Skip to content

Argonne Leadership Computing Facility

Parsl on Polaris

Parsl is a flexible and scalable parallel programming library for Python.

-- Parsl Documentation

For many applications, managing an ensemble of jobs into a workflow is a critical step that can easily become a performance bottleneck. Many tools exist to address this, of which parsl is just one. On this page, we'll highlight some of the key pieces of information about parsl that are relevant to Polaris. Parsl is also extensively documented, has a dedicated Slack Channel, and a large community of users and developers beyond ALCF. We encourage you to engage with the parsl community for support with parsl specific questions, and for Polaris-specific questions or problems, please contact

Getting Parsl on Polaris

You can install parsl building off of the conda modules. You have some flexibility in how you want to extend the conda module to include parsl, but here is an example way to do it:

# Load the Conda Module (needed everytime you use parsl)
module load conda
conda activate

# Create a virtual env that uses the conda env as the system packages.
# Only do the next line on initial set up:
python -m venv --system-site-packages /path/to/your/virtualenv

# Load the virtual env (every time):
source /path/to/your/virtualenv/bin/activate

# Install parsl (only once)
pip install parsl

Using Parsl on Polaris

Parsl has a variety of possible configuration settings. As an example, we provide the configuration below that will run one task per GPU:

from parsl.config import Config

# PBSPro is the right provider for Polaris:
from parsl.providers import PBSProProvider
# The high throughput executor is for scaling to HPC systems:
from parsl.executors import HighThroughputExecutor
# You can use the MPI launcher, but may want the Gnu Parallel launcher, see below
from parsl.launchers import MpiExecLauncher, GnuParallelLauncher
# address_by_interface is needed for the HighThroughputExecutor:
from parsl.addresses import address_by_interface
# For checkpointing:
from parsl.utils import get_all_checkpoints

# Adjust your user-specific options here:

user_opts = {
    "worker_init":      f"source /path/to/your/virtualenv/bin/activate; cd {run_dir}", # load the environment where parsl is installed
    "scheduler_options":"#PBS -l filesystems=home:eagle:grand" , # specify any PBS options here, like filesystems
    "account":          "YOURPROJECT",
    "queue":            "debug-scaling",
    "walltime":         "1:00:00",
    "nodes_per_block":  3, # think of a block as one job on polaris, so to run on the main queues, set this >= 10
    "cpus_per_node":    32, # Up to 64 with multithreading
    "available_accelerators": 4, # Each Polaris node has 4 GPUs, setting this ensures one worker per GPU
    "cores_per_worker": 8, # this will set the number of cpu hardware threads per worker.  

checkpoints = get_all_checkpoints(run_dir)
print("Found the following checkpoints: ", checkpoints)

config = Config(
                available_accelerators=user_opts["available_accelerators"], # if this is set, it will override other settings for max_workers if set
                start_method="spawn",  # Needed to avoid interactions between MPI and os.fork
                    launcher=MpiExecLauncher(bind_cmd="--cpu-bind", overrides="--depth=64 --ppn 1"),
                    # Which launcher to use?  Check out the note below for some details.  Try MPI first!
                    # launcher=GnuParallelLauncher(),
                    # PBS directives (header lines): for array jobs pass '-J' option
                    # Command to be run before starting a worker, such as:
                    # number of compute nodes allocated for each block
                    max_blocks=1, # Can increase more to have more parallel jobs
        checkpoint_files = checkpoints,
        checkpoint_mode = 'task_exit',

Special notes for Polaris

On Polaris, there is a known bug where python applications launched with mpi and that use fork to spawn processes can sometimes have unexplaned hangs. For this reason, it is recommended to use start_method="spawn" on Polaris when using the MpiExecLauncher as is shown in the example config above. Alternatively, another solution is to use the GNUParallelLauncher which uses GNU Parallel to spawn processes. GNU Parallel can be loaded in your environment with the command module load gnu-parallel. Both of these approaches will circumvent the hang issue from using fork.


For parsl versions after July 2023, the address passed in the HighThroughputExecutor needs to be set to address = address_by_interface("bond0"). With parsl versions prior to July 2023, it was recommended to use address = address_by_hostname() on Polaris, but with later versions this will not work on Polaris (or any other machine).