diff --git a/.travis.yml b/.travis.yml
index d7f3a2bea34c..a8a63caee9f2 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -26,14 +26,14 @@ matrix:
         - JDK='Oracle JDK 8'
         - PYTHON=3.5 PYTHONWARNINGS=ignore
       install:
-        - ./.travis/install-dependencies.sh
+        - ./ci/travis/install-dependencies.sh
         - export PATH="$HOME/miniconda/bin:$PATH"
       script:
       - ./java/test.sh
 
       # Test Bazel build
       - rm -rf build
-      - ./.travis/install-bazel.sh
+      - ./ci/travis/install-bazel.sh
       - bazel build ...
 
     - os: linux
@@ -47,10 +47,10 @@ matrix:
         - sudo apt-get install -qq clang-format-3.8
       install: []
       script:
-        - .travis/check-git-clang-format-output.sh
+        - ./ci/travis/check-git-clang-format-output.sh
         # Try generating Sphinx documentation. To do this, we need to install
         # Ray first.
-        - ./.travis/install-dependencies.sh
+        - ./ci/travis/install-dependencies.sh
         - export PATH="$HOME/miniconda/bin:$PATH"
         - cd doc
         - pip install -q -r requirements-doc.txt
@@ -59,7 +59,7 @@ matrix:
         - cd ..
         # Run Python linting, ignore dict vs {} (C408), others are defaults
         - flake8 --exclude=python/ray/core/generated/,doc/source/conf.py,python/ray/cloudpickle/ --ignore=C408,E121,E123,E126,E226,E24,E704,W503,W504,W605
-        - .travis/format.sh --all
+        - ./ci/travis/format.sh --all
         # Make sure that the README is formatted properly.
         - cd python
         - python setup.py check --restructuredtext --strict --metadata
@@ -72,7 +72,7 @@ matrix:
         - export GOPATH="$HOME/go_dir"
         - export PATH="$GOROOT/bin:$GOPATH/bin:$PATH"
         - go get github.com/bazelbuild/buildtools/buildifier
-        - ./.travis/bazel-format.sh
+        - ./ci/travis/bazel-format.sh
 
     - os: linux
       dist: trusty
@@ -81,10 +81,10 @@ matrix:
         - sudo apt-get update -qq
         - sudo apt-get install -qq valgrind
       install:
-        - ./.travis/install-dependencies.sh
+        - ./ci/travis/install-dependencies.sh
         - export PATH="$HOME/miniconda/bin:$PATH"
-        - ./.travis/install-ray.sh
-        - ./.travis/install-with-cache-libs.sh
+        - ./ci/travis/install-ray.sh
+        - ./ci/travis/install-with-cache-libs.sh
 
       script:
         - cd build
@@ -98,10 +98,10 @@ matrix:
 
         # # Python3.5+ only. Otherwise we will get `SyntaxError` regardless of how we set the tester.
         - python -c 'import sys;exit(sys.version_info>=(3,5))' || python -m pytest -v --durations=10 python/ray/experimental/test/async_test.py
-        - python -m pytest -v --durations=10 test/mini_test.py
-        - python -m pytest -v --durations=10 test/array_test.py
-        - python -m pytest -v --durations=10 test/multi_node_test_2.py
-        - python -m pytest -v --durations=10 test/node_manager_test.py
+        - python -m pytest -v --durations=10 python/ray/tests/test_mini.py
+        - python -m pytest -v --durations=10 python/ray/tests/test_array.py
+        - python -m pytest -v --durations=10 python/ray/tests/test_multi_node_2.py
+        - python -m pytest -v --durations=10 python/ray/tests/test_node_manager.py
 
 
     # Build Linux wheels.
@@ -109,22 +109,22 @@ matrix:
       dist: trusty
       env: LINUX_WHEELS=1 PYTHONWARNINGS=ignore
       install:
-        - ./.travis/install-dependencies.sh
+        - ./ci/travis/install-dependencies.sh
         # This command should be kept in sync with ray/python/README-building-wheels.md.
         - docker run --rm -w /ray -v `pwd`:/ray -ti quay.io/xhochy/arrow_manylinux1_x86_64_base:latest /ray/python/build-wheel-manylinux1.sh
       script:
-        - ./.travis/test-wheels.sh
+        - ./ci/travis/test-wheels.sh
 
     # Build MacOS wheels.
     - os: osx
       osx_image: xcode7
       env: MAC_WHEELS=1 PYTHONWARNINGS=ignore
       install:
-        - ./.travis/install-dependencies.sh
+        - ./ci/travis/install-dependencies.sh
         # This command should be kept in sync with ray/python/README-building-wheels.md.
         - ./python/build-wheel-macos.sh
       script:
-        - ./.travis/test-wheels.sh
+        - ./ci/travis/test-wheels.sh
 
     # Test GCS integration
     - os: linux
@@ -136,10 +136,10 @@ matrix:
 
 
 install:
-  - ./.travis/install-dependencies.sh
+  - ./ci/travis/install-dependencies.sh
   - export PATH="$HOME/miniconda/bin:$PATH"
-  - ./.travis/install-ray.sh
-  - ./.travis/install-cython-examples.sh
+  - ./ci/travis/install-ray.sh
+  - ./ci/travis/install-cython-examples.sh
 
