Fenicsx- saving state file to enable restart from the last saved time step in parallel

General
1

Hi,
I have been working with fenicsx 0.6.0. I am trying to save state variables every 100 timesteps, which helps me restarting the simulation from last saved timestep when my simulation crashes - checkpointing in other terms I guess. I am able to do that in serial code using these functions:

def save_state(step, u, un, filename="./stateBackup/simulation_state.npz"):
    # Convert to numpy arrays
    u_array = u.vector[:]
    un_array = un.vector[:]
    # Save using np.savez
    np.savez(filename, step=step, u=u_array, un=un_array)

def load_state(u, un, filename="./stateBackup/simulation_state.npz"):
    # Load the data
    with np.load(filename, allow_pickle=False) as data:
        step = data['step']
        u.vector.setArray(data['u'])
        un.vector.setArray(data['un'])
    return step

                               But while working in parallel, I am facing issues of dof mismatch after restarting with the saved state file. Any leads on how to work with this in parallel environment would be super helpful.                    Thanks in advance.
2

Achieving n-m checkpointing is actually quite challenging. Thankfully @dokken is kind enough to share his excellent work via GitHub - jorgensd/adios4dolfinx: Extending DOLFINx with checkpointing functionality.

This uses adios2 in the backend to store all the information necessary to save/load data from arbitrary partitions into new partitions. Ideal for checkpointing in parallel computations.

3

Thanks Nate!
I came across this earlier, but the issue is adios4dolfinx is compatible with fenicsx v0.7 or higher, and my script is in fenicsx v0.6. Which is why I was looking if there is any workaround for this, before I start writing my script to work with higher versions of fenicsx.

4

There is a 0.6.0 release: Release v0.6.0 · jorgensd/adios4dolfinx · GitHub

However, I would strongly advice to upgrade your installation, as there are many bugs and performance fixes and new features since 2023