diff --git a/python/build-wheel-macos.sh b/python/build-wheel-macos.sh
index 48d646bcdedc..e29c0469f274 100755
--- a/python/build-wheel-macos.sh
+++ b/python/build-wheel-macos.sh
@@ -14,24 +14,36 @@ MACPYTHON_PY_PREFIX=/Library/Frameworks/Python.framework/Versions
 DOWNLOAD_DIR=python_downloads
 
 NODE_VERSION="14"
+# BONSAI changes begin - python version 3.6.1 removed
 PY_VERSIONS=("3.7.0"
              "3.8.2")
+# BONSAI changes end
+# BONSAI changes begin - "python-3.6.1-macosx10.6.pkg" removed
 PY_INSTS=("python-3.7.0-macosx10.6.pkg"
           "python-3.8.2-macosx10.9.pkg")
+# BONSAI changes end
+# BONSAI changes begin - Python 3.6 removed
 PY_MMS=("3.7"
         "3.8")
+# BONSAI changes end
 
 # The minimum supported numpy version is 1.14, see
 # https://issues.apache.org/jira/browse/ARROW-3141
+# BONSAI changes begin
+# NUMPY_VERSIONS=("1.14.5"
+#                 "1.14.5"
+#                 "1.14.5")
 NUMPY_VERSIONS=("1.19.2"
                 "1.19.2")
+# BONSAI changes end
+
 
 ./ci/travis/install-bazel.sh
 
 mkdir -p $DOWNLOAD_DIR
 mkdir -p .whl
-
-# In azure pipelines or github acions, we don't need to install node
+# BONSAI changes begin - added check if Node already installed and don't install in that case
+# In azure pipelines or github actions, we don't need to install node
 if [ -x "$(command -v npm)" ]; then
   echo "Node already installed"
   npm -v
@@ -41,6 +53,7 @@ else
   nvm install $NODE_VERSION
   nvm use node
 fi
+# BONSAI changes end
 
 # Build the dashboard so its static assets can be included in the wheel.
 # TODO(mfitton): switch this back when deleting old dashboard code.
@@ -58,7 +71,9 @@ for ((i=0; i<${#PY_VERSIONS[@]}; ++i)); do
   # The -f flag is passed twice to also run git clean in the arrow subdirectory.
   # The -d flag removes directories. The -x flag ignores the .gitignore file,
   # and the -e flag ensures that we don't remove the .whl directory.
+  # BONSAI changes begin - added code
   git config --global --add safe.directory /ray
+  # BONSAI changes end
   git clean -f -f -x -d -e .whl -e $DOWNLOAD_DIR -e python/ray/new_dashboard/client -e dashboard/client
 
   # Install Python.
@@ -76,7 +91,9 @@ for ((i=0; i<${#PY_VERSIONS[@]}; ++i)); do
 
   pushd python
     # Setuptools on CentOS is too old to install arrow 0.9.0, therefore we upgrade.
-    $PIP_CMD install --upgrade setuptools
+    # BONSAI changes begin - pinned setuptools to 41.0.0
+    $PIP_CMD install setuptools==41.0.0
+    # BONSAI changes end
     # Install setuptools_scm because otherwise when building the wheel for
     # Python 3.6, we see an error.
     $PIP_CMD install -q setuptools_scm==3.1.0
@@ -94,7 +111,9 @@ for ((i=0; i<${#PY_VERSIONS[@]}; ++i)); do
     fi
     # Add the correct Python to the path and build the wheel. This is only
     # needed so that the installation finds the cython executable.
+    # BONSAI changes begin -     PATH=$MACPYTHON_PY_PREFIX/$PY_MM/bin:$PATH $PYTHON_EXE setup.py bdist_wheel
     MACOSX_DEPLOYMENT_TARGET=10.15 PATH=$MACPYTHON_PY_PREFIX/$PY_MM/bin:$PATH $PYTHON_EXE setup.py bdist_wheel
+    # BONSAI changes end
     mv dist/*.whl ../.whl/
   popd
 done
diff --git a/python/build-wheel-manylinux2014.sh b/python/build-wheel-manylinux2014.sh
index 930f551a2f4b..c022c1e68b24 100755
--- a/python/build-wheel-manylinux2014.sh
+++ b/python/build-wheel-manylinux2014.sh
@@ -12,20 +12,31 @@ EOF
 chmod +x /usr/bin/nproc
 
 NODE_VERSION="14"
+# BONSAI changes begin - removed "cp36-cp36m"
 PYTHONS=("cp37-cp37m"
          "cp38-cp38")
+# BONSAI changes end
+
 
 # The minimum supported numpy version is 1.14, see
 # https://issues.apache.org/jira/browse/ARROW-3141
+# BONSAI changes begin
+# NUMPY_VERSIONS=("1.14.5"
+#                 "1.14.5"
+#                 "1.14.5")
 NUMPY_VERSIONS=("1.19.2"
                 "1.19.2")
-
+# BONSAI changes end
+# BONSAI changes begin - added code
 yum -y update
+# BONSAI changes end
 yum -y install unzip zip sudo
 yum -y install java-1.8.0-openjdk java-1.8.0-openjdk-devel xz
 yum -y install openssl
+# BONSAI changes begin - added code
 yum -y install curl
 yum install -y gcc-c++ make
+# BONSAI changes end
 
 java -version
 java_bin=$(readlink -f "$(command -v java)")
@@ -47,6 +58,7 @@ if [[ -n "${RAY_INSTALL_JAVA:-}" ]]; then
   unset RAY_INSTALL_JAVA
 fi
 
+# BONSAI changes begin
 # In azure pipelines or github acions, we don't need to install node
 if [ -x "$(command -v npm)" ]; then
   echo "Node already installed"
@@ -74,16 +86,19 @@ else
   # node –v
   npm -v
 fi
+# BONSAI changes end
 
 # Build the dashboard so its static assets can be included in the wheel.
 # TODO(mfitton): switch this back when deleting old dashboard code.
 pushd python/ray/new_dashboard/client
+  # BONSAI changes begin
   pwd
   echo "Run npm ci"
   npm ci
   echo "Run npm run build"
   npm run build
   echo "Done running npm run build"
+  # BONSAI changes end
 popd
 set -x
 