   - cd build
   - bash ../src/ray/test/run_gcs_tests.sh
@@ -162,7 +162,7 @@ script:
   # class definitions: The main module of tests that are run
   # with pytest have the same name as the test file -- and this
   # module is only found if the test directory is in the PYTHONPATH.
-  - export PYTHONPATH="$PYTHONPATH:./test/"
+  # - export PYTHONPATH="$PYTHONPATH:./ci/"
 
   # ray tune tests
   - python python/ray/tune/test/dependency_test.py
@@ -182,43 +182,46 @@ script:
   # Python3.5+ only. Otherwise we will get `SyntaxError` regardless of how we set the tester.
   - python -c 'import sys;exit(sys.version_info>=(3,5))' || python -m pytest -v --durations=10 python/ray/experimental/test/async_test.py
 
-  - python -m pytest -v --durations=10 python/ray/test/test_global_state.py
-  - python -m pytest -v --durations=10 python/ray/test/test_queue.py
-  - python -m pytest -v --durations=10 python/ray/test/test_ray_init.py
-  - python -m pytest -v --durations=10 test/mini_test.py
-
-  - python -m pytest -v --durations=10 test/runtest.py
-  - python -m pytest -v --durations=10 test/array_test.py
-  - python -m pytest -v --durations=10 test/actor_test.py
-  - python -m pytest -v --durations=10 test/autoscaler_test.py
-  - python -m pytest -v --durations=10 test/tensorflow_test.py
-  - python -m pytest -v --durations=10 test/failure_test.py
-  - python -m pytest -v --durations=10 test/microbenchmarks.py
-  - python -m pytest -v --durations=10 test/stress_tests.py
-  - python -m pytest -v --durations=10 test/component_failures_test.py
-  - python -m pytest -v --durations=10 test/multi_node_test.py
-  - python -m pytest -v --durations=10 test/multi_node_test_2.py
-  - python -m pytest -v --durations=10 test/recursion_test.py
-  - python -m pytest -v --durations=10 test/monitor_test.py
-  - python -m pytest -v --durations=10 test/cython_test.py
-  - python -m pytest -v --durations=10 test/credis_test.py
-  - python -m pytest -v --durations=10 test/node_manager_test.py
-  - python -m pytest -v --durations=10 test/test_signal.py
+  # ray tests
+  # TODO(williamma12): We cannot use pytests built-in test discovery because
+  # it causes a lot of the tests to fail on travis' apple builds even though
+  # it runs without issue on an apple build locally.
+  - python -m pytest -v --durations=10 python/ray/tests/test_global_state.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_queue.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_ray_init.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_mini.py
+
+  - python -m pytest -v --durations=10 python/ray/tests/test_basic.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_array.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_actor.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_autoscaler.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_tensorflow.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_failure.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_microbenchmarks.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_stress.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_component_failures.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_multi_node.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_multi_node_2.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_recursion.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_monitors.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_cython.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_credis.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_node_manager.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_signal.py
   # TODO(yuhguo): object_manager_test.py requires a lot of CPU/memory, and
   # better be put in Jenkins. However, it fails frequently in Jenkins, but
   # works well in Travis. We should consider moving it back to Jenkins once
   # we figure out the reason.
-  - python -m pytest -v --durations=10 test/object_manager_test.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_object_manager.py
 
   # ray temp file tests
-  - python -m pytest -v --durations=10 test/tempfile_test.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_tempfile.py
 
   # ray debug tools tests
-  - python -m pytest -v --durations=10 test/debug_tools_test.py
+  - python -m pytest -v --durations=10 python/ray/tests/test_debug_tools.py
 
   # modin test files
-  - python -m pytest -v --durations=10 python/ray/test/test_modin.py
-
+  - python -m pytest -v --durations=10 python/ray/tests/test_modin.py
 deploy:
   - provider: s3
     access_key_id: AKIAJ2L7XDUSZVTXI5QA
diff --git a/test/jenkins_tests/multi_node_docker_test.py b/ci/jenkins_tests/multi_node_docker_test.py
similarity index 100%
rename from test/jenkins_tests/multi_node_docker_test.py
rename to ci/jenkins_tests/multi_node_docker_test.py
diff --git a/test/jenkins_tests/multi_node_tests/large_memory_test.py b/ci/jenkins_tests/multi_node_tests/large_memory_test.py
similarity index 100%
rename from test/jenkins_tests/multi_node_tests/large_memory_test.py
rename to ci/jenkins_tests/multi_node_tests/large_memory_test.py
diff --git a/test/jenkins_tests/multi_node_tests/many_drivers_test.py b/ci/jenkins_tests/multi_node_tests/many_drivers_test.py
similarity index 95%
rename from test/jenkins_tests/multi_node_tests/many_drivers_test.py
rename to ci/jenkins_tests/multi_node_tests/many_drivers_test.py
index 94eeb4715e66..585c3806103a 100644
--- a/test/jenkins_tests/multi_node_tests/many_drivers_test.py
+++ b/ci/jenkins_tests/multi_node_tests/many_drivers_test.py
@@ -6,8 +6,8 @@
 import time
 
 import ray
-from ray.test.test_utils import (_wait_for_nodes_to_join, _broadcast_event,
-                                 _wait_for_event)
+from ray.tests.utils import (_wait_for_nodes_to_join, _broadcast_event,
+                             _wait_for_event)
 
