feature(nyz): move atari_py to ale-py; split base and env docker build

.github/workflows/deploy.yml

@@ -3,7 +3,7 @@ name: deploy  # deploy docker
 on: [push]
 
 jobs:
-  docker:
+  docker_base:
     runs-on: ubuntu-latest
     if: "contains(github.event.head_commit.message, 'enable docker')"
     strategy:
@@ -29,14 +29,73 @@ jobs:
           if [[ $GITHUB_REF == refs/tags/* ]]; then
             VERSION=${GITHUB_REF#refs/tags/}
           fi
-          if [ "${{ github.event_name }}" = "schedule" ]; then
-            VERSION=nightly
-          fi
           TAGS="${DOCKER_IMAGE}:${VERSION}"
+          NIGHTLY_TAGS="${DOCKER_IMAGE}:nightly"
           if [[ $VERSION =~ ^v[0-9]{1,3}\.[0-9]{1,3}\.[0-9]{1,3}$ ]]; then
             TAGS="$TAGS,${DOCKER_IMAGE}:latest"
           fi
           echo ::set-output name=tags::${TAGS}
+          echo ::set-output name=nightlytags::${NIGHTLY_TAGS}
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v1
+        with:
+          platforms: all
+
+      - name: Set up Docker Buildx
+        id: buildx
+        uses: docker/setup-buildx-action@v1
+
+      - name: Cache Docker layers
+        uses: actions/cache@v2
+        with:
+          path: /tmp/.buildx-cache
+          key: ${{ runner.os }}-buildx-${{ github.sha }}
+          restore-keys: |
+            ${{ runner.os }}-buildx-
+      - name: Login to DockerHub
+        uses: docker/login-action@v1
+        with:
+          username: ${{ secrets.DOCKERIO_USERNAME }}
+          password: ${{ secrets.DOCKERIO_PASSWORD }}
+
+      - name: Build and push
+        id: docker_build
+        uses: docker/build-push-action@v2
+        with:
+          builder: ${{ steps.buildx.outputs.name }}
+          context: ./
+          file: ./Dockerfile.base
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: |
+            ${{ steps.prep.outputs.tags }}
+            ${{ steps.prep.outputs.nightlytags }}
+          cache-from: type=local,src=/tmp/.buildx-cache
+          cache-to: type=local,dest=/tmp/.buildx-cache
+
+      - name: Image digest
+        run: echo ${{ steps.docker_build.outputs.digest }}
+
+  docker_env:
+    runs-on: ubuntu-latest
+    needs: docker_base
+    if: "contains(github.event.head_commit.message, 'enable docker')"
+    strategy:
+      matrix:
+        platform: [linux/amd64]
+        # python-version: [3.6, 3.7, 3.8]
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v2
+      - name: Prepare
+        id: prep
+        env:
+          DOCKERIO_ORG: opendilab
+          TARGET: ding
+        run: |
+          DOCKER_IMAGE=$DOCKERIO_ORG/$TARGET
+          VERSION=nightly-atari
+          ATARI_TAGS="${DOCKER_IMAGE}:${VERSION}"
+          echo ::set-output name=ataritags::${ATARI_TAGS}
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v1
         with:
@@ -65,11 +124,12 @@ jobs:
         with:
           builder: ${{ steps.buildx.outputs.name }}
           context: ./
-          file: ./Dockerfile
+          file: ./Dockerfile.env
           push: ${{ github.event_name != 'pull_request' }}
-          tags: ${{ steps.prep.outputs.tags }}
+          tags: ${{ steps.prep.outputs.ataritags }}
           cache-from: type=local,src=/tmp/.buildx-cache
           cache-to: type=local,dest=/tmp/.buildx-cache
+          target: atari
 