@@ -96,7 +111,9 @@ for ((i=0; i<${#PYTHONS[@]}; ++i)); do
   # The -d flag removes directories. The -x flag ignores the .gitignore file,
   # and the -e flag ensures that we don't remove the .whl directory, the
   # dashboard directory and jars directory.
+  # BONSAI changes begin - added code
   git config --global --add safe.directory /ray
+  # BONSAI changes end
   git clean -f -f -x -d -e .whl -e python/ray/new_dashboard/client -e dashboard/client -e python/ray/jars
 
   pushd python
@@ -110,9 +127,12 @@ for ((i=0; i<${#PYTHONS[@]}; ++i)); do
       echo "TRAVIS_COMMIT variable not set - required to populated ray.__commit__."
       exit 1
     fi
-
+    # BONSAI changes begin
+    # PATH=/opt/python/${PYTHON}/bin:/root/bazel-3.2.0/output:$PATH \
+    #    /opt/python/"${PYTHON}"/bin/python setup.py bdist_wheel
     PATH=/opt/python/${PYTHON}/bin:/root/bazel-3.4.1/output:$PATH \
     /opt/python/"${PYTHON}"/bin/python setup.py bdist_wheel
+    # BONSAI changes end
     # In the future, run auditwheel here.
     mv dist/*.whl ../.whl/
   popd
@@ -127,5 +147,7 @@ for path in .whl/*.whl; do
 done
 
 # Clean the build output so later operations is on a clean directory.
+# BONSAI changes begin - added code
 git config --global --add safe.directory /ray
+# BONSAI changes end
 git clean -f -f -x -d -e .whl -e python/ray/dashboard/client
diff --git a/python/ray/_private/memory_monitor.py b/python/ray/_private/memory_monitor.py
index 5f0789061db9..0628908c6d46 100644
--- a/python/ray/_private/memory_monitor.py
+++ b/python/ray/_private/memory_monitor.py
@@ -38,6 +38,7 @@ def get_message(used_gb, total_gb, threshold):
         pids = psutil.pids()
         proc_stats = []
         for pid in pids:
+            # BONSAI changes begin
             try:
                 proc = psutil.Process(pid)
                 proc_stats.append((get_rss(proc.memory_info()), pid,
@@ -47,6 +48,7 @@ def get_message(used_gb, total_gb, threshold):
                 # respect 100 character limit
                 msg = msg[:100]
                 proc_stats.append(0, pid, msg)
+            # BONSAI changes end
         proc_str = "PID\tMEM\tCOMMAND"
         for rss, pid, cmdline in sorted(proc_stats, reverse=True)[:10]:
             proc_str += "\n{}\t{}GiB\t{}".format(
diff --git a/python/ray/test_utils.py b/python/ray/test_utils.py
index 112892103397..78ac664dcce6 100644
--- a/python/ray/test_utils.py
+++ b/python/ray/test_utils.py
@@ -193,8 +193,12 @@ def run_string_as_driver(driver_script):
         stdout=subprocess.PIPE,
         stderr=subprocess.STDOUT)
     with proc:
+        # BONSAI changes begin
+        # output = proc.communicate(driver_script.encode("ascii"))[0]
+        #   if proc.returncode:
         output, error = proc.communicate(driver_script.encode("ascii"))
         if proc.returncode or error is not None:
+        # BONSAI changes end
             print(ray._private.utils.decode(output))
             raise subprocess.CalledProcessError(proc.returncode, proc.args,
                                                 output, proc.stderr)
diff --git a/python/ray/tests/BUILD b/python/ray/tests/BUILD
index 8f681ccd6e9c..80c75ce6051d 100644
--- a/python/ray/tests/BUILD
+++ b/python/ray/tests/BUILD
@@ -5,9 +5,11 @@ SRCS = [] + select({
         # TODO(mehrdadn): This should be added for all platforms once resulting errors are fixed
         "**/conftest.py",
     ]),
+    # BONSAI changes begin - "//conditions:default": [],
     "//conditions:default": glob([
         "**/conftest.py",
     ]),
+    # BONSAI changes end
 })
 
 py_test_module_list(
@@ -40,10 +42,14 @@ py_test_module_list(
     "test_global_gc.py",
     "test_mldataset.py",
   ],
+  # BONSAI changes begin - size = "medium",
   size = "large",
+  # BONSAI changes end
   extra_srcs = SRCS,
   tags = ["exclusive", "medium_size_python_tests_a_to_j"],
+  # BONSAI changes begin - added code
   data = glob(["mnist_784_100_samples.pkl"]),
+  # BONSAI changes end
   deps = ["//:ray_lib"],
 )
 
@@ -114,6 +120,7 @@ py_test_module_list(
   extra_srcs = SRCS,
   tags = ["exclusive"],
   deps = ["//:ray_lib"],
+  # BONSAI changes begin - added code
   data = [
     "additional_property.yaml",
     "//:python/ray/autoscaler/aws/example-full.yaml",
@@ -125,6 +132,7 @@ py_test_module_list(
     "//:python/ray/autoscaler/aws/example-minimal.yaml",
     "//:python/ray/autoscaler/aws/example-multi-node-type.yaml",
   ],
+  # BONSAI changes end
 )
 
 py_test_module_list(
@@ -139,11 +147,12 @@ py_test_module_list(
   deps = ["//:ray_lib"],
   data = glob(["test_cli_patterns/*.*"], allow_empty=False),
 )
-
+# BONSAI changes begin - added code
 TEST_RUNTIME_ENV_SOURCES = [
     "//:python/ray/experimental/packaging/example_pkg/my_pkg/stubs.py",
     "//:python/ray/experimental/packaging/example_pkg/my_pkg/impl/__init__.py",
 ]
+# BONSAI changes end
 
 py_test_module_list(
   files = [
@@ -152,9 +161,13 @@ py_test_module_list(
     "test_runtime_env.py",
   ],
   size = "large",
+  # BONSAI changes begin -   extra_srcs = SRCS,
   extra_srcs = SRCS + TEST_RUNTIME_ENV_SOURCES,
+  # BONSAI changes end
   deps = ["//:ray_lib"],
+  # BONSAI changes begin - added code
   data = ["//:python/ray/experimental/packaging/example_pkg/ray_pkg.yaml"],
+  # BONSAI changes end
 )
 
 # TODO(barakmich): aws/ might want its own buildfile, or
diff --git a/python/ray/tests/test_autoscaler_yaml.py b/python/ray/tests/test_autoscaler_yaml.py
index 810316eb4613..37ad518509c0 100644
--- a/python/ray/tests/test_autoscaler_yaml.py
+++ b/python/ray/tests/test_autoscaler_yaml.py
@@ -178,8 +178,12 @@ def testValidateDefaultConfigAWSMultiNodeTypes(self):
                     "Config did not pass multi node types auto fill test!")
 
