create per environment Dockerfiles. (#70)

* create per environment Dockerfiles. Adjust CI setup to better parallelize runs. Fix a couple of issues in golden and trace tests. Update a few of the docs. * bugfix in mmc agent. Also install kubectl for CI, update badge branch. * remove integration test parallelism.
2025-12-18 11:40:18 +01:00 · 2018-11-14 07:40:22 -08:00
parent a849c17e46
commit 524f8436a2
20 changed files with 448 additions and 139 deletions
--- a/.circleci/config.yml
+++ b/.circleci/config.yml
@@ -30,10 +30,12 @@ aliases:
        sudo curl -o /usr/local/bin/aws-iam-authenticator https://amazon-eks.s3-us-west-2.amazonaws.com/1.10.3/2018-07-26/bin/linux/amd64/aws-iam-authenticator
        sudo chmod a+x /usr/local/bin/aws-iam-authenticator
        aws eks update-kubeconfig --name coach-aws-cicd
+        sudo curl -o /usr/local/bin/kubectl https://storage.googleapis.com/kubernetes-release/release/$(curl -s https://storage.googleapis.com/kubernetes-release/release/stable.txt)/bin/linux/amd64/kubectl
+        sudo chmod a+x /usr/local/bin/kubectl

 version: 2
 jobs:
-  build:
+  build_base:
    <<: *executor_prep
    steps:
      - checkout
@@ -41,7 +43,7 @@ jobs:
      - *restore_cache
      - *aws_prep
      - run:
-          name: Build and push container
+          name: Build and push base and main container
          command: |
            REGISTRY=316971102342.dkr.ecr.us-west-2.amazonaws.com
            TAG=$(git describe --tags --always --dirty)
@@ -53,10 +55,64 @@ jobs:

            docker tag ${REGISTRY}/coach-base:${TAG} coach-base:master

-            docker build -t ${REGISTRY}/coach:${TAG} -f docker/Dockerfile .
+            docker build --build-arg MUJOCO_KEY=${MUJOCO_KEY} -t ${REGISTRY}/coach:${TAG} -f docker/Dockerfile .
            docker push ${REGISTRY}/coach:${TAG}
          no_output_timeout: 30m

+  build_gym_env:
+    <<: *executor_prep
+    steps:
+      - checkout
+      - *remote_docker
+      - *restore_cache
+      - *aws_prep
+      - run:
+          name: Build and push gym environment container
+          command: |
+            REGISTRY=316971102342.dkr.ecr.us-west-2.amazonaws.com
+            TAG=$(git describe --tags --always --dirty)
+            docker pull ${REGISTRY}/coach-base:${MASTER_BRANCH}
+            docker tag ${REGISTRY}/coach-base:${MASTER_BRANCH} coach-base:master
+            docker build --cache-from ${REGISTRY}/coach-base:${MASTER_BRANCH} -t ${REGISTRY}/coach-gym_environment:${TAG} -f docker/Dockerfile.gym_environment .
+            docker push ${REGISTRY}/coach-gym_environment:${TAG}
+          no_output_timeout: 10m
+
+  build_doom_env:
+    <<: *executor_prep
+    steps:
+      - checkout
+      - *remote_docker
+      - *restore_cache
+      - *aws_prep
+      - run:
+          name: Build and push vizdoom environment container
+          command: |
+            REGISTRY=316971102342.dkr.ecr.us-west-2.amazonaws.com
+            TAG=$(git describe --tags --always --dirty)
+            docker pull ${REGISTRY}/coach-base:${MASTER_BRANCH}
+            docker tag ${REGISTRY}/coach-base:${MASTER_BRANCH} coach-base:master
+            docker build --cache-from ${REGISTRY}/coach-base:${MASTER_BRANCH} -t ${REGISTRY}/coach-doom_environment:${TAG} -f docker/Dockerfile.doom_environment .
+            docker push ${REGISTRY}/coach-doom_environment:${TAG}
+          no_output_timeout: 10m
+
+  build_mujoco_env:
+    <<: *executor_prep
+    steps:
+      - checkout
+      - *remote_docker
+      - *restore_cache
+      - *aws_prep
+      - run:
+          name: Build and push mujoco environment container
+          command: |
+            REGISTRY=316971102342.dkr.ecr.us-west-2.amazonaws.com
+            TAG=$(git describe --tags --always --dirty)
+            docker pull ${REGISTRY}/coach-base:${MASTER_BRANCH}
+            docker tag ${REGISTRY}/coach-base:${MASTER_BRANCH} coach-base:master
+            docker build --cache-from ${REGISTRY}/coach-base:${MASTER_BRANCH} --build-arg MUJOCO_KEY=${MUJOCO_KEY} -t ${REGISTRY}/coach-mujoco_environment:${TAG} -f docker/Dockerfile.mujoco_environment .
+            docker push ${REGISTRY}/coach-mujoco_environment:${TAG}
+          no_output_timeout: 10m
+
  unit_tests:
    <<: *executor_prep
    steps:
@@ -70,6 +126,14 @@ jobs:
            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
            python3 rl_coach/tests/test_eks.py  -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn unit-test -tc 'make unit_tests_without_docker' -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
+      - run:
+          name: cleanup
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            kubectl delete --all pods --namespace=unit-test-${CIRCLE_BUILD_NUM} || true
+            kubectl delete ns unit-test-${CIRCLE_BUILD_NUM} || true
+          when: always

