Create a pool of MAPDL instances

PyMAPDL contains the MapdlPool class to simplify creating multiple local instances of the Mapdl class for batch processing. This can be used for the batch processing of a set of input files, convergence analysis, or other batch related processes.

This code creates a pool:

>>> from ansys.mapdl.core import MapdlPool
>>> pool = MapdlPool(10)
'MAPDL Pool with 10 active instances'
>>> pool.exit(block=True)

You can supply additional keyword arguments when creating the pool. This code creates several instances with one CPU each running at the current directory within their own isolated directories:

>>> import os
>>> my_path = os.getcmd()
>>> pool = MapdlPool(10, nproc=1, run_location=my_path)
Creating Pool: 100%|########| 10/10 [00:01<00:00,  1.43it/s]

Additionally, you can group already running MAPDL instances into an MapdlPool instance by specifying their ports when creating the pool.

>>> from ansys.mapdl.core import MapdlPool
>>> pool = MapdlPool(port=[50082, 50083, 50084, 50085, 50086])
'MAPDL Pool with 5 active instances'
>>> pool.exit(block=True)

You can also specify a list of IP addresses to connect to:

>>> pool = MapdlPool(ip=["127.0.0.2", "127.0.0.3", "127.0.0.4"])
Creating Pool: 100%|########| 3/3 [00:01<00:00,  1.43it/s]

You can access each individual MAPDL instance with this code:

>>> pool[0]
<ansys.mapdl.core.mapdl.MapdlGrpc object at 0x7f66270cc8d0>

Note that this is a self-healing pool. If an instance of MAPDL dies during a batch process, that instance is automatically restarted. You can turn off this behavior by setting restart_failed=False when creating the pool.

Run a set of input files

You can use the pool to run a set of pre-generated input files using the MapdlPool.run_batch method. For example, this code would run the first set of 20 verification files:

>>> from ansys.mapdl.core import examples
>>> files = [examples.vmfiles["vm%d" % i] for i in range(1, 21)]
>>> outputs = pool.run_batch(files)
>>> len(outputs)
20

Run an user function

You can use the pool to run a custom user function on each MAPDL instance over a set of inputs. As in the example for the MapdlPool.run_batch method, the following code uses a set of verification files. However, it implements it as a function and outputs the final routine instead of the text output from MAPDL.

completed_indices = []


def func(mapdl, input_file, index):
    # input_file, index = args
    mapdl.clear()
    output = mapdl.input(input_file)
    completed_indices.append(index)
    return mapdl.parameters.routine


inputs = [(examples.vmfiles["vm%d" % i], i) for i in range(1, 10)]
output = pool.map(func, inputs, progress_bar=True, wait=True)
[
    "Begin level",
    "Begin level",
    "Begin level",
    "Begin level",
    "Begin level",
    "Begin level",
    "Begin level",
    "Begin level",
    "Begin level",
]

# Close the PyMAPDL pool.
pool.exit()

Using next available instances

When working with many multiple instances, it might be more convenient to use the MapdlPool class within a context manager. This can be accomplished using the MapdlPool.next() method as follows:

with pool.next() as mapdl:
    mapdl.prep7()
    ...

This context manager makes sure to set the instance as busy or locked while code is executing the block. Once the execution exits the context manager, the instance is set free or unlocked. This context manager is particularly interesting when using it with threads.

from ansys.mapdl.core import MapdlPool
from threading import Thread

loads = [1e6, 2e6, 3e6]
solutions = {each_load: None for each_load in loads}

pool = MapdlPool(2)


def calculate_model(mapdl, load):
    mapdl.prep7()
    mapdl.et(1, "SOLID5")
    mapdl.block(0, 10, 0, 20, 0, 30)
    mapdl.esize(10)
    mapdl.vmesh("ALL")
    mapdl.units("SI")  # SI - International system (m, kg, s, K).

    # Define a material (nominal steel in SI)
    mapdl.mp("EX", 1, 210e9)  # Elastic moduli in Pa (kg/(m*s**2))
    mapdl.mp("DENS", 1, 7800)  # Density in kg/m3
    mapdl.mp("PRXY", 1, 0.3)  # Poisson's Ratio

    # Fix the left-hand side.
    mapdl.nsel("S", "LOC", "Z", 0)
    mapdl.d("ALL", "UX")
    mapdl.d("ALL", "UY")
    mapdl.d("ALL", "UZ")

    mapdl.nsel("S", "LOC", "Z", 30)
    mapdl.f("ALL", "FX", load)

    mapdl.allsel()
    mapdl.solu()
    mapdl.antype("STATIC")
    mapdl.solve()
    mapdl.finish()

    # Get maximum displacement in the X direction on the top surface.
    mapdl.nsel("S", "LOC", "Z", 30)
    solutions[load] = mapdl.post_processing.nodal_displacement("X").max()


def threaded_function(load):
    with pool.next() as mapdl:
        value = calculate_model(mapdl, load)


if __name__ == "__main__":
    threads = []
    for load in loads:
        t = Thread(target=threaded_function, args=[load])
        t.start()
        threads.append(t)

    for thread in threads:
        thread.join()

    for k, v in solutions.items():
        print(f"Load: {k:5.2f}\tDisplacement: {v:8.6f}")

You can also use the MapdlPool.next_available() method to obtain an available Mapdl instance, but in that case, you must manage the lock with the Mapdl.locked method.

pool = MapdlPool(4)

mapdl, i = pool.next_available(return_index=True)

mapdl.locked = True

mapdl.prep7()
# More code...
# ...
#
mapdl.locked = False  # Important for the instance to be seen as available.

Close the PyMAPDL pool

You can close the PyMAPDL pool with the MapdlPool.exit() command.

>>> pool.exit()

API description

For a comprehensive description, see MAPDL pool.