     def testValidateNetworkConfig(self):
+        # BONSAI changes begin
+        # web_yaml = "https://raw.githubusercontent.com/ray-project/ray/" \
+        #   "master/python/ray/autoscaler/aws/example-full.yaml"
         web_yaml = "https://raw.githubusercontent.com/BonsaiAI/ray/" \
             "master/python/ray/autoscaler/aws/example-full.yaml"
+        # BONSAI changes end
         response = urllib.request.urlopen(web_yaml, timeout=5)
         content = response.read()
         with tempfile.TemporaryFile() as f:
diff --git a/python/ray/tests/test_basic.py b/python/ray/tests/test_basic.py
index 1bc11ab49b06..4d2dd20cfb62 100644
--- a/python/ray/tests/test_basic.py
+++ b/python/ray/tests/test_basic.py
@@ -6,7 +6,9 @@
 import time
 
 import numpy as np
+# BONSAI changes begin - added import
 import platform
+# BONSAI changes end
 import pytest
 
 import ray.cluster_utils
@@ -251,9 +253,11 @@ def foo():
             }).remote())
 
     to_check = ["CPU", "GPU", "memory", "custom1"]
+    # BONSAI changes begin - added if clause
     if "linux" in platform.platform().lower():
         # TODO(Edi): we have to fix the agents used by the pipelines to allow this
         #            kind of tests in mac.
+    # BONSAI changes end
         for key in to_check:
             assert without_options[key] != with_options[key], key
         assert without_options != with_options
diff --git a/python/ray/tests/test_client.py b/python/ray/tests/test_client.py
index 555cebd2f06f..675983e5ab76 100644
--- a/python/ray/tests/test_client.py
+++ b/python/ray/tests/test_client.py
@@ -338,6 +338,7 @@ def print_on_stderr_and_stdout(s):
         time.sleep(1)
         print_on_stderr_and_stdout.remote("Hello world")
         time.sleep(1)
+        # BONSAI changes begin
         # NOTE (Ruofan): The test is flakey and Ray fixes this test in Ray 1+:
         # https://github.com/ray-project/ray/pull/19232/files
         # We apply the fix here. Later once we updated the Ray version, we SHOULD
@@ -347,6 +348,7 @@ def print_on_stderr_and_stdout(s):
             if "Hello world" in msg:
                 num_hello += 1
         assert num_hello == 2, f"Invalid logs: {log_msgs}"
+        # BONSAI changes end
 
 
 @pytest.mark.skipif(sys.platform == "win32", reason="Failing on Windows.")
diff --git a/python/ray/tests/test_debug_tools.py b/python/ray/tests/test_debug_tools.py
index 6bb29c6bd87b..31a78406a849 100644
--- a/python/ray/tests/test_debug_tools.py
+++ b/python/ray/tests/test_debug_tools.py
@@ -25,8 +25,10 @@ def ray_gdb_start():
 
 @pytest.mark.skipif(
     sys.platform != "linux" and sys.platform != "linux2",
+    # BONSAI changes begin - added decorator
     reason="TODO(Edi): This test requires Linux.")
 # @pytest.mark.skip(reason="Too flaky.")
+# BONSAI changes end
 def test_raylet_gdb(ray_gdb_start):
     # ray_gdb_start yields the expected process name
     ray.init(num_cpus=1)
diff --git a/python/ray/tests/test_iter.py b/python/ray/tests/test_iter.py
index aae0283cda58..46a2993d7ccc 100644
--- a/python/ray/tests/test_iter.py
+++ b/python/ray/tests/test_iter.py
@@ -1,18 +1,24 @@
+# BONSAI changes begin - added import
 from __future__ import generator_stop
+# BONSAI changes end
 import sys
 import time
 import collections
 from collections import Counter
+# BONSAI changes begin - added import
 from typing import Iterator, Any, Callable, Generator, List, Optional
+# BONSAI changes end
 
 import pytest
 
 import ray
+# BONSAI changes begin - added was_cause_by_step_iteration import
 from ray.util.iter import from_items, from_iterators, from_range, \
     from_actors, ParallelIteratorWorker, LocalIterator, was_cause_by_stop_iteration
+# BONSAI changes end
 from ray.test_utils import Semaphore
 
-
+# BONSAI changes begin - added Class
 class _ReusableGenerator(Iterator[int]):
     def __init__(self, sequence: List[int], sleep: Optional[float] = None):
         self.sequence = sequence
@@ -44,8 +50,7 @@ def __next__(self) -> int:
                 self.wait = True
                 self.last_full_epoch_time = time.perf_counter()
             raise StopIteration
-
-
+# BONSAI changes end
 
 def test_select_shards(ray_start_regular_shared):
     it = from_items([1, 2, 3, 4], num_shards=4)
@@ -167,10 +172,13 @@ def verify_metrics(x):
     it12 = it1.union(it2, deterministic=True)
     it123 = it12.union(it3, deterministic=True)
     out = it123.for_each(verify_metrics)
+    # BONSAI changes begin
+    # assert out.take(20) == [1, 1, 1, 2, 2, 3, 2, 4, 3, 3, 4, 4]
     taken = out.take(20)
     expected = [1, 1, 1, 2, 2, 3, 2, 4, 3, 3, 4, 4]
     assert len(taken) == len(expected)
     assert taken == expected
+    # BONSAI changes end
 
 
 def test_from_items(ray_start_regular_shared):
@@ -434,7 +442,7 @@ def test_gather_async_optimized(ray_start_regular_shared):
     it = it.gather_async(batch_ms=100, num_async=4)
     assert sorted(it) == list(range(100))
 
-
+# BONSAI changes begin - added function
 def test_gather_async_empty(ray_start_regular_shared):
     # Finite sequences
     it = from_iterators([range(10), []], is_infinite_sequence=False)
@@ -483,7 +491,7 @@ def test_gather_async_empty(ray_start_regular_shared):
                 raise ex
     assert len(sorted(collected)) == 0
     assert [c > 0 for c in attempts_collect_counts].count(True) == 0
-
+# BONSAI changes end
 
 def test_get_shard_optimized(ray_start_regular_shared):
     it = from_range(6, num_shards=3)
@@ -532,13 +540,16 @@ def test_gather_async_optimized_benchmark(ray_start_regular_shared):
 
 
 def test_batch_across_shards(ray_start_regular_shared):
+    # BONSAI changes begin - is_infinite_sequence=False added
     # Finite sequences
     it = from_iterators([[0, 1], [2, 3]], is_infinite_sequence=False)
+    # BONSAI changes end
     it = it.batch_across_shards()
     assert (
         repr(it) == "LocalIterator[ParallelIterator[from_iterators[shards=2]]"
         ".batch_across_shards()]")
     assert sorted(it) == [[0, 2], [1, 3]]
+    # BONSAI changes begin - added code
     # Infinite sequences
     it = from_iterators([_ReusableGenerator([0, 1]), _ReusableGenerator([2, 3])],
                         is_infinite_sequence=True)
@@ -560,8 +571,9 @@ def test_batch_across_shards(ray_start_regular_shared):
     for l in [[0, 2], [1, 3]]:
         assert collected.count(l) >= 2
     assert [c > 0 for c in attempts_collect_counts].count(True) > 1
+    # BONSAI changes begin - added end
 
