[RLlib] Create ci/rllib_tests and organize a little (#11342)

ci/rllib_tests/regresssion_tests/ray-project/cluster.yaml

@@ -31,6 +31,7 @@ setup_commands:
     - conda uninstall -y terminado
     - source activate tensorflow_p36 && pip install -U ray-{{ray_version}}-cp36-cp36m-manylinux1_x86_64.whl
     - source activate tensorflow_p36 && pip install ray[rllib] ray[debug]
+    - source activate tensorflow_p36 && pip install torch==1.6 torchvision
     - source activate tensorflow_p36 && pip install boto3==1.4.8 cython==0.29.0
 
 # Command to start ray on the head node. You don't need to change this.

ci/rllib_tests/regresssion_tests/ray-project/project.yaml

@@ -42,7 +42,6 @@ commands:
 
   - name: run-regression-tests
     command: |
-      source activate tensorflow_p36 && pip install torch torchvision
       source activate tensorflow_p36 && rllib train -f compact-regression-test.yaml
     help: "Run rllib regression tests"
 
@@ -55,4 +54,4 @@ output_files: [
   # For example, uncomment this to save the logs from the
   # last ray job.
   # "/tmp/ray/session_latest",
-]
+]

ci/rllib_tests/regresssion_tests/run.sh

@@ -1,4 +1,3 @@
 #!/usr/bin/env bash
 
-source activate tensorflow_p36 && pip install torch==1.4 torchvision
 source activate tensorflow_p36 && rllib train -f compact-regression-test.yaml

ci/rllib_tests/unit_gpu_tests/ray-project/cluster.yaml

@@ -0,0 +1,42 @@
+cluster_name: ray-rllib-regression-tests
+
+min_workers: 0
+max_workers: 0
+
+# Cloud-provider specific configuration.
+provider:
+    type: aws
+    region: us-west-2
+    availability_zone: us-west-2a
+    cache_stopped_nodes: False
+
+# How Ray will authenticate with newly launched nodes.
+auth:
+    ssh_user: ubuntu
+
+head_node:
+    InstanceType: p2.xlarge  # Cheaper 1GPU K80 instance
+    ImageId: ami-07728e9e2742b0662 # Deep Learning AMI (Ubuntu 16.04)
+
+    # Set primary volume to 25 GiB
+    BlockDeviceMappings:
+        - DeviceName: /dev/sda1
+          Ebs:
+              VolumeSize: 100
+
+
+# List of shell commands to run to set up nodes.
+setup_commands:
+    - wget --quiet https://s3-us-west-2.amazonaws.com/ray-wheels/{{ray_branch}}/{{commit}}/ray-{{ray_version}}-cp36-cp36m-manylinux1_x86_64.whl
+    - conda uninstall -y terminado
+    - source activate tensorflow_p36 && pip install -U ray-{{ray_version}}-cp36-cp36m-manylinux1_x86_64.whl
+
+# Command to start ray on the head node. You don't need to change this.
+head_start_ray_commands:
+    - source activate tensorflow_p36 && ray stop
+    - ulimit -n 65536; source activate tensorflow_p36 && OMP_NUM_THREADS=1 ray start --head --redis-port=6379 --object-manager-port=8076 --autoscaling-config=~/ray_bootstrap_config.yaml
+
+# Command to start ray on worker nodes. You don't need to change this.
+worker_start_ray_commands:
+    - source activate tensorflow_p36 && ray stop
+    - ulimit -n 65536; source activate tensorflow_p36 && OMP_NUM_THREADS=1 ray start --address=$RAY_HEAD_IP:6379 --object-manager-port=8076

ci/rllib_tests/unit_gpu_tests/ray-project/requirements.txt

@@ -0,0 +1,6 @@
+ray[rllib]
+ray[debug]
+torch==1.6+cu101
+torchvision==0.7.0+cu101
+boto3==1.4.8
+cython==0.29.0

ci/rllib_tests/unit_gpu_tests/run.sh

@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+
+# Run all test cases, but with a forced num_gpus=1.
+# TODO: (sven) chose correct dir and run over all RLlib tests and example scripts!
+source activate tensorflow_p36 && export RAY_FORCE_NUM_GPUS=1 && cd ~ && python -m pytest test_attention_net_learning.py