  integration_tests:
    <<: *executor_prep
@@ -84,8 +148,16 @@ jobs:
            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
            python3 rl_coach/tests/test_eks.py  -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn integration-test -tc 'make integration_tests_without_docker' -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
+      - run:
+          name: cleanup
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            kubectl delete --all pods --namespace=integration-test-${CIRCLE_BUILD_NUM} || true
+            kubectl delete ns integration-test-${CIRCLE_BUILD_NUM} || true
+          when: always

-  golden_tests:
+  golden_test_gym:
    <<: *executor_prep
    steps:
      - checkout
@@ -93,13 +165,23 @@ jobs:
      - *restore_cache
      - *aws_prep
      - run:
-          name: run golden tests
+          name: run gym related golden tests
          command: |
            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
-            python3 rl_coach/tests/test_eks.py  -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn golden-test -tc 'make golden_tests_without_docker' -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
+            export PRESETS='CartPole_A3C,CartPole_Dueling_DDQN,CartPole_NStepQ,CartPole_DQN,CartPole_DFP,CartPole_PG,CartPole_NEC,CartPole_ClippedPPO,CartPole_PAL'
+            python3 rl_coach/tests/test_eks.py -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn golden-test-gym -tc "export PRESETS=${PRESETS} && make golden_tests_without_docker" -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach-gym_environment:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
+          no_output_timeout: 30m
+      - run:
+          name: cleanup
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            kubectl delete --all pods --namespace=golden-test-gym-${CIRCLE_BUILD_NUM} || true
+            kubectl delete ns golden-test-gym-${CIRCLE_BUILD_NUM} || true
+          when: always