-
+# BONSAI changes begin - added test
 def test_batch_across_unbalanced_shards(ray_start_regular_shared):
     # Finite sequences
     it = from_iterators([[0, 1, 2], [3, 4, 5, 6]], is_infinite_sequence=False)
@@ -592,6 +604,7 @@ def test_batch_across_unbalanced_shards(ray_start_regular_shared):
     for x in [0, 1, 2, 3, 4, 5, 6]:
         assert flat_collected.count(x) >= 2
     assert [c > 0 for c in attempts_collect_counts].count(True) > 1
+# BONSAI changes end
 
 
 def test_remote(ray_start_regular_shared):
@@ -645,14 +658,17 @@ def gen_slow():
             time.sleep(0.3)
             print("PRODUCE SLOW", i)
             yield i
-
+    # BONSAI changes begin - added  is_infinite_sequence=False to from_iterators
     it1 = from_iterators([gen_fast],
                          is_infinite_sequence=False).for_each(lambda x: ("fast", x))
     it2 = from_iterators([gen_slow],
                          is_infinite_sequence=False).for_each(lambda x: ("slow", x))
+    # BONSAI changes end
+
     it = it1.union(it2)
     results = list(it.gather_async())
     assert all(x[0] == "slow" for x in results[-3:]), results
+    # BONSAI changes begin - added code
     # Infinite sequences
     it1 = from_iterators([_ReusableGenerator(list(range(10)), 0.05)],
                          is_infinite_sequence=True).for_each(lambda x: ("fast", x))
@@ -679,6 +695,7 @@ def gen_slow():
     assert slow_count <= (len(collected_categories)/2)
     assert (len(collected_categories) - slow_count) >= 8
     assert [c > 0 for c in attempts_collect_counts].count(True) > 1
+    # BONSAI changes end
 
 
 def test_union_local_async(ray_start_regular_shared):
@@ -694,17 +711,19 @@ def gen_slow():
             time.sleep(0.3)
             print("PRODUCE SLOW", i)
             yield i
-
+    # BONSAI changes begin - added is_infinite_sequence=False to from_iterators
     it1 = from_iterators([gen_fast],
                          is_infinite_sequence=False).for_each(lambda x: ("fast", x))
     it2 = from_iterators([gen_slow],
                          is_infinite_sequence=False).for_each(lambda x: ("slow", x))
+    # BONSAI changes end
     it = it1.gather_async().union(it2.gather_async())
     assert (repr(it) == "LocalIterator[LocalUnion[LocalIterator["
             "ParallelIterator[from_iterators[shards=1].for_each()]"
             ".gather_async()], LocalIterator[ParallelIterator["
             "from_iterators[shards=1].for_each()].gather_async()]]]")
     results = list(it)
+    # BONSAI changes begin - added code
     slow_count = sum(1 for x in results if x[0] == "slow")
     assert slow_count >= 1
     assert (len(results) - slow_count) >= 8
@@ -735,8 +754,9 @@ def gen_slow():
     assert slow_count >= 1
     assert (len(collected_categories) - slow_count) >= 8
     assert [c > 0 for c in attempts_collect_counts].count(True) > 1
+    # BONSAI changes begin
 
-
+# BONSAI changes begin - added test
 def test_union_local_async_empty_iter(ray_start_regular_shared):
     # Finite sequences
     def gen_fast():
@@ -787,8 +807,9 @@ def gen_nothing():
     assert fast_count >= 1
     assert (len(collected_categories) - fast_count) == 0
     assert [c > 0 for c in attempts_collect_counts].count(True) > 1
+# BONSAI changes end
 
-
+# BONSAI changes begin - added test
 def test_union_local_async_strict(ray_start_regular_shared):
     # Finite sequences
     def gen_fast():
@@ -843,8 +864,9 @@ def gen_slow():
     assert slow_count >= 1
     assert (len(collected_categories) - slow_count) >= 8
     assert [c > 0 for c in attempts_collect_counts].count(True) > 1
+# BONSAI changes end
 
-
+# BONSAI changes begin - added test
 def test_union_local_async_strict_empty_iter(ray_start_regular_shared):
     # Finite sequences
     def gen_fast():
@@ -896,6 +918,7 @@ def gen_nothing():
     assert fast_count >= 1
     assert (len(collected_categories) - fast_count) == 0
     assert [c > 0 for c in attempts_collect_counts].count(True) > 1
+# BONSAI changes end
 
 
 def test_serialization(ray_start_regular_shared):
diff --git a/python/ray/tests/test_memory_scheduling.py b/python/ray/tests/test_memory_scheduling.py
index d82a74d4898e..682b2967e263 100644
--- a/python/ray/tests/test_memory_scheduling.py
+++ b/python/ray/tests/test_memory_scheduling.py
@@ -8,11 +8,13 @@
 
 
 def object_store_memory(a, delta=MB):
+    # BONSAI changes begin - b = ray.available_resources()["object_store_memory"]
     r = ray.available_resources()
     if r and "object_store_memory" in r:
         b = ray.available_resources()["object_store_memory"]
     else:
         b = a
+    # BONSAI changes end
     return abs(a - b) < delta
 
 
diff --git a/python/ray/tests/test_multi_tenancy.py b/python/ray/tests/test_multi_tenancy.py
index 0bac8dd2e65e..bd49563cbaf2 100644
--- a/python/ray/tests/test_multi_tenancy.py
+++ b/python/ray/tests/test_multi_tenancy.py
@@ -263,11 +263,14 @@ def nested(i):
 
     # Wait for worker capping. worker capping should be triggered
     # every 10 ms, but we wait long enough to avoid a flaky test.
+    # BONSAI changes begin - time.sleep(1)
     time.sleep(5)
+    # BONSAI changes end
     ref2 = ray.get(nested.remote(0))
 
     # New workers shouldn't be registered because we reused the
     # previous workers that own objects.
+    # BONSAI changes begin
     # NOTE (Ruofan): The test is flakey and Ray fixes this test in Ray 1+:
     # https://github.com/ray-project/ray/pull/15571/files
     # We apply the fix here. Later once we updated the Ray version, we SHOULD
@@ -277,6 +280,7 @@ def nested(i):
     # occasionally flaky with that check.
     assert abs(num_workers - cur_num_workers) < 2, \
         (num_workers, cur_num_workers)
+    # BONSAI changes end
     assert len(ref2) == expected_num_workers
     assert len(ref) == expected_num_workers
 
diff --git a/python/ray/tests/test_runtime_env.py b/python/ray/tests/test_runtime_env.py
index 5a842de53c92..77444bb5b692 100644
--- a/python/ray/tests/test_runtime_env.py
+++ b/python/ray/tests/test_runtime_env.py
@@ -76,7 +76,7 @@ def test_single_node(ray_start_cluster_head, working_dir):
     execute_statement = "print(sum(ray.get([run_test.remote()] * 1000)))"
     script = driver_script.format(**locals())
     out = run_string_as_driver(script)
-
+    # BONSAI changes begin
     # Colorama logging may add ANSI codes to the end of the output if resetting colors
     # So we look for the ANSI code starting with "\033" and remove it before checking
     tokens = out.strip().split()
@@ -87,6 +87,7 @@ def test_single_node(ray_start_cluster_head, working_dir):
         output = tokens[-1]
 
     assert output == "1000"
+    # BONSAI changes end
     from ray._private.runtime_env import PKG_DIR
     assert len(list(Path(PKG_DIR).iterdir())) == 1
 
