test larger cpu runner (#1170)

* test larger cpu runner

* update omsf runner version

* nvidia-smi throws an error if there is no GPU

* Try a large

* just run slow tets

* try xlarge (should fail but thats okay)

* lets see if the intel flavor is any faster

* see if this changes the runtime at all

* go back to working runner instance type

* bump size of disk

* check time on a t3 med

* try xlarge

* try t3a.xlarge

* now t3a.2xl

* now t3.2xl

* use platforms that actually will work

* need empty set not dict

* pin buggy mypy

* fix mypy pin

* see if large works

* check t3.2xlarge

* make sure we load the marks correctly

* I think we are using the wrong instance family

* don't run itergration tests on the CPU

Author: mikemhenry

.github/workflows/cpu-long-tests.yaml

@@ -21,8 +21,9 @@ jobs:
         uses: omsf/start-aws-gha-runner@v1.0.0
         with:
           aws_image_id: ami-0b7f661c228e6a4bb
-          aws_instance_type: t3a.small  # TODO try t3a.medium next
+          aws_instance_type: c7i.xlarge
           aws_home_dir: /home/ubuntu
+          aws_root_device_size: 125 
         env:
           GH_PAT: ${{ secrets.GH_PAT }}
 
@@ -46,9 +47,6 @@ jobs:
       - name: Print Docker details
         run: "docker version || true"
 
-      - name: Check for nvidia-smi
-        run: "nvidia-smi"
-
       - name: "Setup Micromamba"
         uses: mamba-org/setup-micromamba@v2
         with:
@@ -88,12 +86,11 @@ jobs:
 
       - name: "Run tests"
         env:
-          # Set the OFE_SLOW_TESTS to True if running a Cron job
           OFE_SLOW_TESTS: "true"
           DUECREDIT_ENABLE: 'yes'
           OFE_INTEGRATION_TESTS: FALSE
         run: |
-          pytest -n logical -vv --durations=10
+          pytest -n logical -vv --durations=10 openfecli/tests/ openfe/tests/
 
   stop-aws-runner:
     runs-on: ubuntu-latest

.github/workflows/mypy.yaml

@@ -43,7 +43,7 @@ jobs:
       - name: "Install steps"
         run: |
           python -m pip install --no-deps git+https://github.com/OpenFreeEnergy/gufe@main
-          python -m pip install mypy
+          python -m pip install "mypy<1.16.0"
           python -m pip install types-setuptools
           python -m pip install --no-deps -e .
 

openfe/tests/protocols/conftest.py

@@ -223,7 +223,7 @@ def afe_solv_transformation_json() -> str:
     """
     d = resources.files('openfe.tests.data.openmm_afe')
     fname = "AHFEProtocol_json_results.gz"
-    
+
     with gzip.open((d / fname).as_posix(), 'r') as f:  # type: ignore
         return f.read().decode()  # type: ignore
 
@@ -240,3 +240,31 @@ def md_json() -> str:
 
     with gzip.open((d / fname).as_posix(), 'r') as f:  # type: ignore
         return f.read().decode()  # type: ignore
+
+@pytest.fixture
+def get_available_openmm_platforms() -> set[str]:
+    """
+    OpenMM Platforms that are available and functional on system
+    """
+    import openmm
+    from openmm import Platform
+    # Get platforms that openmm was built with
+    platforms = {Platform.getPlatform(i).getName() for i in range(Platform.getNumPlatforms())}
+
+    # Now check if we can actually use the platforms
+    working_platforms = set()
+    for platform in platforms:
+        system = openmm.System()
+        system.addParticle(1.0)
+        integrator = openmm.VerletIntegrator(0.001)
+        try:
+            context = openmm.Context(system, integrator, Platform.getPlatformByName(platform))
+            working_platforms.add(platform)
+            del context
+        except openmm.OpenMMException:
+            continue
+        finally:
+            del system, integrator
+
+
+    return working_platforms

openfe/tests/protocols/openmm_ahfe/test_ahfe_slow.py

@@ -15,11 +15,11 @@
 @pytest.mark.parametrize('platform', ['CPU', 'CUDA'])
 def test_openmm_run_engine(
     platform,
-    available_platforms,
+    get_available_openmm_platforms,
     benzene_modifications,
     tmpdir
 ):
-    if platform not in available_platforms:
+    if platform not in get_available_openmm_platforms:
         pytest.skip(f"OpenMM Platform: {platform} not available")
 
     # Run a really short calculation to check everything is going well

openfe/tests/protocols/openmm_rfe/test_hybrid_top_slow.py

@@ -17,11 +17,11 @@
 def test_openmm_run_engine(
     benzene_vacuum_system,
     platform,
-    available_platforms,
+    get_available_openmm_platforms,
     benzene_modifications,
     tmpdir
 ):
-    if platform not in available_platforms:
+    if platform not in get_available_openmm_platforms:
         pytest.skip(f"OpenMM Platform: {platform} not available")
     # this test actually runs MD
     # these settings are a small self to self sim, that has enough eq that