Skip to content

Commit

Permalink
mpi4py: run the spawn and dynamic process tests
Browse files Browse the repository at this point in the history 
Split the mpi4py Github Action into 4 parts:

1. build: do everything to build, configure, and install Open MPI and
   mpi4py
2. run: run all the mpi4py tests with its defaults.  As of March 2024,
   this disables the spawn and dynamic tests, which means that the
   entire block of tests should pass.
3. run_spawn: run all the mpi4py tests, including the spawn tests.  As
   of March 2024, we know some of these tests are failing.
4. run_dynamic: run all the mpi4py tests, including the dynamic tests.
   As of March 2024, we know some of these tests are failing.

The spawn and dynamic failures are different, so we split them up and
run them separately.

Signed-off-by: Jeff Squyres <jeff@squyres.com>
  • Loading branch information
@jsquyres
jsquyres committed Mar 21, 2024
1 parent f88283d commit 0ad934c
Showing 1 changed file with 224 additions and 5 deletions.
229 changes: 224 additions & 5 deletions .github/workflows/ompi_mpi4py.yaml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3,12 +3,10 @@ name: mpi4py
on: [pull_request]

jobs:
mpi4py:
build:
runs-on: ubuntu-latest
timeout-minutes: 60

timeout-minutes: 30
steps:

- name: Configure hostname
run: echo 127.0.0.1 `hostname` | sudo tee -a /etc/hosts > /dev/null
if: ${{ runner.os == 'Linux' || runner.os == 'macOS' }}
Expand Down Expand Up @@ -36,7 +34,8 @@ jobs:
--disable-sphinx
--disable-mpi-fortran
--disable-oshmem
LDFLAGS=-Wl,-rpath,/usr/local/lib
--prefix=/opt/openmpi
LDFLAGS=-Wl,-rpath,/opt/openmpi/lib
working-directory: mpi-build

- name: Build MPI
Expand All @@ -47,6 +46,9 @@ jobs:
run: sudo make install
working-directory: mpi-build

- name: Add Open MPI to PATH
run: echo /opt/openmpi/bin >> $GITHUB_PATH

- name: Tweak MPI
run: |
# Tweak MPI
Expand All @@ -65,6 +67,18 @@ jobs:
- name: Show MPICC
run: mpicc -show

- name: Save the Open MPI installation for other jobs
uses: actions/upload-artifact@v4
with:
# I dislike hard-coding /home/runner, but I couldn't get $HOME
# or ${{ env.HOME }} to expand properly.
path: |
/opt/openmpi
/home/runner/.openmpi
/home/runner/.prte
name: ompi-build
retention-days: 2

- name: Use Python
uses: actions/setup-python@v5
with:
Expand All @@ -89,6 +103,56 @@ jobs:
env:
CFLAGS: "-O0"

- name: Find where python's site-packages directory is located
run: python3 -c 'import site; print(f"pysite={site.getsitepackages()[0]}")' >> $GITHUB_ENV

- name: Save the mpi4py installation for other jobs
uses: actions/upload-artifact@v4
with:
# mpi4py was checked out in the same directory as OMPI.
# But we only need the "test" and "demo" directories from mpi4py.
path: |
test
demo
${{ env.pysite }}/mpi4py
name: mpi4py-build
retention-days: 2

#==============================================

run:
# This whole set of tests run with mpi4py's defaults. As of March
# 2024, this means disabling the spawn and dynamic tests. We want
# this block of tests to pass.
needs: [build]
runs-on: ubuntu-latest
timeout-minutes: 30
steps:
- name: Get OMPI build
uses: actions/download-artifact@v4
with:
name: ompi-build
path: /
- name: Restore executable permissions
run: chmod a+x /opt/openmpi/bin/*
- name: Add Open MPI to PATH
run: echo /opt/openmpi/bin >> $GITHUB_PATH

- name: Use Python
uses: actions/setup-python@v5
with:
python-version: 3
architecture: x64
- name: Find where python's site-packages directory is located
run: python3 -c 'import site; print(f"pysite={site.getsitepackages()[0]}")' >> $GITHUB_ENV
- name: Get mpi4py build
uses: actions/download-artifact@v4
with:
name: mpi4py-build
path: /

#----------------------------------------------

- name: Test mpi4py (singleton)
run: python test/main.py -v
if: ${{ true }}
Expand Down Expand Up @@ -118,3 +182,158 @@ jobs:
run: python demo/test-run/test_run.py -v
if: ${{ true }}
timeout-minutes: 10

