Support for GlobusComputeExecutor

.github/workflows/gce_test.yaml

@@ -0,0 +1,97 @@
+name: GlobusComputeExecutor tests
+
+on:
+  pull_request:
+    types:
+      - opened
+      - synchronize
+
+jobs:
+  main-test-suite:
+    runs-on: ubuntu-20.04
+    timeout-minutes: 60
+
+    steps:
+    - uses: actions/checkout@master
+
+    - name: Set up Python 3.11
+      uses: actions/setup-python@v4
+      with:
+        python-version: 3.11
+
+    - name: Collect Job Information
+      id: job-info
+      run: |
+        echo "Python Version: 3.11" >> ci_job_info.txt
+        echo "CI Triggering Event: ${{ github.event_name }}" >> ci_job_info.txt
+        echo "Triggering Git Ref: ${{ github.ref }}" >> ci_job_info.txt
+        echo "Triggering Git SHA: ${{ github.sha }}" >> ci_job_info.txt
+        echo "Workflow Run: ${{ github.run_number }}" >> ci_job_info.txt
+        echo "Workflow Attempt: ${{ github.run_attempt }}" >> ci_job_info.txt
+        as_ascii="$(echo "${{ github.ref_name }}" | perl -pe "s/[^A-z0-9-]+/-/g; s/^-+|-+\$//g; s/--+/-/g;")"
+        echo "as-ascii=$as_ascii" >> $GITHUB_OUTPUT
+
+    - name: setup virtual env
+      run: |
+        make virtualenv
+        source .venv/bin/activate
+
+    - name: Non-requirements based install
+      run: |
+        # mpich: required by mpi4py which is in test-requirements for radical-pilot
+        sudo apt-get update -q
+        sudo apt-get install -qy mpich
+
+    - name: make deps clean_coverage
+      run: |
+        source .venv/bin/activate
+        make deps
+        make clean_coverage
+
+        # Temporary fix until fixes make it to a release
+        git clone -b main https://github.com/globus/globus-compute.git
+        pip3 install globus-compute/compute_sdk globus-compute/compute_endpoint
+
+    - name: start globus_compute_endpoint
+      env:
+        GLOBUS_COMPUTE_CLIENT_ID: ${{ secrets.GLOBUS_COMPUTE_CLIENT_ID }}
+        GLOBUS_COMPUTE_CLIENT_SECRET: ${{ secrets.GLOBUS_COMPUTE_SECRET_KEY }}
+      run: |
+        source /home/runner/work/parsl/parsl/.venv/bin/activate
+        globus-compute-endpoint configure default
+        which globus-compute-endpoint
+        python3 -c "import globus_compute_sdk; print(globus_compute_sdk.__version__)"
+        python3 -c "import globus_compute_endpoint; print(globus_compute_endpoint.__version__)"
+        cat << EOF > /home/runner/.globus_compute/default/config.yaml
+        engine:
+          type: ThreadPoolEngine
+          max_workers: 4
+        EOF
+        cat /home/runner/.globus_compute/default/config.yaml
+        mkdir ~/.globus_compute/default/tasks_working_dir
+        globus-compute-endpoint start default
+        globus-compute-endpoint list
+    - name: make test
+      env:
+        GLOBUS_COMPUTE_CLIENT_ID: ${{ secrets.GLOBUS_COMPUTE_CLIENT_ID }}
+        GLOBUS_COMPUTE_CLIENT_SECRET: ${{ secrets.GLOBUS_COMPUTE_SECRET_KEY }}
+      run: |
+        source .venv/bin/activate
+        export GLOBUS_COMPUTE_ENDPOINT=$(globus-compute-endpoint list | grep default | cut -c 3-38) 
+        echo "GLOBUS_COMPUTE_ENDPOINT = $GLOBUS_COMPUTE_ENDPOINT"
+
+        export PARSL_TEST_PRESERVE_NUM_RUNS=7
+
+        make gce_test
+        ln -s pytest-parsl/parsltest-current test_runinfo
+
+    - name: Archive runinfo logs
+      if: ${{ always() }}
+      uses: actions/upload-artifact@v4
+      with:
+        name: runinfo-3.11-${{ steps.job-info.outputs.as-ascii }}-${{ github.sha }}
+        path: |
+          runinfo/
+          pytest-parsl/
+          ci_job_info.txt
+        compression-level: 9