 # This test should be run with 5 nodes, which have 0, 0, 5, 6, and 50 GPUs for
 # a total of 61 GPUs. It should be run with a large number of drivers (e.g.,
diff --git a/test/jenkins_tests/multi_node_tests/remove_driver_test.py b/ci/jenkins_tests/multi_node_tests/remove_driver_test.py
similarity index 98%
rename from test/jenkins_tests/multi_node_tests/remove_driver_test.py
rename to ci/jenkins_tests/multi_node_tests/remove_driver_test.py
index 08a100670997..1cd10195b607 100644
--- a/test/jenkins_tests/multi_node_tests/remove_driver_test.py
+++ b/ci/jenkins_tests/multi_node_tests/remove_driver_test.py
@@ -6,8 +6,8 @@
 import time
 
 import ray
-from ray.test.test_utils import (_wait_for_nodes_to_join, _broadcast_event,
-                                 _wait_for_event, wait_for_pid_to_exit)
+from ray.tests.utils import (_wait_for_nodes_to_join, _broadcast_event,
+                             _wait_for_event, wait_for_pid_to_exit)
 
 # This test should be run with 5 nodes, which have 0, 1, 2, 3, and 4 GPUs for a
 # total of 10 GPUs. It should be run with 7 drivers. Drivers 2 through 6 must
diff --git a/test/jenkins_tests/multi_node_tests/test_0.py b/ci/jenkins_tests/multi_node_tests/test_0.py
similarity index 100%
rename from test/jenkins_tests/multi_node_tests/test_0.py
rename to ci/jenkins_tests/multi_node_tests/test_0.py
diff --git a/test/jenkins_tests/multi_node_tests/test_wait_hanging.py b/ci/jenkins_tests/multi_node_tests/test_wait_hanging.py
similarity index 100%
rename from test/jenkins_tests/multi_node_tests/test_wait_hanging.py
rename to ci/jenkins_tests/multi_node_tests/test_wait_hanging.py
diff --git a/test/jenkins_tests/run_asv.sh b/ci/jenkins_tests/run_asv.sh
similarity index 100%
rename from test/jenkins_tests/run_asv.sh
rename to ci/jenkins_tests/run_asv.sh
diff --git a/test/jenkins_tests/run_multi_node_tests.sh b/ci/jenkins_tests/run_multi_node_tests.sh
similarity index 100%
rename from test/jenkins_tests/run_multi_node_tests.sh
rename to ci/jenkins_tests/run_multi_node_tests.sh
diff --git a/test/jenkins_tests/run_rllib_asv.sh b/ci/jenkins_tests/run_rllib_asv.sh
similarity index 100%
rename from test/jenkins_tests/run_rllib_asv.sh
rename to ci/jenkins_tests/run_rllib_asv.sh
diff --git a/test/stress_tests/application_cluster_template.yaml b/ci/stress_tests/application_cluster_template.yaml
similarity index 100%
rename from test/stress_tests/application_cluster_template.yaml
rename to ci/stress_tests/application_cluster_template.yaml
diff --git a/test/stress_tests/run_application_stress_tests.sh b/ci/stress_tests/run_application_stress_tests.sh
similarity index 100%
rename from test/stress_tests/run_application_stress_tests.sh
rename to ci/stress_tests/run_application_stress_tests.sh
diff --git a/test/stress_tests/run_stress_tests.sh b/ci/stress_tests/run_stress_tests.sh
similarity index 100%
rename from test/stress_tests/run_stress_tests.sh
rename to ci/stress_tests/run_stress_tests.sh
diff --git a/test/stress_tests/stress_testing_config.yaml b/ci/stress_tests/stress_testing_config.yaml
similarity index 100%
rename from test/stress_tests/stress_testing_config.yaml
rename to ci/stress_tests/stress_testing_config.yaml
diff --git a/test/stress_tests/test_dead_actors.py b/ci/stress_tests/test_dead_actors.py
similarity index 100%
rename from test/stress_tests/test_dead_actors.py
rename to ci/stress_tests/test_dead_actors.py
diff --git a/test/stress_tests/test_many_tasks_and_transfers.py b/ci/stress_tests/test_many_tasks_and_transfers.py
similarity index 100%
rename from test/stress_tests/test_many_tasks_and_transfers.py
rename to ci/stress_tests/test_many_tasks_and_transfers.py
diff --git a/.travis/bazel-format.sh b/ci/travis/bazel-format.sh
similarity index 91%
rename from .travis/bazel-format.sh
rename to ci/travis/bazel-format.sh
index ec5fa5bb0cf5..f3b4b1a9aad8 100755
--- a/.travis/bazel-format.sh
+++ b/ci/travis/bazel-format.sh
@@ -43,7 +43,7 @@ while [[ $# > 0 ]]; do
   shift
 done
 