@@ -99,7 +100,7 @@ def test_two_node(two_node_cluster, working_dir):
     execute_statement = "print(sum(ray.get([run_test.remote()] * 1000)))"
     script = driver_script.format(**locals())
     out = run_string_as_driver(script)
-
+    # BONSAI changes begin
     # Colorama logging may add ANSI codes to the end of the output if resetting colors
     # So we look for the ANSI code starting with "\033" and remove it before checking
     tokens = out.strip().split()
@@ -108,6 +109,7 @@ def test_two_node(two_node_cluster, working_dir):
         output = tokens[-2]
     else:
         output = tokens[-1]
+    # BONSAI changes end
 
     from ray._private.runtime_env import PKG_DIR
     assert len(list(Path(PKG_DIR).iterdir())) == 1
@@ -122,7 +124,7 @@ def test_two_node_module(two_node_cluster, working_dir):
     script = driver_script.format(**locals())
     print(script)
     out = run_string_as_driver(script)
-
+    # BONSAI changes begin
     # Colorama logging may add ANSI codes to the end of the output if resetting colors
     # So we look for the ANSI code starting with "\033" and remove it before checking
     tokens = out.strip().split()
@@ -131,6 +133,7 @@ def test_two_node_module(two_node_cluster, working_dir):
         output = tokens[-2]
     else:
         output = tokens[-1]
+    # BONSAI changes end
 
     from ray._private.runtime_env import PKG_DIR
     assert len(list(Path(PKG_DIR).iterdir())) == 1
@@ -151,7 +154,7 @@ def test_two_node_uri(two_node_cluster, working_dir):
         execute_statement = "print(sum(ray.get([run_test.remote()] * 1000)))"
     script = driver_script.format(**locals())
     out = run_string_as_driver(script)
-
+    # BONSAI changes begin
     # Colorama logging may add ANSI codes to the end of the output if resetting colors
     # So we look for the ANSI code starting with "\033" and remove it before checking
     tokens = out.strip().split()
@@ -160,6 +163,7 @@ def test_two_node_uri(two_node_cluster, working_dir):
         output = tokens[-2]
     else:
         output = tokens[-1]
+    # BONSAI changes end
 
     from ray._private.runtime_env import PKG_DIR
     assert len(list(Path(PKG_DIR).iterdir())) == 1
@@ -176,7 +180,7 @@ def test_regular_actors(ray_start_cluster_head, working_dir):
 """
     script = driver_script.format(**locals())
     out = run_string_as_driver(script)
-
+    # BONSAI changes begin
     # Colorama logging may add ANSI codes to the end of the output if resetting colors
     # So we look for the ANSI code starting with "\033" and remove it before checking
     tokens = out.strip().split()
@@ -185,6 +189,7 @@ def test_regular_actors(ray_start_cluster_head, working_dir):
         output = tokens[-2]
     else:
         output = tokens[-1]
+    # BONSAI changes end
 
     from ray._private.runtime_env import PKG_DIR
     assert len(list(Path(PKG_DIR).iterdir())) == 1
@@ -201,7 +206,7 @@ def test_detached_actors(ray_start_cluster_head, working_dir):
 """
     script = driver_script.format(**locals())
     out = run_string_as_driver(script)
-
+    # BONSAI changes begin
     # Colorama logging may add ANSI codes to the end of the output if resetting colors
     # So we look for the ANSI code starting with "\033" and remove it before checking
     tokens = out.strip().split()
@@ -210,6 +215,7 @@ def test_detached_actors(ray_start_cluster_head, working_dir):
         output = tokens[-2]
     else:
         output = tokens[-1]
+    # BONSAI changes end
 
     from ray._private.runtime_env import PKG_DIR
     # It's a detached actors, so it should still be there
@@ -252,8 +258,10 @@ def create_tf_env(tf_version: str):
         command_str = command_separator.join(commands)
         subprocess.run([command_str], shell=True)
 
+    # BONSAI changes begin - tf_versions was ["2.2.0", "2.3.0"]
     tf_versions = ["2.4.1"]
     ray.get([create_tf_env.remote(version) for version in tf_versions])
+    # BONSAI changes end
     ray.shutdown()
     yield
 
@@ -280,7 +288,9 @@ def test_task_conda_env(conda_envs, shutdown_only):
     def get_tf_version():
         return tf.__version__
 
+    # BONSAI changes begin - tf_versions was ["2.2.0", "2.3.0"]
     tf_versions = ["2.4.1"]
+    # BONSAI changes end
     for tf_version in tf_versions:
         runtime_env = {"conda": f"tf-{tf_version}"}
         task = get_tf_version.options(runtime_env=runtime_env)
@@ -299,7 +309,9 @@ class TfVersionActor:
         def get_tf_version(self):
             return tf.__version__
 
+    # BONSAI changes begin - tf_versions was ["2.2.0", "2.3.0"]
     tf_versions = ["2.4.1"]
+    # BONSAI changes end
     for tf_version in tf_versions:
         runtime_env = {"conda": f"tf-{tf_version}"}
         actor = TfVersionActor.options(runtime_env=runtime_env).remote()
@@ -326,7 +338,9 @@ class TfVersionActor:
         def get_tf_version(self):
             return ray.get(wrapped_tf_version.remote())
 
+    # BONSAI changes begin - tf_versions was ["2.2.0", "2.3.0"]
     tf_versions = ["2.4.1"]
+    # BONSAI changes end
     for tf_version in tf_versions:
         runtime_env = {"conda": f"tf-{tf_version}"}
         task = wrapped_tf_version.options(runtime_env=runtime_env)
@@ -347,7 +361,9 @@ def test_job_config_conda_env(conda_envs):
     def get_conda_env():
         return tf.__version__
 