Makefile

@@ -51,6 +51,10 @@ clean_coverage:
 mypy: ## run mypy checks
 	MYPYPATH=$(CWD)/mypy-stubs mypy parsl/
 
+.PHONY: gce_test
+gce_test: ## Run tests with GlobusComputeExecutor
+	pytest -v -k "not shared_fs and not issue_3620 and not staging_required" --config parsl/tests/configs/globus_compute.py parsl/tests/ --random-order --durations 10
+
 .PHONY: local_thread_test
 local_thread_test: ## run all tests with local_thread config
 	pytest parsl/tests/ -k "not cleannet" --config parsl/tests/configs/local_threads.py --random-order --durations 10

docs/reference.rst

@@ -77,6 +77,7 @@ Executors
     parsl.executors.taskvine.TaskVineExecutor
     parsl.executors.FluxExecutor
     parsl.executors.radical.RadicalPilotExecutor
+    parsl.executors.globus_compute.GlobusComputeExecutor
 
 Manager Selectors
 =================

docs/userguide/execution.rst

@@ -87,6 +87,9 @@ Parsl currently supports the following executors:
 4. `parsl.executors.taskvine.TaskVineExecutor`: This executor uses `TaskVine <https://ccl.cse.nd.edu/software/taskvine/>`_ as the execution backend. TaskVine scales up to tens of thousands of cores and actively uses local storage on compute nodes to offer a diverse array of performance-oriented features, including: smart caching and sharing common large files between tasks and compute nodes, reliable execution of tasks, dynamic resource sizing, automatic Python environment detection and sharing.
 These executors cover a broad range of execution requirements. As with other Parsl components, there is a standard interface (ParslExecutor) that can be implemented to add support for other executors.
 
+5. `parsl.executors.globus_compute.GlobusComputeExecutor`: This executor uses `Globus Compute <https://globus-compute.readthedocs.io/en/latest/index.html>`_
+as the execution backend to run tasks on remote systems.
+
 .. note::
    Refer to :ref:`configuration-section` for information on how to configure these executors.
 

mypy.ini

@@ -177,6 +177,9 @@ ignore_missing_imports = True
 #[mypy-multiprocessing.synchronization.*]
 #ignore_missing_imports = True
 
+[mypy-globus_compute_sdk.*]
+ignore_missing_imports = True
+
 [mypy-pandas.*]
 ignore_missing_imports = True
 

parsl/executors/__init__.py

@@ -1,4 +1,5 @@
 from parsl.executors.flux.executor import FluxExecutor
+from parsl.executors.globus_compute import GlobusComputeExecutor
 from parsl.executors.high_throughput.executor import HighThroughputExecutor
 from parsl.executors.high_throughput.mpi_executor import MPIExecutor
 from parsl.executors.threads import ThreadPoolExecutor
@@ -8,4 +9,5 @@
            'HighThroughputExecutor',
            'MPIExecutor',
            'WorkQueueExecutor',
-           'FluxExecutor']
+           'FluxExecutor',
+           'GlobusComputeExecutor']