-pushd $ROOT_DIR/..
+pushd $ROOT_DIR/../..
 BAZEL_FILES="bazel/BUILD bazel/BUILD.plasma bazel/ray.bzl BUILD.bazel WORKSPACE"
-buildifier -mode=$RUN_TYPE -diff_command="diff -u" $BAZEL_FILES 
+buildifier -mode=$RUN_TYPE -diff_command="diff -u" $BAZEL_FILES
 popd
diff --git a/.travis/check-git-clang-format-output.sh b/ci/travis/check-git-clang-format-output.sh
similarity index 80%
rename from .travis/check-git-clang-format-output.sh
rename to ci/travis/check-git-clang-format-output.sh
index 1b78e8175e6b..4d621848fdd1 100755
--- a/.travis/check-git-clang-format-output.sh
+++ b/ci/travis/check-git-clang-format-output.sh
@@ -8,7 +8,7 @@ else
   base_commit="$TRAVIS_BRANCH"
   echo "Running clang-format against branch $base_commit, with hash $(git rev-parse $base_commit)"
 fi
-output="$(.travis/git-clang-format --binary clang-format-3.8 --commit $base_commit --diff --exclude '(.*thirdparty/|.*redismodule.h|.*.js)')"
+output="$(ci/travis/git-clang-format --binary clang-format-3.8 --commit $base_commit --diff --exclude '(.*thirdparty/|.*redismodule.h|.*.js)')"
 if [ "$output" == "no modified files to format" ] || [ "$output" == "clang-format did not modify any files" ] ; then
   echo "clang-format passed."
   exit 0
diff --git a/.travis/format.sh b/ci/travis/format.sh
similarity index 100%
rename from .travis/format.sh
rename to ci/travis/format.sh
diff --git a/.travis/git-clang-format b/ci/travis/git-clang-format
similarity index 100%
rename from .travis/git-clang-format
rename to ci/travis/git-clang-format
diff --git a/.travis/install-bazel.sh b/ci/travis/install-bazel.sh
similarity index 100%
rename from .travis/install-bazel.sh
rename to ci/travis/install-bazel.sh
diff --git a/.travis/install-cython-examples.sh b/ci/travis/install-cython-examples.sh
similarity index 93%
rename from .travis/install-cython-examples.sh
rename to ci/travis/install-cython-examples.sh
index bc742c132cb6..58f44a15e102 100755
--- a/.travis/install-cython-examples.sh
+++ b/ci/travis/install-cython-examples.sh
@@ -7,7 +7,7 @@ ROOT_DIR=$(cd "$(dirname "${BASH_SOURCE:-$0}")"; pwd)
 
 echo "PYTHON is $PYTHON"
 
-cython_examples="$ROOT_DIR/../examples/cython"
+cython_examples="$ROOT_DIR/../../examples/cython"
 
 if [[ "$PYTHON" == "2.7" ]]; then
 
diff --git a/.travis/install-dependencies.sh b/ci/travis/install-dependencies.sh
similarity index 100%
rename from .travis/install-dependencies.sh
rename to ci/travis/install-dependencies.sh
diff --git a/.travis/install-ray.sh b/ci/travis/install-ray.sh
similarity index 84%
rename from .travis/install-ray.sh
rename to ci/travis/install-ray.sh
index 28aeeef3267a..79ca25d7abae 100755
--- a/.travis/install-ray.sh
+++ b/ci/travis/install-ray.sh
@@ -9,21 +9,21 @@ echo "PYTHON is $PYTHON"
 
 if [[ "$PYTHON" == "2.7" ]]; then
 
-  pushd "$ROOT_DIR/../python"
+  pushd "$ROOT_DIR/../../python"
     python setup.py install --user
   popd
 
 elif [[ "$PYTHON" == "3.5" ]]; then
   export PATH="$HOME/miniconda/bin:$PATH"
 
-  pushd "$ROOT_DIR/../python"
+  pushd "$ROOT_DIR/../../python"
     python setup.py install --user
   popd
 
 elif [[ "$LINT" == "1" ]]; then
   export PATH="$HOME/miniconda/bin:$PATH"
 
-  pushd "$ROOT_DIR/../python"
+  pushd "$ROOT_DIR/../../python"
     python setup.py install --user
   popd
 
diff --git a/.travis/install-with-cache-libs.sh b/ci/travis/install-with-cache-libs.sh
similarity index 65%
rename from .travis/install-with-cache-libs.sh
rename to ci/travis/install-with-cache-libs.sh
index f95a260c7e00..9758e7c89816 100755
--- a/.travis/install-with-cache-libs.sh
+++ b/ci/travis/install-with-cache-libs.sh
@@ -7,15 +7,15 @@ ROOT_DIR=$(cd "$(dirname "${BASH_SOURCE:-$0}")"; pwd)
 
 # Copy the libs to thirdparty/external_project_libs.
 # -n means do not build again.
-bash $ROOT_DIR/../thirdparty/scripts/collect_dependent_libs.sh -n
+bash $ROOT_DIR/../../thirdparty/scripts/collect_dependent_libs.sh -n
 
 # Import all the libs cached in local to environment variables.
-source $ROOT_DIR/../thirdparty/external_project_libs/resource.txt
+source $ROOT_DIR/../../thirdparty/external_project_libs/resource.txt
 
 # Clean the arrow building libs and use the cache libs.
-pushd $ROOT_DIR/../build/external/arrow/src/arrow_ep
+pushd $ROOT_DIR/../../build/external/arrow/src/arrow_ep
 echo "Clean Arrow building files."
-rm -rf $ROOT_DIR/../build/external/arrow/src/arrow_ep-build/
+rm -rf $ROOT_DIR/../../build/external/arrow/src/arrow_ep-build/
 popd
 
