Memory leak caused by eq.solve at each step?

[aritra-mukhopadhyay]

I am running a simple code for diffusion in two dimensions, as shown below. However, the memory consumed increases quickly within the for loop, as much as 30GB RAM within a minute. Increasing the number of time steps leads it to eventually being 'Killed' since it runs out of 64GB memory. Is this because somehow the value of the target variable 'C' is stored for all time steps instead of the current one? I am quite new to FiPy, so any clue as to why this is happening would be very helpful :)

Code:

#Diffusion in 2D with periodic grid

#Import necessary modules
from fipy import *
from tqdm import tqdm

#Define the simulation parameters
Lx = 10.0 # length of the domain in x direction
Ly = 10.0 # length of the domain in y direction
nx = 100 # number of cells in x direction
ny = 100 # number of cells in y direction
dx = Lx / nx # cell size in x direction
dy = Ly / ny # cell size in y direction
D = 1.0 # diffusion coefficient
dt = 0.0001 # time step size
time_steps = 5000 # number of time steps to take

#Create the 2D periodic grid
mesh = PeriodicGrid2D(nx=nx, ny=ny, dx=dx, dy=dy)

#Define the target variable with some initial value
C = CellVariable(name="concentration", mesh=mesh)

#Define a random noise initial conditions
noise = UniformNoiseVariable(mesh=mesh,minimum=0.,maximum=1.)
C.setValue(noise)

#Define the PDE
eq = (TransientTerm() == DiffusionTerm(coeff=D))

#Solve the PDE
for i in tqdm(range(time_steps)):
    eq.solve(var=C, dt=dt)

I am using Ubuntu 22.04 and a conda python 3.10 environment.

[guyer]

What version of FiPy are you running?

python -c "import fipy; print(fipy.__version__)"

This may be an issue that was reported (and fixed) a couple of months ago: #896

[aritra-mukhopadhyay]

I am using FiPy version 3.4.3. Interestingly, I found that the problem did not arise while running it in Google Colab. In Colab, I installed FiPy via pip, whereas in my machine I am using a conda installation.

[aritra-mukhopadhyay]

I am attaching the package list of my conda environment below if it would be helpful.
environment_packages.txt

[guyer]

Thank you for forwarding your environment. We're delinquent in doing a release since fixing #896. Please try a git checkout of FiPy.

[aritra-mukhopadhyay]

Thanks a lot! I can confirm that the issue was fixed by using the latest Git version of Fipy (3.4.3+5.g4db188c1), but only together with the line PETSc.garbage_cleanup() inside the for loop (after importing: from petsc4py import PETSc) as suggested by you in #896 (comment).