More debugging belt and braces

Author: ggorman

.github/workflows/pytest-gpu.yml

@@ -87,7 +87,10 @@ jobs:
 
       - name: Build docker image
         run: |
+          docker buildx create --use --name $RUNNER_NAME || true
           docker build . \
+            --builder $RUNNER_NAME \
+            --label ci-run=$GITHUB_RUN_ID \
             --rm --pull \
             --file docker/Dockerfile.devito \
             --tag "${DOCKER_IMAGE}" \
@@ -102,19 +105,40 @@ jobs:
           docker run ${{ matrix.flags }} "${DOCKER_IMAGE}" ${{ matrix.test_drive_cmd }}
 
       - name: Test with pytest
+        env:
+          # Exported earlier in the job; needed inside the container for codecov
+          CODECOV_TOKEN: ${{ env.CODECOV_TOKEN }}
         run: |
+          # ------------------------------------------------------------------
+          # 1.  Add Codecov’s environment variables (GITHUB_SHA, etc.)
+          # ------------------------------------------------------------------
           ci_env=$(bash <(curl -s https://codecov.io/env))
 
-          # docker run ${{ matrix.flags }} $ci_env -e CI=true "${DOCKER_IMAGE}" \
-          docker run ${{ matrix.flags }} $ci_env -e CI=true  \
+          # ------------------------------------------------------------------
+          # 2.  Make sure the runner really gave us a GPU mask
+          # ------------------------------------------------------------------
+          : "${CUDA_VISIBLE_DEVICES:?CUDA_VISIBLE_DEVICES not set}"
+
+          # ------------------------------------------------------------------
+          # 3.  Run the test suite inside the image we just built
+          # ------------------------------------------------------------------
+          docker run \
+            --init --rm -t \
+            --name "${CONTAINER_BASENAME}" \
+            --env CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES}" \
+            --gpus "device=${CUDA_VISIBLE_DEVICES}" \
+            ${ci_env} \
+            -e CI=true \
             -e PYTHONFAULTHANDLER=1 \
             -e CUDA_LAUNCH_BLOCKING=1 \
             -e NV_ACC_NOTIFY=3 \
             -e NV_ACC_DEBUG=1 \
             -e DEVITO_LOGGING=DEBUG \
+            -e CODECOV_TOKEN \
             "${DOCKER_IMAGE}" \
             pytest -vv -ra -l -s --full-trace --maxfail=1 \
-            --cov --cov-config=.coveragerc --cov-report=xml ${{ matrix.test_files }}
+                   --cov --cov-config=.coveragerc --cov-report=xml \
+                   ${{ matrix.test_files }}
 
       - name: Test examples
         run: |
@@ -128,6 +152,9 @@ jobs:
       - name: Clean up test image
         if: always()
         run: |
-          docker rmi -f "${DOCKER_IMAGE}" || true     # remove the image
-          docker system prune -f                      # free dangling layers
+          # Remove only the image we just built
+          docker rmi -f "${DOCKER_IMAGE}" || true
 
+          # Guard global prune the layers we created
+          # (label ci-run=$GITHUB_RUN_ID)
+          docker image prune -f --filter label=ci-run=$GITHUB_RUN_ID