parsl/executors/globus_compute.py

@@ -0,0 +1,171 @@
+from __future__ import annotations
+
+import copy
+import uuid
+from concurrent.futures import Future
+from typing import Any, Callable, Dict, Optional, Union
+
+import typeguard
+
+from parsl.errors import OptionalModuleMissing
+from parsl.executors.base import ParslExecutor
+from parsl.utils import RepresentationMixin
+
+try:
+    from globus_compute_sdk import Client, Executor
+    _globus_compute_enabled = True
+except ImportError:
+    _globus_compute_enabled = False
+
+UUID_LIKE_T = Union[uuid.UUID, str]
+
+
+class GlobusComputeExecutor(ParslExecutor, RepresentationMixin):
+    """ GlobusComputeExecutor enables remote execution on Globus Compute endpoints
+
+    GlobusComputeExecutor is a thin wrapper over globus_compute_sdk.Executor
+    Refer to `globus-compute user documentation <https://globus-compute.readthedocs.io/en/latest/executor.html>`_
+    and `reference documentation <https://globus-compute.readthedocs.io/en/latest/reference/executor.html>`_
+    for more details.
+
+    .. note::
+       As a remote execution system, Globus Compute relies on serialization to ship
+       tasks and results between the Parsl client side and the remote Globus Compute
+       Endpoint side. Serialization is unreliable across python versions, and
+       wrappers used by Parsl assume identical Parsl versions across on both sides.
+       We recommend using matching Python, Parsl and Globus Compute version on both
+       the client side and the endpoint side for stable behavior.
+
+    """
+
+    @typeguard.typechecked
+    def __init__(
+            self,
+            endpoint_id: UUID_LIKE_T,
+            task_group_id: Optional[UUID_LIKE_T] = None,
+            resource_specification: Optional[Dict[str, Any]] = None,
+            user_endpoint_config: Optional[Dict[str, Any]] = None,
+            label: str = "GlobusComputeExecutor",
+            batch_size: int = 128,
+            amqp_port: Optional[int] = None,
+            client: Optional[Client] = None,
+            **kwargs,
+    ):
+        """
+        Parameters
+        ----------
+
+        endpoint_id:
+            id of the endpoint to which to submit tasks
+
+        task_group_id:
+            The Task Group to which to associate tasks.  If not set,
+            one will be instantiated.
+
+        resource_specification:
+            Specify resource requirements for individual task execution.
+
+        user_endpoint_config:
+            User endpoint configuration values as described
+            and allowed by endpoint administrators. Must be a JSON-serializable dict
+            or None. Refer docs from `globus-compute
+            <https://globus-compute.readthedocs.io/en/latest/endpoints/endpoints.html#templating-endpoint-configuration>`_
+            for more info.
+
+        label:
+            a label to name the executor
+
+        batch_size:
+            the maximum number of tasks to coalesce before
+            sending upstream [min: 1, default: 128]
+
+        amqp_port:
+            Port to use when connecting to results queue. Note that the
+            Compute web services only support 5671, 5672, and 443.
+
+        client:
+            instance of globus_compute_sdk.Client to be used by the executor.
+            If not provided, the executor will instantiate one with default arguments.
+
+        kwargs:
+            Other kwargs listed will be passed through to globus_compute_sdk.Executor
+            as is. Refer to `globus-compute docs
+            <https://globus-compute.readthedocs.io/en/latest/reference/executor.html#globus-compute-executor>`_
+        """
+        super().__init__()
+        self.endpoint_id = endpoint_id
+        self.task_group_id = task_group_id
+        self.resource_specification = resource_specification
+        self.user_endpoint_config = user_endpoint_config
+        self.label = label
+        self.batch_size = batch_size
+        self.amqp_port = amqp_port
+        self.client = client
+        self.executor_kwargs = kwargs
+
+        if not _globus_compute_enabled:
+            raise OptionalModuleMissing(
+                ['globus-compute-sdk'],
+                "GlobusComputeExecutor requires globus-compute-sdk installed"
+            )
+
+    def start(self) -> None:
+        """ Start the Globus Compute Executor """
+
+        self._executor: Executor = Executor(
+            endpoint_id=self.endpoint_id,
+            task_group_id=self.task_group_id,
+            resource_specification=self.resource_specification,
+            user_endpoint_config=self.user_endpoint_config,
+            label=self.label,
+            batch_size=self.batch_size,
+            amqp_port=self.amqp_port,
+            client=self.client,
+            **self.executor_kwargs
+        )
+
+    def submit(self, func: Callable, resource_specification: Dict[str, Any], *args: Any, **kwargs: Any) -> Future:
+        """ Submit func to globus-compute
+
+
+        Parameters
+        ----------
+
+        func: Callable
+            Python function to execute remotely
+
+        resource_specification: Dict[str, Any]
+            Resource specification can be used specify MPI resources required by MPI applications on
+            Endpoints configured to use globus compute's MPIEngine. GCE also accepts *user_endpoint_config*
+            to configure endpoints when the endpoint is a `Multi-User Endpoint
+            <https://globus-compute.readthedocs.io/en/latest/endpoints/endpoints.html#templating-endpoint-configuration>`_
+
+        args:
+            Args to pass to the function
+
+        kwargs:
+            kwargs to pass to the function
+
+        Returns
+        -------
+
+        Future
+        """
+        res_spec = copy.deepcopy(resource_specification or self.resource_specification)
+        # Pop user_endpoint_config since it is illegal in resource_spec for globus_compute
+        if res_spec:
+            user_endpoint_config = res_spec.pop('user_endpoint_config', self.user_endpoint_config)
+        else:
+            user_endpoint_config = self.user_endpoint_config
+
+        self._executor.resource_specification = res_spec
+        self._executor.user_endpoint_config = user_endpoint_config
+        return self._executor.submit(func, *args, **kwargs)
+
+    def shutdown(self):
+        """Clean-up the resources associated with the Executor.
+
+        GCE.shutdown will cancel all futures that have not yet registered with
+        Globus Compute and will not wait for the launched futures to complete.
+        """
+        self._executor.shutdown(wait=False, cancel_futures=True)