+    # BONSAI changes begin - for tf_version in ["2.2.0", "2.3.0"]:
     for tf_version in ["2.4.1"]:
+        # BONSAI changes end
         runtime_env = {"conda": f"tf-{tf_version}"}
         ray.init(job_config=JobConfig(runtime_env=runtime_env))
         assert ray.get(get_conda_env.remote()) == tf_version
diff --git a/python/ray/tune/requirements-dev.txt b/python/ray/tune/requirements-dev.txt
index bb7d8209318a..4f3d03844c03 100644
--- a/python/ray/tune/requirements-dev.txt
+++ b/python/ray/tune/requirements-dev.txt
@@ -1,11 +1,15 @@
 flake8==3.7.7
 flake8-quotes
 gym
+# BONSAI changes begin opencv-python added; opencv-python-headles was 4.3.0.36
 opencv-python==4.5.2.52
 opencv-python-headless==4.5.2.52
+# BONSAI changes end
 pandas
 requests
 tabulate
+# BONSAI changes begin - tensorflow version pinned; tensorflow_probability dependency added
 tensorflow==2.5.0
 tensorflow_probability==0.11.1
+# BONSAI changes end
 yapf==0.23.0
diff --git a/python/ray/tune/tests/test_horovod.py b/python/ray/tune/tests/test_horovod.py
index 0ab0bd9a8316..3b1d79288dd5 100644
--- a/python/ray/tune/tests/test_horovod.py
+++ b/python/ray/tune/tests/test_horovod.py
@@ -38,8 +38,9 @@ def ray_connect_cluster():
     # The code after the yield will run as teardown code.
     ray.shutdown()
 
-
+# BONSAI changes begin - added decorator
 @pytest.mark.skip("TODO(Edi): Fix ImportError: horovod/torch/mpi_lib_v2.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN5torch3jit6tracer9addInputsEPNS0_4NodeEPKcRKN3c1013TensorOptionsE")
+# BONSAI changes end
 def test_single_step(ray_start_2_cpus):
     trainable_cls = DistributedTrainableCreator(
         _train_simple, num_hosts=1, num_slots=2)
@@ -47,8 +48,9 @@ def test_single_step(ray_start_2_cpus):
     trainer.train()
     trainer.stop()
 
-
+# BONSAI changes begin - added decorator
 @pytest.mark.skip("TODO(Edi): Fix ImportError: horovod/torch/mpi_lib_v2.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN5torch3jit6tracer9addInputsEPNS0_4NodeEPKcRKN3c1013TensorOptionsE")
+# BONSAI changes end
 def test_step_after_completion(ray_start_2_cpus):
     trainable_cls = DistributedTrainableCreator(
         _train_simple, num_hosts=1, num_slots=2)
@@ -66,8 +68,9 @@ def bad_func(a, b, c):
     with pytest.raises(ValueError):
         t_cls()
 
-
+# BONSAI changes begin - added decorator
 @pytest.mark.skip("TODO(Edi): Fix ImportError: horovod/torch/mpi_lib_v2.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN5torch3jit6tracer9addInputsEPNS0_4NodeEPKcRKN3c1013TensorOptionsE")
+# BONSAI changes end
 def test_set_global(ray_start_2_cpus):
     trainable_cls = DistributedTrainableCreator(_train_simple, num_slots=2)
     trainable = trainable_cls()
@@ -75,8 +78,9 @@ def test_set_global(ray_start_2_cpus):
     trainable.stop()
     assert result["rank"] == 0
 
-
+# BONSAI changes begin - added decorator
 @pytest.mark.skip("TODO(Edi): Fix ImportError: horovod/torch/mpi_lib_v2.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN5torch3jit6tracer9addInputsEPNS0_4NodeEPKcRKN3c1013TensorOptionsE")
+# BONSAI changes end
 @pytest.mark.parametrize("enabled_checkpoint", [True, False])
 def test_simple_tune(ray_start_4_cpus, enabled_checkpoint):
     trainable_cls = DistributedTrainableCreator(_train_simple, num_slots=2)
diff --git a/python/ray/tune/tests/test_progress_reporter.py b/python/ray/tune/tests/test_progress_reporter.py
index 66ee865b9156..b5f0700a09fa 100644
--- a/python/ray/tune/tests/test_progress_reporter.py
+++ b/python/ray/tune/tests/test_progress_reporter.py
@@ -456,12 +456,14 @@ def testVerboseReporting(self):
             try:
                 self.assertIn(VERBOSE_EXP_OUT_1, output)
                 self.assertIn(VERBOSE_EXP_OUT_2, output)
+                # BONSAI changes begin
                 # TODO(Edi): revert this after figuring out the error
                 # self.assertIn(VERBOSE_TRIAL_NORM, output)
                 # self.assertNotIn(VERBOSE_TRIAL_DETAIL, output)
                 trial_norm = VERBOSE_TRIAL_NORM in output
                 trial_detail = VERBOSE_TRIAL_DETAIL in output
                 self.assertTrue(trial_norm or trial_detail)
+                # BONSAI changes end
             except Exception:
                 print("*** BEGIN OUTPUT ***")
                 print(output)
@@ -473,12 +475,14 @@ def testVerboseReporting(self):
             try:
                 self.assertIn(VERBOSE_EXP_OUT_1, output)
                 self.assertIn(VERBOSE_EXP_OUT_2, output)
+                # BONSAI changes begin
                 # TODO(Edi): revert this after figuring out the error
                 # self.assertNotIn(VERBOSE_TRIAL_NORM, output)
                 # self.assertIn(VERBOSE_TRIAL_DETAIL, output)
                 trial_norm = VERBOSE_TRIAL_NORM in output
                 trial_detail = VERBOSE_TRIAL_DETAIL in output
                 self.assertTrue(trial_norm or trial_detail)
+                # BONSAI changes end
             except Exception:
                 print("*** BEGIN OUTPUT ***")
                 print(output)
diff --git a/python/ray/util/iter.py b/python/ray/util/iter.py
index 1a277ccf4945..74748c8226f6 100644
--- a/python/ray/util/iter.py
+++ b/python/ray/util/iter.py
@@ -1,22 +1,27 @@
 from __future__ import generator_stop
 from contextlib import contextmanager
 import collections
+# BONSAI changes begin - added import
 import logging
+# BONSAI changes end
 import random
 import threading
 import time
+# BONSAI changes begin - added Iterator import
 from typing import TypeVar, Generic, Iterable, List, Callable, Any, Iterator
+# BONSAI changes end
 
 import ray
 from ray.util.iter_metrics import MetricsContext, SharedMetrics
-
+# BONSAI changes begin - added code
 logger = logging.getLogger(__name__)
+# BONSAI changes end
 