 # Rebuild ray with libs cache environment variables, which is fast.
diff --git a/test/travis-ci/install.sh b/ci/travis/install.sh
similarity index 100%
rename from test/travis-ci/install.sh
rename to ci/travis/install.sh
diff --git a/test/travis-ci/run_test.sh b/ci/travis/run_test.sh
similarity index 100%
rename from test/travis-ci/run_test.sh
rename to ci/travis/run_test.sh
diff --git a/.travis/test-wheels.sh b/ci/travis/test-wheels.sh
similarity index 85%
rename from .travis/test-wheels.sh
rename to ci/travis/test-wheels.sh
index f7870ea52d49..1f4e97e795b5 100755
--- a/.travis/test-wheels.sh
+++ b/ci/travis/test-wheels.sh
@@ -21,7 +21,7 @@ else
   exit 1
 fi
 
-TEST_SCRIPT=$ROOT_DIR/../test/microbenchmarks.py
+TEST_SCRIPT=$ROOT_DIR/../../python/ray/tests/test_microbenchmarks.py
 
 if [[ "$platform" == "linux" ]]; then
   # First test Python 2.7.
@@ -31,7 +31,7 @@ if [[ "$platform" == "linux" ]]; then
   bash miniconda2.sh -b -p $HOME/miniconda2
 
   # Find the right wheel by grepping for the Python version.
-  PYTHON_WHEEL=$(find $ROOT_DIR/../.whl -type f -maxdepth 1 -print | grep -m1 '27')
+  PYTHON_WHEEL=$(find $ROOT_DIR/../../.whl -type f -maxdepth 1 -print | grep -m1 '27')
 
   # Install the wheel.
   $HOME/miniconda2/bin/pip install $PYTHON_WHEEL
@@ -46,7 +46,7 @@ if [[ "$platform" == "linux" ]]; then
   bash miniconda3.sh -b -p $HOME/miniconda3
 
   # Find the right wheel by grepping for the Python version.
-  PYTHON_WHEEL=$(find $ROOT_DIR/../.whl -type f -maxdepth 1 -print | grep -m1 '36')
+  PYTHON_WHEEL=$(find $ROOT_DIR/../../.whl -type f -maxdepth 1 -print | grep -m1 '36')
 
   # Install the wheel.
   $HOME/miniconda3/bin/pip install $PYTHON_WHEEL
@@ -55,7 +55,7 @@ if [[ "$platform" == "linux" ]]; then
   $HOME/miniconda3/bin/python $TEST_SCRIPT
 
   # Check that the other wheels are present.
-  NUMBER_OF_WHEELS=$(ls -1q $ROOT_DIR/../.whl/*.whl | wc -l)
+  NUMBER_OF_WHEELS=$(ls -1q $ROOT_DIR/../../.whl/*.whl | wc -l)
   if [[ "$NUMBER_OF_WHEELS" != "5" ]]; then
     echo "Wrong number of wheels found."
     ls -l $ROOT_DIR/../.whl/
@@ -84,7 +84,7 @@ elif [[ "$platform" == "macosx" ]]; then
     PIP_CMD="$(dirname $PYTHON_EXE)/pip$PY_MM"
 
     # Find the appropriate wheel by grepping for the Python version.
-    PYTHON_WHEEL=$(find $ROOT_DIR/../.whl -type f -maxdepth 1 -print | grep -m1 "$PY_WHEEL_VERSION")
+    PYTHON_WHEEL=$(find $ROOT_DIR/../../.whl -type f -maxdepth 1 -print | grep -m1 "$PY_WHEEL_VERSION")
 
     # Install the wheel.
     $PIP_CMD install $PYTHON_WHEEL
diff --git a/.travis/upgrade-syn.sh b/ci/travis/upgrade-syn.sh
similarity index 100%
rename from .travis/upgrade-syn.sh
rename to ci/travis/upgrade-syn.sh
diff --git a/python/ray/rllib/train.py b/python/ray/rllib/train.py
index 3b043e1a944f..d08f7ef14a37 100755
--- a/python/ray/rllib/train.py
+++ b/python/ray/rllib/train.py
@@ -8,7 +8,7 @@
 import yaml
 
 import ray
-from ray.test.cluster_utils import Cluster
+from ray.tests.cluster_utils import Cluster
 from ray.tune.config_parser import make_parser
 from ray.tune.trial import resources_to_json
 from ray.tune.tune import _make_scheduler, run_experiments
diff --git a/python/ray/test/__init__.py b/python/ray/tests/__init__.py
similarity index 100%
rename from python/ray/test/__init__.py
rename to python/ray/tests/__init__.py
diff --git a/python/ray/test/cluster_utils.py b/python/ray/tests/cluster_utils.py
similarity index 100%
rename from python/ray/test/cluster_utils.py
rename to python/ray/tests/cluster_utils.py
diff --git a/test/actor_test.py b/python/ray/tests/test_actor.py
similarity index 99%
rename from test/actor_test.py
rename to python/ray/tests/test_actor.py
index 16642540c117..fd5ce7bfbc57 100644
--- a/test/actor_test.py
+++ b/python/ray/tests/test_actor.py
@@ -14,8 +14,8 @@
 
 import ray
 import ray.ray_constants as ray_constants
-import ray.test.test_utils
-import ray.test.cluster_utils
+import ray.tests.utils
+import ray.tests.cluster_utils
 
 
 @pytest.fixture
@@ -44,7 +44,7 @@ def shutdown_only():
 