-  trace_tests:
+  golden_test_doom:
    <<: *executor_prep
    steps:
      - checkout
@@ -107,30 +189,197 @@ jobs:
      - *restore_cache
      - *aws_prep
      - run:
-          name: run trace tests
+          name: run doom related golden tests
          command: |
            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
-            python3 rl_coach/tests/test_eks.py  -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn trace-test -tc 'make trace_tests_without_docker' -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
+            export PRESETS='Doom_Basic_DQN,Doom_Basic_A3C,Doom_Health_DFP'
+            python3 rl_coach/tests/test_eks.py -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn golden-test-doom -tc "export PRESETS=${PRESETS} && make golden_tests_without_docker" -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach-doom_environment:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
+          no_output_timeout: 30m
+      - run:
+          name: cleanup
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            kubectl delete --all pods --namespace=golden-test-doom-${CIRCLE_BUILD_NUM} || true
+            kubectl delete ns golden-test-doom-${CIRCLE_BUILD_NUM} || true
+          when: always
+
+  golden_test_mujoco:
+    <<: *executor_prep
+    steps:
+      - checkout
+      - *remote_docker
+      - *restore_cache
+      - *aws_prep
+      - run:
+          name: run mujoco related golden tests
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            export PRESETS='BitFlip_DQN_HER,BitFlip_DQN,Mujoco_A3C,Mujoco_A3C_LSTM,Mujoco_PPO,Mujoco_ClippedPPO,Mujoco_DDPG'
+            python3 rl_coach/tests/test_eks.py -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn golden-test-mujoco -tc "export PRESETS=${PRESETS} && make golden_tests_without_docker" -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach-mujoco_environment:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
+          no_output_timeout: 30m
+      - run:
+          name: cleanup
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            kubectl delete --all pods --namespace=golden-test-mujoco-${CIRCLE_BUILD_NUM} || true
+            kubectl delete ns golden-test-mujoco-${CIRCLE_BUILD_NUM} || true
+          when: always
+
+  trace_test_gym:
+    <<: *executor_prep
+    steps:
+      - checkout
+      - *remote_docker
+      - *restore_cache
+      - *aws_prep
+      - run:
+          name: run gym related trace tests
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            export PRESETS='CartPole_A3C,CartPole_Dueling_DDQN,CartPole_NStepQ,CartPole_DQN,CartPole_DFP,CartPole_PG,CartPole_NEC,CartPole_ClippedPPO,CartPole_PAL'
+            python3 rl_coach/tests/test_eks.py -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn trace-test-gym -tc "export PRESETS=${PRESETS} && make trace_tests_without_docker" -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach-gym_environment:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
+          no_output_timeout: 30m
+      - run:
+          name: cleanup
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            kubectl delete --all pods --namespace=trace-test-gym-${CIRCLE_BUILD_NUM} || true
+            kubectl delete ns trace-test-gym-${CIRCLE_BUILD_NUM} || true
+          when: always
+
+  trace_test_doom:
+    <<: *executor_prep
+    steps:
+      - checkout
+      - *remote_docker
+      - *restore_cache
+      - *aws_prep
+      - run:
+          name: run doom related trace tests
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            export PRESETS='Doom_Basic_DQN,Doom_Basic_A3C,Doom_Health_DFP'
+            python3 rl_coach/tests/test_eks.py -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn trace-test-doom -tc "export PRESETS=${PRESETS} && make trace_tests_without_docker" -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach-doom_environment:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
+          no_output_timeout: 30m
+      - run:
+          name: cleanup
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            kubectl delete --all pods --namespace=trace-test-doom-${CIRCLE_BUILD_NUM} || true
+            kubectl delete ns trace-test-doom-${CIRCLE_BUILD_NUM} || true
+          when: always
+
+  trace_test_mujoco:
+    <<: *executor_prep
+    steps:
+      - checkout
+      - *remote_docker
+      - *restore_cache
+      - *aws_prep
+      - run:
+          name: run mujoco related trace tests
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            export PRESETS='BitFlip_DQN_HER,BitFlip_DQN,Mujoco_A3C,Mujoco_A3C_LSTM,Mujoco_PPO,Mujoco_ClippedPPO,Mujoco_DDPG'
+            python3 rl_coach/tests/test_eks.py -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn trace-test-mujoco -tc "export PRESETS=${PRESETS} && make trace_tests_without_docker" -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach-mujoco_environment:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
+          no_output_timeout: 30m
+      - run:
+          name: cleanup
+          command: |
+            export AWS_ACCESS_KEY_ID=`echo ${AWS_ACCESS_KEY_ID} | base64 --decode`
+            export AWS_SECRET_ACCESS_KEY=`echo ${AWS_SECRET_ACCESS_KEY} | base64 --decode`
+            kubectl delete --all pods --namespace=trace-test-mujoco-${CIRCLE_BUILD_NUM} || true
+            kubectl delete ns trace-test-mujoco-${CIRCLE_BUILD_NUM} || true
+          when: always
+
+  container_deploy:
+    <<: *executor_prep
+    steps:
+      - checkout
+      - *remote_docker
+      - *restore_cache
+      - *aws_prep
+      - run:
+          name: Tag and push updated base and main container
+          command: |
+            REGISTRY=316971102342.dkr.ecr.us-west-2.amazonaws.com
+            TAG=$(git describe --tags --always --dirty)
+
+            docker pull ${REGISTRY}/coach-base:${TAG}
+            docker tag ${REGISTRY}/coach-base:${TAG} coach-base:${MASTER_BRANCH}
+            docker push ${REGISTRY}/coach-base:${MASTER_BRANCH}
+
+            docker pull ${REGISTRY}/coach:${TAG}
+            docker tag ${REGISTRY}/coach:${TAG} coach:${MASTER_BRANCH}
+            docker push ${REGISTRY}/coach:${MASTER_BRANCH}

 workflows:
  version: 2
  build_and_test:
    jobs:
-      - build
+      - build_base
      - unit_tests:
          requires:
-            - build
+            - build_base
      - integration_tests:
          requires:
-            - build
+            - build_base
      - e2e_approval:
          type: approval
          requires:
-            - build
-      - golden_tests:
+            - build_base
+      - build_gym_env:
          requires:
            - e2e_approval
-      - trace_tests:
+      - build_doom_env:
          requires:
            - e2e_approval
+      - build_mujoco_env:
+          requires:
+            - e2e_approval
+      - gym_approval:
+          type: approval
+          requires:
+            - golden_test_gym
+      - doom_approval:
+          type: approval
+          requires:
+            - golden_test_doom
+      - mujoco_approval:
+          type: approval
+          requires:
+            - golden_test_mujoco
+      - golden_test_gym:
+          requires:
+            - build_gym_env
+      - golden_test_doom:
+          requires:
+            - build_doom_env
+      - golden_test_mujoco:
+          requires:
+            - build_mujoco_env
+      - trace_test_gym:
+          requires:
+            - gym_approval
+      - trace_test_doom:
+          requires:
+            - doom_approval
+      - trace_test_mujoco:
+          requires:
+            - mujoco_approval
+      - container_deploy:
+          requires:
+            - unit_tests
+            - integration_tests
+          filters:
+            branches:
+              only: 0.11.0-release