       - name: Image digest
         run: echo ${{ steps.docker_build.outputs.digest }}

Dockerfile → Dockerfile.base

@@ -18,14 +18,4 @@ ADD dizoo dizoo
 ADD ding ding
 
 RUN python3 -m pip install --upgrade pip \
-    && python3 -m pip install --no-cache-dir .[common_env]
-
-# install ROMs with rar
-RUN curl -sL https://www.rarlab.com/rar/rarlinux-x64-6.0.1.tar.gz | tar -zxvf - \
-    && cd rar && make \
-    && cd .. && rm -rf rar \
-    && mkdir roms && cd roms \ 
-    && curl -O http://www.atarimania.com/roms/Roms.rar && unrar e Roms.rar \
-    && python -m atari_py.import_roms . \
-    && cd .. && rm -rf roms \
-    && rm /usr/local/bin/rar /usr/local/bin/unrar /etc/rarfiles.lst /usr/local/lib/default.sfx
+    && python3 -m pip install --no-cache-dir .[fast]

Dockerfile.env

@@ -0,0 +1,8 @@
+FROM opendilab/ding:nightly as atari
+
+WORKDIR /ding
+
+RUN python3 -m pip install --upgrade pip \
+    && python3 -m pip install --no-cache-dir .[common_env] \
+    && pip install autorom \
+    && AutoROM --accept-license

dizoo/atari/envs/__init__.py

@@ -1,2 +1 @@
 from .atari_env import AtariEnv, AtariEnvMR
-from .atari_multi_discrete_env import AtariMultiDiscreteEnv

dizoo/atari/envs/atari_env.py

@@ -185,6 +185,7 @@ def seed(self, seed: int, dynamic_seed: bool = True) -> None:
 
     def step(self, action: np.ndarray) -> BaseEnvTimestep:
         assert isinstance(action, np.ndarray), type(action)
+        action = action.item()
         obs, rew, done, info = self._env.step(action)
         self._final_eval_reward += rew
         obs = to_ndarray(obs)
@@ -240,7 +241,7 @@ class AtariEnvMR(AtariEnv):
 
     def reset(self) -> np.ndarray:
         if not self._init_flag:
-            self._make_env(only_info=False)
+            self._env = self._make_env(only_info=False)
             self._init_flag = True
         if hasattr(self, '_seed'):
             np_seed = 100 * np.random.randint(1, 1000)
@@ -251,7 +252,7 @@ def reset(self) -> np.ndarray:
         return obs
 
     def _make_env(self, only_info=False):
-        self._env = wrap_deepmind_mr(
+        return wrap_deepmind_mr(
             self._cfg.env_id,
             frame_stack=self._cfg.frame_stack,
             episode_life=self._cfg.is_train,
@@ -269,6 +270,7 @@ def info(self) -> BaseEnvInfo:
             info.obs_space.shape = obs_shape
             info.act_space.shape = act_shape
             info.rew_space.shape = rew_shape
+            return info
         else:
             raise NotImplementedError('{} not found in ATARIENV_INFO_DICT [{}]'\
                 .format(self._cfg.env_id, ATARIENV_INFO_DICT.keys()))

dizoo/atari/envs/test_atari_env.py

@@ -5,7 +5,7 @@
 from dizoo.atari.envs import AtariEnv, AtariEnvMR
 
 
-@pytest.mark.unittest
+@pytest.mark.envtest
 class TestAtariEnv:
 
     def test_pong(self):

setup.py

@@ -47,7 +47,7 @@
         'numpy>=1.10',
         'requests>=2.25.1',
         'six',
-        'gym>=0.15.3',  # pypy incompatible
+        'gym>=0.20.0',  # pypy incompatible
         'torch>=1.3.1,<=1.8.0',
         'pyyaml',
         'easydict==1.9',
@@ -88,11 +88,13 @@
         'fast': [
             'numpy-stl',
             'numba>=0.53.0',
+        ],
+        'dist': [
             'redis==3.5.3',
             'redis-py-cluster==2.1.0',
         ],
         'common_env': [
-            'atari_py',
+            'ale-py',  # atari
             'box2d-py',
             'cmake>=3.18.4',
             'opencv-python',  # pypy incompatible