 @pytest.fixture()
 def ray_start_cluster():
-    cluster = ray.test.cluster_utils.Cluster()
+    cluster = ray.tests.cluster_utils.Cluster()
     yield cluster
 
     # The code after the yield will run as teardown code.
@@ -58,7 +58,7 @@ def two_node_cluster():
         "initial_reconstruction_timeout_milliseconds": 200,
         "num_heartbeats_timeout": 10,
     })
-    cluster = ray.test.cluster_utils.Cluster()
+    cluster = ray.tests.cluster_utils.Cluster()
     for _ in range(2):
         remote_node = cluster.add_node(
             num_cpus=1, _internal_config=internal_config)
@@ -73,7 +73,7 @@ def two_node_cluster():
 @pytest.fixture
 def head_node_cluster(request):
     timeout = getattr(request, "param", 200)
-    cluster = ray.test.cluster_utils.Cluster(
+    cluster = ray.tests.cluster_utils.Cluster(
         initialize_head=True,
         connect=True,
         head_node_args={
@@ -418,7 +418,7 @@ def __init__(self):
     assert len(actor_keys) == 1
     actor_class_info = r.hgetall(actor_keys[0])
     assert actor_class_info[b"class_name"] == b"Foo"
-    assert actor_class_info[b"module"] == b"actor_test"
+    assert actor_class_info[b"module"] == b"ray.tests.test_actor"
 
 
 def test_multiple_return_values(ray_start_regular):
@@ -483,13 +483,13 @@ def getpid(self):
     a = Actor.remote()
     pid = ray.get(a.getpid.remote())
     a = None
-    ray.test.test_utils.wait_for_pid_to_exit(pid)
+    ray.tests.utils.wait_for_pid_to_exit(pid)
 
     actors = [Actor.remote() for _ in range(10)]
     pids = ray.get([a.getpid.remote() for a in actors])
     a = None
     actors = None
-    [ray.test.test_utils.wait_for_pid_to_exit(pid) for pid in pids]
+    [ray.tests.utils.wait_for_pid_to_exit(pid) for pid in pids]
 
     @ray.remote
     class Actor(object):
diff --git a/test/array_test.py b/python/ray/tests/test_array.py
similarity index 99%
rename from test/array_test.py
rename to python/ray/tests/test_array.py
index 7ed73811991d..1ddd9d45c69a 100644
--- a/test/array_test.py
+++ b/python/ray/tests/test_array.py
@@ -10,7 +10,7 @@
 import ray
 import ray.experimental.array.remote as ra
 import ray.experimental.array.distributed as da
-import ray.test.cluster_utils
+import ray.tests.cluster_utils
 
 if sys.version_info >= (3, 0):
     from importlib import reload
@@ -75,7 +75,7 @@ def ray_start_two_nodes():
     ]:
         reload(module)
     # Start the Ray processes.
-    cluster = ray.test.cluster_utils.Cluster()
+    cluster = ray.tests.cluster_utils.Cluster()
     for _ in range(2):
         cluster.add_node(num_cpus=10)
     ray.init(redis_address=cluster.redis_address)
diff --git a/test/autoscaler_test.py b/python/ray/tests/test_autoscaler.py
similarity index 100%
rename from test/autoscaler_test.py
rename to python/ray/tests/test_autoscaler.py
diff --git a/test/runtest.py b/python/ray/tests/test_basic.py
similarity index 99%
rename from test/runtest.py
rename to python/ray/tests/test_basic.py
index 1ad95233bc52..370dfc018809 100644
--- a/test/runtest.py
+++ b/python/ray/tests/test_basic.py
@@ -25,8 +25,8 @@
 import pytest
 
 import ray
-import ray.test.cluster_utils
-import ray.test.test_utils
+import ray.tests.cluster_utils
+import ray.tests.utils
 from ray.utils import _random_string
 
 logger = logging.getLogger(__name__)
@@ -1139,7 +1139,7 @@ def f():
 
 @pytest.fixture()
 def ray_start_cluster():
-    cluster = ray.test.cluster_utils.Cluster()
+    cluster = ray.tests.cluster_utils.Cluster()
     yield cluster
 
     # The code after the yield will run as teardown code.
@@ -2281,7 +2281,7 @@ def f():
         return os.getpid()
 
     pid = ray.get(f.remote())
-    ray.test.test_utils.wait_for_pid_to_exit(pid)
+    ray.tests.utils.wait_for_pid_to_exit(pid)
 
     @ray.remote(max_calls=2)
     def f():
@@ -2290,7 +2290,7 @@ def f():
     pid1 = ray.get(f.remote())
     pid2 = ray.get(f.remote())
     assert pid1 == pid2
-    ray.test.test_utils.wait_for_pid_to_exit(pid1)
+    ray.tests.utils.wait_for_pid_to_exit(pid1)
 
 
 def attempt_to_load_balance(remote_function,
@@ -2443,9 +2443,9 @@ def f(*xs):
     assert task_spec["DriverID"] == driver_id
     assert task_spec["ReturnObjectIDs"] == [result_id]
     function_table_entry = function_table[task_spec["FunctionID"]]
-    assert function_table_entry["Name"] == "runtest.f"
+    assert function_table_entry["Name"] == "ray.tests.test_basic.f"
     assert function_table_entry["DriverID"] == driver_id
-    assert function_table_entry["Module"] == "runtest"
+    assert function_table_entry["Module"] == "ray.tests.test_basic"
 
     assert task_table[task_id] == ray.global_state.task_table(task_id)
 