 # The type of an iterator element.
 T = TypeVar("T")
 U = TypeVar("U")
 
-
+# BONSAI changes begin - added function
 def was_cause_by_stop_iteration(ex) -> bool:
     if isinstance(ex, StopIteration):
         return True
@@ -24,6 +29,7 @@ def was_cause_by_stop_iteration(ex) -> bool:
         return was_cause_by_stop_iteration(ex.__cause__)
     else:
         return False
+# BONSAI changes end
 
 
 def from_items(items: List[T], num_shards: int = 2,
@@ -43,8 +49,10 @@ def from_items(items: List[T], num_shards: int = 2,
     name = "from_items[{}, {}, shards={}{}]".format(
         items and type(items[0]).__name__ or "None", len(items), num_shards,
         ", repeat=True" if repeat else "")
+    # BONSAI changes begin -  return from_iterators(shards, repeat=repeat, name=name)
     return from_iterators(shards, repeat=repeat, name=name,
                           is_infinite_sequence=False)
+    # BONSAI changes begin
 
 
 def from_range(n: int, num_shards: int = 2,
@@ -73,14 +81,18 @@ def from_range(n: int, num_shards: int = 2,
         generators,
         repeat=repeat,
         name=name,
+        # BONSAI changes begin - added keyword parameter
         is_infinite_sequence=False
+        # BONSAI changes end
     )
 
 
 def from_iterators(generators: List[Iterable[T]],
                    repeat: bool = False,
                    name=None,
+                   # BONSAI changes begin - added keyword parameter
                    is_infinite_sequence: bool = True) -> "ParallelIterator[T]":
+                   # BONSAI changes end
     """Create a parallel iterator from a list of iterables.
     An iterable can be a conatiner (list, str, tuple, set, etc.),
     a generator, or a custom class that implements __iter__ or __getitem__.
@@ -106,25 +118,34 @@ def from_iterators(generators: List[Iterable[T]],
             but a lambda that returns it can be.
         repeat (bool): Whether to cycle over the iterators forever.
         name (str): Optional name to give the iterator.
+        # BONSAI changes begin - added documentation
         is_infinite_sequence (bool): Whether the sequence generated by
         item_generator should be consider an infinite sequence of items.
         For the sake of the parallel iterators, one that hold a infinite
         sequence, could be called again after a stop iteration message.
         In other words, an StopIteration for a infinite sequence must be
         seen as a "no items available" message.
+        # BONSAI changes end
+
     """
     worker_cls = ray.remote(ParallelIteratorWorker)
+    # BONSAI changes begin - actors = [worker_cls.remote(g, repeat) for g in generators]
     actors = [worker_cls.remote(g, repeat, is_infinite_sequence = is_infinite_sequence)
               for g in generators]
+    # BONSAI changes end
     if not name:
         name = "from_iterators[shards={}{}]".format(
             len(generators), ", repeat=True" if repeat else "")
+    # BONSAI changes begin - return from_actors(actors, name=name)
     return from_actors(actors, name=name, is_infinite_sequence=is_infinite_sequence)
+    # BONSAI changes end
 
 
 def from_actors(actors: List["ray.actor.ActorHandle"],
                 name=None,
+                # BONSAI changes begin - added keyword parameter
                 is_infinite_sequence: bool = True) -> "ParallelIterator[T]":
+                # BONSAI changes end
     """Create a parallel iterator from an existing set of actors.
 
     Each actor must subclass the ParallelIteratorWorker interface.
@@ -133,17 +154,21 @@ def from_actors(actors: List["ray.actor.ActorHandle"],
         actors (list): List of actors that each implement
             ParallelIteratorWorker.
         name (str): Optional name to give the iterator.
+         # BONSAI changes begin - added documentation
         is_infinite_sequence (bool): Whether the sequence generated by
         item_generator should be consider an infinite sequence of items.
         For the sake of the parallel iterators, one that hold a infinite
         sequence, could be called again after a stop iteration message.
         In other words, an StopIteration for a infinite sequence must be
         seen as a "no items available" message.
+         # BONSAI changes end
     """
     if not name:
         name = f"from_actors[shards={len(actors)}]"
+    # BONSAI changes begin - return ParallelIterator([_ActorSet(actors, [])], name, parent_iterators=[])
     return ParallelIterator([_ActorSet(actors, [])], name, parent_iterators=[],
                             is_infinite_sequence = is_infinite_sequence)
+    # BONSAI changes end
 
 
 class ParallelIterator(Generic[T]):
@@ -197,7 +222,9 @@ class ParallelIterator(Generic[T]):
 
     def __init__(self, actor_sets: List["_ActorSet"], name: str,
                  parent_iterators: List["ParallelIterator[Any]"],
+                 # BONSAI changes begin - added kwyword parameter
                  is_infinite_sequence: bool = True):
+                 # BONSAI changes end
         """Create a parallel iterator (this is an internal function).
 
         Args:
@@ -215,8 +242,9 @@ def __init__(self, actor_sets: List["_ActorSet"], name: str,
 
         # keep explicit reference to parent iterator for repartition
         self.parent_iterators = parent_iterators
-
+        # BONSAI changes begin - added code
         self.is_infinite_sequence = is_infinite_sequence
+        # BONSAI changes wns
 
     def __iter__(self):
         raise TypeError(
@@ -235,7 +263,9 @@ def _with_transform(self, local_it_fn, name):
             [a.with_transform(local_it_fn) for a in self.actor_sets],
             name=self.name + name,
             parent_iterators=self.parent_iterators,
+            # BONSAI changes begin - added keyword parameter
             is_infinite_sequence=self.is_infinite_sequence
+            # BONSAI changes end
         )
 
     def transform(self, fn: Callable[[Iterable[T]], Iterable[U]]
@@ -457,11 +487,15 @@ def base_iterator(num_partitions, partition_index, timeout=None):
                             batch_ms=batch_ms)] = actor
                         for item in batch:
                             yield item
+                    # BONSAI changes begin
+                    # except StopIteration:
+                    #   pass
                     except (StopIteration, RuntimeError) as ex:
                         if was_cause_by_stop_iteration(ex):
                             pass
                         else:
                             raise ex
+                    # BONSAI changes end
                 # Always yield after each round of wait with timeout.
                 if timeout is not None:
                     yield _NextValueNotReady()
@@ -477,7 +511,9 @@ def make_gen_i(i):
         # need explicit reference to self so actors in this instance do not die
         return ParallelIterator(
             [_ActorSet(actors, [])], name, parent_iterators=[self],
+            # BONSAI changes begin - added keyword parameter
             is_infinite_sequence=self.is_infinite_sequence
+            # BONSAI changes end
         )
 
     def gather_sync(self) -> "LocalIterator[T]":
@@ -525,6 +561,7 @@ def base_iterator(timeout=None):
                         yield _NextValueNotReady()
                 except TimeoutError:
                     yield _NextValueNotReady()
+                # BONSAI changes begin
                 except (StopIteration, RuntimeError) as ex:
                     if was_cause_by_stop_iteration(ex):
                         # If we are streaming (infinite sequence) then
@@ -552,10 +589,12 @@ def base_iterator(timeout=None):
                         futures = [a.par_iter_next.remote() for a in active]
                     else:
                         raise ex
-
+                # BONSAI changes end
         name = f"{self}.batch_across_shards()"
         return LocalIterator(base_iterator, SharedMetrics(), name=name,
+                             # BONSAI changes begin - added keyword parameter
                              is_infinite_sequence=self.is_infinite_sequence)
+                             # BONSAI changes end
 
