march=native" makenek cd - # Run case cd src/phill/ nekmpi $CASE <nb_procs> # foreground nekbmpi $CASE <nb_procs> # background cd - # Clean makenek clean
-
Use of Snakemake which is similar to GNU Make, but allows one to blend Bash and Python scripting and uses simple YAML files for managing custom configurations of compilers and flags for different computers.
- User friendly, modular, object oriented API
- Reuse of code (inheritance)
- Tested with a good code coverage (>90%)
- Yet another layer... with the possible associated bugs :-)
- Requires some basic knowledge of Python to use (not really a big issue, to be honest).
- Deep modification of solvers requires learning how Snakemake functions and how to write Jinja templates (which are not so hard, btw)
Contributions are welcome! You can help by testing out the code, filing issues and submitting patches. See contributing guidelines.