@@ -2722,7 +2722,8 @@ def f(self):
 
     @ray.remote
     def unique_1():
-        assert setproctitle.getproctitle() == "ray_worker:runtest.unique_1()"
+        assert setproctitle.getproctitle(
+        ) == "ray_worker:ray.tests.test_basic.unique_1()"
 
     actor = UniqueName.remote()
     ray.get(actor.f.remote())
diff --git a/test/component_failures_test.py b/python/ray/tests/test_component_failures.py
similarity index 99%
rename from test/component_failures_test.py
rename to python/ray/tests/test_component_failures.py
index 6279f258b6c9..4d9bbd80d5e7 100644
--- a/test/component_failures_test.py
+++ b/python/ray/tests/test_component_failures.py
@@ -13,8 +13,8 @@
 
 import ray
 import ray.ray_constants as ray_constants
-from ray.test.cluster_utils import Cluster
-from ray.test.test_utils import run_string_as_driver_nonblocking
+from ray.tests.cluster_utils import Cluster
+from ray.tests.utils import run_string_as_driver_nonblocking
 
 
 @pytest.fixture
diff --git a/test/credis_test.py b/python/ray/tests/test_credis.py
similarity index 100%
rename from test/credis_test.py
rename to python/ray/tests/test_credis.py
diff --git a/test/cython_test.py b/python/ray/tests/test_cython.py
similarity index 100%
rename from test/cython_test.py
rename to python/ray/tests/test_cython.py
diff --git a/test/debug_tools_test.py b/python/ray/tests/test_debug_tools.py
similarity index 100%
rename from test/debug_tools_test.py
rename to python/ray/tests/test_debug_tools.py
diff --git a/test/failure_test.py b/python/ray/tests/test_failure.py
similarity index 99%
rename from test/failure_test.py
rename to python/ray/tests/test_failure.py
index 543cb714060a..19dfbbc6cb45 100644
--- a/test/failure_test.py
+++ b/python/ray/tests/test_failure.py
@@ -16,7 +16,7 @@
 import ray
 import ray.ray_constants as ray_constants
 from ray.utils import _random_string
-from ray.test.cluster_utils import Cluster
+from ray.tests.cluster_utils import Cluster
 
 
 def relevant_errors(error_type):
diff --git a/python/ray/test/test_global_state.py b/python/ray/tests/test_global_state.py
similarity index 98%
rename from python/ray/test/test_global_state.py
rename to python/ray/tests/test_global_state.py
index 0a627ca03b1f..11920a21c6a4 100644
--- a/python/ray/test/test_global_state.py
+++ b/python/ray/tests/test_global_state.py
@@ -11,7 +11,7 @@
 import time
 
 import ray
-from ray.test.cluster_utils import Cluster
+from ray.tests.cluster_utils import Cluster
 
 
 @pytest.fixture
diff --git a/test/microbenchmarks.py b/python/ray/tests/test_microbenchmarks.py
similarity index 100%
rename from test/microbenchmarks.py
rename to python/ray/tests/test_microbenchmarks.py
diff --git a/test/mini_test.py b/python/ray/tests/test_mini.py
similarity index 100%
rename from test/mini_test.py
rename to python/ray/tests/test_mini.py
diff --git a/python/ray/test/test_modin.py b/python/ray/tests/test_modin.py
similarity index 100%
rename from python/ray/test/test_modin.py
rename to python/ray/tests/test_modin.py
diff --git a/test/monitor_test.py b/python/ray/tests/test_monitors.py
similarity index 98%
rename from test/monitor_test.py
rename to python/ray/tests/test_monitors.py
index 248aa6fb0e31..486d8ca022ae 100644
--- a/test/monitor_test.py
+++ b/python/ray/tests/test_monitors.py
@@ -10,7 +10,7 @@
 
 import ray
 
-from ray.test.test_utils import run_and_get_output
+from ray.tests.utils import run_and_get_output
 
 
 def _test_cleanup_on_driver_exit(num_redis_shards):
diff --git a/test/multi_node_test.py b/python/ray/tests/test_multi_node.py
similarity index 98%
rename from test/multi_node_test.py
rename to python/ray/tests/test_multi_node.py
index 239789d13e50..719e473808e5 100644
--- a/test/multi_node_test.py
+++ b/python/ray/tests/test_multi_node.py
@@ -8,8 +8,8 @@
 import time
 
 import ray
-from ray.test.test_utils import (run_and_get_output, run_string_as_driver,
-                                 run_string_as_driver_nonblocking)
+from ray.tests.utils import (run_and_get_output, run_string_as_driver,
+                             run_string_as_driver_nonblocking)
 