#==============================================

run_spawn:
# This whole set of tests runs explicitly with setting "enable the
# spawn tests". As of March 2024, we know that Open MPI is
# failing these tests.
#
# Note that we state here that we need the build *and* run jobs,
# because when a job fails, it will kill any concurrently-running
# jobs. Since we want the "run" job to complete, we make it a
# dependency of this job.
needs: [build, run]
runs-on: ubuntu-latest
timeout-minutes: 30
steps:
- name: Get OMPI build
uses: actions/download-artifact@v4
with:
name: ompi-build
path: /
- name: Restore executable permissions
run: chmod a+x /opt/openmpi/bin/*
- name: Add Open MPI to PATH
run: echo /opt/openmpi/bin >> $GITHUB_PATH

- name: Use Python
uses: actions/setup-python@v5
with:
python-version: 3
architecture: x64
- name: Find where python's site-packages directory is located
run: python3 -c 'import site; print(f"pysite={site.getsitepackages()[0]}")' >> $GITHUB_ENV
- name: Get mpi4py build
uses: actions/download-artifact@v4
with:
name: mpi4py-build
path: /

#----------------------------------------------

- name: Test mpi4py (singleton)
run: python test/main.py -v
if: ${{ true }}
env:
MPI4PY_TEST_SPAWN: "1"
timeout-minutes: 10
- name: Test mpi4py (np=1)
run: mpiexec -n 1 python test/main.py -v
env:
MPI4PY_TEST_SPAWN: "1"
if: ${{ true }}
timeout-minutes: 10
- name: Test mpi4py (np=2)
run: mpiexec -n 2 python test/main.py -v -f
env:
MPI4PY_TEST_SPAWN: "1"
if: ${{ true }}
timeout-minutes: 10
- name: Test mpi4py (np=3)
run: mpiexec -n 3 python test/main.py -v -f
env:
MPI4PY_TEST_SPAWN: "1"
if: ${{ true }}
timeout-minutes: 10
- name: Test mpi4py (np=4)
run: mpiexec -n 4 python test/main.py -v -f
env:
MPI4PY_TEST_SPAWN: "1"
if: ${{ true }}
timeout-minutes: 10
- name: Test mpi4py (np=5)
run: mpiexec -n 5 python test/main.py -v -f
env:
MPI4PY_TEST_SPAWN: "1"
if: ${{ true }}
timeout-minutes: 10

#==============================================

run_dynamic:
# This whole set of tests runs explicitly with setting "enable the
# dynamic tests". As of March 2024, we know that Open MPI is
# failing these tests.
#
# Note that we state here that we need the build *and* run jobs,
# because when a job fails, it will kill any concurrently-running
# jobs. Since we want the "run" job to complete, we make it a
# dependency of this job.
needs: [build, run]
runs-on: ubuntu-latest
timeout-minutes: 30
steps:
- name: Get OMPI build
uses: actions/download-artifact@v4
with:
name: ompi-build
path: /
- name: Restore executable permissions
run: chmod a+x /opt/openmpi/bin/*
- name: Add Open MPI to PATH
run: echo /opt/openmpi/bin >> $GITHUB_PATH

- name: Use Python
uses: actions/setup-python@v5
with:
python-version: 3
architecture: x64
- name: Find where python's site-packages directory is located
run: python3 -c 'import site; print(f"pysite={site.getsitepackages()[0]}")' >> $GITHUB_ENV
- name: Get mpi4py build
uses: actions/download-artifact@v4
with:
name: mpi4py-build
path: /

#----------------------------------------------

- name: Test mpi4py (singleton)
run: python test/main.py -v
if: ${{ true }}
env:
MPI4PY_TEST_DYNPROC: "1"
timeout-minutes: 10
- name: Test mpi4py (np=1)
run: mpiexec -n 1 python test/main.py -v
env:
MPI4PY_TEST_DYNPROC: "1"
if: ${{ true }}
timeout-minutes: 10
- name: Test mpi4py (np=2)
run: mpiexec -n 2 python test/main.py -v -f
env:
MPI4PY_TEST_DYNPROC: "1"
if: ${{ true }}
timeout-minutes: 10
- name: Test mpi4py (np=3)
run: mpiexec -n 3 python test/main.py -v -f
env:
MPI4PY_TEST_DYNPROC: "1"
if: ${{ true }}
timeout-minutes: 10
- name: Test mpi4py (np=4)
run: mpiexec -n 4 python test/main.py -v -f
env:
MPI4PY_TEST_DYNPROC: "1"
if: ${{ true }}
timeout-minutes: 10
- name: Test mpi4py (np=5)
run: mpiexec -n 5 python test/main.py -v -f
env:
MPI4PY_TEST_DYNPROC: "1"
if: ${{ true }}
timeout-minutes: 10

0 comments on commit 0ad934c

Please sign in to comment.