Build macOS wheels and integration tests

This is pretty simple, mostly just adding macos-13 (x86) and macos-14
(arm64) runners to the matrix. We also need to explicitly disable MPS
for PyTorch to run correctly on the memory-limited GHA runners.

Signed-off-by: James Bornholt <[email protected]>

Author: jamesbornholt

.github/workflows/ci-integration-main.yml

@@ -2,7 +2,7 @@ name: Integration tests (Main)
 
 on:
   push:
-    branches: [ "main", "feature/*" ]
+    branches: [ "main", "feature/*", "workflow/*" ]
   merge_group:
     types: [ "checks_requested" ]
 

.github/workflows/python-integration.yml

@@ -19,11 +19,12 @@ env:
 jobs:
   integration-test:
     name: Integration tests
-    runs-on: ubuntu-22.04
+    runs-on: ${{ matrix.runner }}
     environment: ${{ inputs.environment }}
     strategy:
       fail-fast: false
       matrix:
+        runner: [ubuntu-22.04, macos-13]
         python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
         test-run:
           - name: "S3"
@@ -53,17 +54,13 @@ jobs:
           aws-region: ${{ vars.S3_REGION }}
 
       - name: Set up stable Rust
-        uses: actions-rs/toolchain@v1
-        with:
-          toolchain: stable
-          override: true
+        uses: dtolnay/rust-toolchain@stable
 
       - name: Restore Cargo cache
         id: restore-cargo-cache
         uses: actions/cache/restore@v3
         with:
           path: |
-            ~/.cargo/bin/
             ~/.cargo/registry/index/
             ~/.cargo/registry/cache/
             ~/.cargo/git/db/
@@ -114,7 +111,6 @@ jobs:
         if: inputs.environment != 'integration-tests'
         with:
           path: |
-            ~/.cargo/bin/
             ~/.cargo/registry/index/
             ~/.cargo/registry/cache/
             ~/.cargo/git/db/

.github/workflows/rust-checks.yml

@@ -32,7 +32,6 @@ jobs:
         with:
           toolchain: stable
           override: true
-          submodules: true
 
       - name: Run cargo deny
         uses: EmbarkStudios/cargo-deny-action@v1
@@ -69,8 +68,11 @@ jobs:
         run: cargo clippy --all-targets --all-features --manifest-path s3torchconnectorclient/Cargo.toml
 
   tests:
-    runs-on: ubuntu-22.04
+    runs-on: ${{ matrix.runner }}
     name: Rust tests
+    strategy:
+      matrix:
+        runner: [ubuntu-22.04, macos-13]
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -80,7 +82,6 @@ jobs:
         with:
           toolchain: stable
           override: true
-          submodules: true
 
       - name: Build Rust tests
         run: cargo test --no-default-features --no-run --manifest-path s3torchconnectorclient/Cargo.toml

.github/workflows/wheels.yml

@@ -19,24 +19,33 @@ jobs:
     uses: ./.github/workflows/generate_third_party_licenses.yml
 
   build_wheels:
-    name: Build wheels for ${{ matrix.build_prefix }} - ${{ matrix.builder.arch }}
+    name: Wheels for ${{ matrix.python }} - ${{ matrix.builder.kind }} - ${{ matrix.builder.arch }}
     runs-on: ${{ matrix.builder.runner }}
     needs: generate_third_party_licenses
     strategy:
       matrix:
-        build_prefix:
+        python:
           - cp38
           - cp39
           - cp310
           - cp311
           - cp312
         builder:
           - runner: codebuild-${{ vars.CODEBUILD_PROJECT_NAME }}-${{ github.run_id }}-${{ github.run_attempt }}-ubuntu-7.0-large
+            kind: manylinux
             arch: x86_64
           - runner: codebuild-${{ vars.CODEBUILD_PROJECT_NAME }}-${{ github.run_id }}-${{ github.run_attempt }}-arm-3.0-large
+            kind: manylinux
             arch: aarch64
 #          - runner: ubuntu-20.04
+#            kind: manylinux
 #            arch: x86_64
+          - runner: macos-13
+            kind: macosx
+            arch: x86_64
+          - runner: macos-14
+            kind: macosx
+            arch: arm64
     permissions:
       id-token: write
       contents: read
@@ -80,7 +89,7 @@ jobs:
           cibuildwheel
           "s3torchconnectorclient"
           --output-dir "wheelhouse"
-          --only "${{ matrix.build_prefix }}-manylinux_${{ matrix.builder.arch }}"
+          --only "${{ matrix.python }}-${{ matrix.builder.kind }}_${{ matrix.builder.arch }}"
           2>&1
         shell: bash
 

s3torchconnector/tst/e2e/test_e2e_s3_lightning_checkpoint.py

@@ -22,6 +22,9 @@
 from models.lightning_transformer import LightningTransformer, L
 
 
+LIGHTNING_ACCELERATOR = "cpu"
+
+
 def test_save_and_load_checkpoint(checkpoint_directory):
     tensor = torch.rand(3, 10, 10)
     s3_lightning_checkpoint = S3LightningCheckpoint(region=checkpoint_directory.region)
@@ -77,7 +80,7 @@ def test_load_trained_checkpoint(checkpoint_directory):
     dataset = WikiText2(data_dir=Path(f"/tmp/data/{nonce}"))
     dataloader = DataLoader(dataset, num_workers=3)
     model = LightningTransformer(vocab_size=dataset.vocab_size)
-    trainer = L.Trainer(fast_dev_run=2)
+    trainer = L.Trainer(accelerator=LIGHTNING_ACCELERATOR, fast_dev_run=2)
     trainer.fit(model=model, train_dataloaders=dataloader)
     checkpoint_name = "lightning_module_training_checkpoint.pt"
     s3_uri = f"{checkpoint_directory.s3_uri}{checkpoint_name}"
@@ -96,6 +99,7 @@ def test_compatibility_with_trainer_plugins(checkpoint_directory):
     s3_lightning_checkpoint = S3LightningCheckpoint(region=checkpoint_directory.region)
     _verify_user_agent(s3_lightning_checkpoint)
     trainer = L.Trainer(
+        accelerator=LIGHTNING_ACCELERATOR,
         default_root_dir=checkpoint_directory.s3_uri,
         plugins=[s3_lightning_checkpoint],
         max_epochs=1,
@@ -130,6 +134,7 @@ def test_compatibility_with_checkpoint_callback(checkpoint_directory):
         enable_version_counter=True,
     )
     trainer = L.Trainer(
+        accelerator=LIGHTNING_ACCELERATOR,
         plugins=[s3_lightning_checkpoint],
         callbacks=[checkpoint_callback],
         min_epochs=4,
@@ -163,6 +168,7 @@ def test_compatibility_with_async_checkpoint_io(checkpoint_directory):
     async_s3_lightning_checkpoint = AsyncCheckpointIO(s3_lightning_checkpoint)
 
     trainer = L.Trainer(
+        accelerator=LIGHTNING_ACCELERATOR,
         default_root_dir=checkpoint_directory.s3_uri,
         plugins=[async_s3_lightning_checkpoint],
         min_epochs=4,
@@ -189,6 +195,7 @@ def test_compatibility_with_lightning_checkpoint_load(checkpoint_directory):
     model = LightningTransformer(vocab_size=dataset.vocab_size)
     s3_lightning_checkpoint = S3LightningCheckpoint(region=checkpoint_directory.region)
     trainer = L.Trainer(
+        accelerator=LIGHTNING_ACCELERATOR,
         default_root_dir=checkpoint_directory.s3_uri,
         plugins=[s3_lightning_checkpoint],
         max_epochs=1,