 
 @pytest.fixture
diff --git a/test/multi_node_test_2.py b/python/ray/tests/test_multi_node_2.py
similarity index 98%
rename from test/multi_node_test_2.py
rename to python/ray/tests/test_multi_node_2.py
index 20a9b6d000b4..28e93050c535 100644
--- a/test/multi_node_test_2.py
+++ b/python/ray/tests/test_multi_node_2.py
@@ -9,7 +9,7 @@
 
 import ray
 import ray.ray_constants as ray_constants
-from ray.test.cluster_utils import Cluster
+from ray.tests.cluster_utils import Cluster
 
 logger = logging.getLogger(__name__)
 
diff --git a/test/node_manager_test.py b/python/ray/tests/test_node_manager.py
similarity index 94%
rename from test/node_manager_test.py
rename to python/ray/tests/test_node_manager.py
index 2f18d7b7308e..5a6b71f18fad 100644
--- a/test/node_manager_test.py
+++ b/python/ray/tests/test_node_manager.py
@@ -5,8 +5,8 @@
 import pytest
 
 import ray
-from ray.test.cluster_utils import Cluster
-from ray.test.test_utils import run_string_as_driver
+from ray.tests.cluster_utils import Cluster
+from ray.tests.utils import run_string_as_driver
 
 
 @pytest.fixture()
diff --git a/test/object_manager_test.py b/python/ray/tests/test_object_manager.py
similarity index 99%
rename from test/object_manager_test.py
rename to python/ray/tests/test_object_manager.py
index b51f54d1a803..757be18979a3 100644
--- a/test/object_manager_test.py
+++ b/python/ray/tests/test_object_manager.py
@@ -11,7 +11,7 @@
 import warnings
 
 import ray
-from ray.test.cluster_utils import Cluster
+from ray.tests.cluster_utils import Cluster
 
 if (multiprocessing.cpu_count() < 40
         or ray.utils.get_system_memory() < 50 * 10**9):
diff --git a/python/ray/test/test_queue.py b/python/ray/tests/test_queue.py
similarity index 100%
rename from python/ray/test/test_queue.py
rename to python/ray/tests/test_queue.py
diff --git a/python/ray/test/test_ray_init.py b/python/ray/tests/test_ray_init.py
similarity index 97%
rename from python/ray/test/test_ray_init.py
rename to python/ray/tests/test_ray_init.py
index 2b17ce35ef28..ef50b013077a 100644
--- a/python/ray/test/test_ray_init.py
+++ b/python/ray/tests/test_ray_init.py
@@ -7,7 +7,7 @@
 import redis
 
 import ray
-from ray.test.cluster_utils import Cluster
+from ray.tests.cluster_utils import Cluster
 
 
 @pytest.fixture
diff --git a/test/recursion_test.py b/python/ray/tests/test_recursion.py
similarity index 74%
rename from test/recursion_test.py
rename to python/ray/tests/test_recursion.py
index c3f7db59af03..0b3b68351cee 100644
--- a/test/recursion_test.py
+++ b/python/ray/tests/test_recursion.py
@@ -6,9 +6,21 @@
 from __future__ import division
 from __future__ import print_function
 
+import pytest
+
 import ray
 
-ray.init(num_cpus=1)
+
+@pytest.fixture
+def ray_start():
+    # Start ray instance
+    ray.init(num_cpus=1)
+
+    # Run test using this fixture
+    yield None
+
+    # Shutdown ray instance
+    ray.shutdown()
 
 
 @ray.remote
@@ -18,7 +30,7 @@ def factorial(n):
     return n * ray.get(factorial.remote(n - 1))
 
 
-def test_recursion():
+def test_recursion(ray_start):
     assert ray.get(factorial.remote(0)) == 1
     assert ray.get(factorial.remote(1)) == 1
     assert ray.get(factorial.remote(2)) == 2
diff --git a/test/test_signal.py b/python/ray/tests/test_signal.py
similarity index 100%
rename from test/test_signal.py
rename to python/ray/tests/test_signal.py
diff --git a/test/stress_tests.py b/python/ray/tests/test_stress.py
similarity index 99%
rename from test/stress_tests.py
rename to python/ray/tests/test_stress.py
index 0cf712a1a93d..cdb5684e30cb 100644
--- a/test/stress_tests.py
+++ b/python/ray/tests/test_stress.py
@@ -9,7 +9,7 @@
 import time
 
 import ray
-from ray.test.cluster_utils import Cluster
+from ray.tests.cluster_utils import Cluster
 import ray.ray_constants as ray_constants
 
 
diff --git a/test/tempfile_test.py b/python/ray/tests/test_tempfile.py
similarity index 100%
rename from test/tempfile_test.py
rename to python/ray/tests/test_tempfile.py
diff --git a/test/tensorflow_test.py b/python/ray/tests/test_tensorflow.py
similarity index 100%
rename from test/tensorflow_test.py
rename to python/ray/tests/test_tensorflow.py
diff --git a/python/ray/test/test_utils.py b/python/ray/tests/utils.py
similarity index 100%
rename from python/ray/test/test_utils.py
rename to python/ray/tests/utils.py
diff --git a/python/ray/tune/test/cluster_tests.py b/python/ray/tune/test/cluster_tests.py
index 05a597e06e11..96a66dbef6a6 100644
--- a/python/ray/tune/test/cluster_tests.py
+++ b/python/ray/tune/test/cluster_tests.py
@@ -16,8 +16,8 @@
 import ray
 from ray import tune
 from ray.rllib import _register_all
-from ray.test.cluster_utils import Cluster
-from ray.test.test_utils import run_string_as_driver_nonblocking
+from ray.tests.cluster_utils import Cluster
+from ray.tests.utils import run_string_as_driver_nonblocking
 from ray.tune.error import TuneError
 from ray.tune.experiment import Experiment
 from ray.tune.trial import Trial