parsl/tests/configs/globus_compute.py

@@ -0,0 +1,18 @@
+import os
+
+from parsl.config import Config
+from parsl.executors import GlobusComputeExecutor
+
+
+def fresh_config():
+
+    endpoint_id = os.environ["GLOBUS_COMPUTE_ENDPOINT"]
+
+    return Config(
+        executors=[
+            GlobusComputeExecutor(
+                label="globus_compute",
+                endpoint_id=endpoint_id
+            )
+        ]
+    )

parsl/tests/conftest.py

@@ -163,6 +163,10 @@ def pytest_configure(config):
         'markers',
         'shared_fs: Marks tests that require a shared_fs between the workers are the test client'
     )
+    config.addinivalue_line(
+        'markers',
+        'issue_3620: Marks tests that do not work correctly on GlobusComputeExecutor (ref: issue 3620)'
+    )
 
 
 @pytest.fixture(autouse=True, scope='session')

parsl/tests/test_error_handling/test_resource_spec.py

@@ -1,3 +1,5 @@
+import pytest
+
 import parsl
 from parsl.app.app import python_app
 from parsl.executors import WorkQueueExecutor
@@ -11,6 +13,7 @@ def double(x, parsl_resource_specification={}):
     return x * 2
 
 
+@pytest.mark.issue_3620
 def test_resource(n=2):
     executors = parsl.dfk().executors
     executor = None

setup.py

@@ -34,6 +34,7 @@
     'flux': ['pyyaml', 'cffi', 'jsonschema'],
     'proxystore': ['proxystore'],
     'radical-pilot': ['radical.pilot==1.60', 'radical.utils==1.60'],
+    'globus_compute': ['globus_compute_sdk>=2.27.1'],
     # Disabling psi-j since github direct links are not allowed by pypi
     # 'psij': ['psi-j-parsl@git+https://github.com/ExaWorks/psi-j-parsl']
 }

test-requirements.txt

@@ -10,6 +10,7 @@ types-mock
 types-python-dateutil
 types-requests
 mpi4py
+globus-compute-sdk>=2.27.1
 
 # sqlalchemy is needed for typechecking, so it's here
 # as well as at runtime for optional monitoring execution