     def gather_async(self, batch_ms=0, num_async=1) -> "LocalIterator[T]":
         """Returns a local iterable for asynchronous iteration.
@@ -599,7 +638,9 @@ def base_iterator(timeout=None):
             for _ in range(num_async):
                 for a in all_actors:
                     futures[a.par_iter_next_batch.remote(batch_ms)] = a
+            # BONSAI changes begin - added code
             active_actors = set(all_actors)
+            # BONSAI changes end
             while futures:
                 pending = list(futures)
                 if timeout is None:
@@ -619,9 +660,14 @@ def base_iterator(timeout=None):
                         batch = ray.get(obj_ref)
                         futures[actor.par_iter_next_batch.remote(
                             batch_ms)] = actor
+                        # BONSAI changes begin - added code
                         active_actors.add(actor)
+                        # BONSAI changes end
                         for item in batch:
                             yield item
+                    # BONSAI changes begin
+                    # except StopIteration:
+                    #   pass
                     except (StopIteration, RuntimeError) as ex:
                         if was_cause_by_stop_iteration(ex):
                             # If we are streaming (infinite sequence) then
@@ -638,13 +684,16 @@ def base_iterator(timeout=None):
                                 pass
                         else:
                             raise ex
+                    # BONSAI changes end
                 # Always yield after each round of wait with timeout.
                 if timeout is not None:
                     yield _NextValueNotReady()
 
         name = f"{self}.gather_async()"
         local_iter = LocalIterator(base_iterator, SharedMetrics(), name=name,
+        # BONSAI changes begin - added keyword parameter
                                    is_infinite_sequence=self.is_infinite_sequence)
+        # BONSAI changes end
         return local_iter
 
     def take(self, n: int) -> List[T]:
@@ -663,18 +712,22 @@ def union(self, other: "ParallelIterator[T]") -> "ParallelIterator[T]":
         actor_sets = []
         actor_sets.extend(self.actor_sets)
         actor_sets.extend(other.actor_sets)
+        # BONSAI changes begin - added code
         is_infinite_sequence = self.is_infinite_sequence
         if is_infinite_sequence != self.is_infinite_sequence:
             logger.warning("One iterator is a infinite sequence and the other is not. "
                            "Assuming the union as a infinite sequence.")
             is_infinite_sequence = True
+        # BONSAI changes end
         # if one of these iterators is a result of a repartition, we need to
         # keep an explicit reference to its parent iterator
         return ParallelIterator(
             actor_sets,
             f"ParallelUnion[{self}, {other}]",
             parent_iterators=self.parent_iterators + other.parent_iterators,
+            # BONSAI changes begin - added keyword parameter
             is_infinite_sequence=is_infinite_sequence
+            # BONSAI changes end
         )
 
     def select_shards(self,
@@ -699,7 +752,9 @@ def select_shards(self,
             [new_actor_set],
             f"{self}.select_shards({len(shards_to_keep)} total)",
             parent_iterators=self.parent_iterators,
+            # BONSAI changes begin - added keyword parameter
             is_infinite_sequence=self.is_infinite_sequence
+            # BONSAI changes end
         )
 
     def num_shards(self) -> int:
@@ -762,15 +817,21 @@ def base_iterator(timeout=None):
                         yield _NextValueNotReady()
                 except TimeoutError:
                     yield _NextValueNotReady()
+                # BONSAI changes begin
+                # except StopIteration:
+                #   break
                 except (StopIteration, RuntimeError) as ex:
                     if was_cause_by_stop_iteration(ex):
                         break
                     else:
                         raise ex
+                # BONSAI changes end
 
         name = self.name + f".shard[{shard_index}]"
         return LocalIterator(base_iterator, SharedMetrics(), name=name,
+                             # BONSAI changes begin - added keyword parameter
                              is_infinite_sequence=self.is_infinite_sequence)
+                             # BONSAI changes end
 
 
 class LocalIterator(Generic[T]):
@@ -787,18 +848,20 @@ class LocalIterator(Generic[T]):
     # we will call it at the beginning of each data fetch call. This can be
     # used to measure the underlying wait latency for measurement purposes.
     ON_FETCH_START_HOOK_NAME = "_on_fetch_start"
-
+    # BONSAI changes begin - added constant
     # If a function passed to LocalIterator.for_each() has this method,
     # we will call it each not ready value condition. This can be
     # used to implement an early stop of the iterator by stoping
     # the iteration, or any default value by returning something.
     # This method should expect the same arguments of the __call__.
     HANDLE_NEXT_VALUE_NOT_READY_HOOK_NAME = "_handle_next_value_not_ready"
+    # BONSAI changes end
 
     thread_local = threading.local()
-
+    # BONSAI changes begin - added constants
     UNION_MAX_PULL = 1000
     UNION_PULL_DELAY_SEC = 1e-7
+    # BONSAI changes end
 
     def __init__(self,
                  base_iterator: Callable[[], Iterable[T]],
@@ -806,7 +869,9 @@ def __init__(self,
                  local_transforms: List[Callable[[Iterable], Any]] = None,
                  timeout: int = None,
                  name=None,
+                 # BONSAI changes begin - added keyword parameter
                  is_infinite_sequence: bool = True):
+                 # BONSAI changes end
         """Create a local iterator (this is an internal function).
 
         Args:
@@ -824,12 +889,14 @@ def __init__(self,
                 which _NextValueNotReady will be returned. This avoids
                 blocking.
             name (str): Optional name for this iterator.
+            # BONSAI changes begin - added documentation
             is_infinite_sequence (bool): Whether the sequence generated by
             item_generator should be consider an infinite sequence of items.
             For the sake of the parallel iterators, one that hold a infinite
             sequence, could be called again after a stop iteration message.
             In other words, an StopIteration for a infinite sequence must be
             seen as a "no items available" message.
+            # BONSAI changes end
         """
         assert isinstance(shared_metrics, SharedMetrics)
         self.base_iterator = base_iterator
@@ -838,7 +905,9 @@ def __init__(self,
         self.shared_metrics = shared_metrics
         self.timeout = timeout
         self.name = name or "unknown"
+        # BONSAI changes begin - added code
         self.is_infinite_sequence = is_infinite_sequence
+        # BONSAI changes end
 
     @staticmethod
     def get_metrics() -> MetricsContext:
@@ -864,10 +933,13 @@ def _metrics_context(self):
 
     def __iter__(self):
         self._build_once()
+        # BONSAI changes begin - return self.built_iterator
         return self
+        # BONSAI changes end
 
     def __next__(self):
         self._build_once()
+        # BONSAI changes begin - return next(self.built_iterator)
         try:
             return next(self.built_iterator)
         except (StopIteration, RuntimeError) as ex:
@@ -876,6 +948,7 @@ def __next__(self):
                 if self.is_infinite_sequence:
                     self.built_iterator = None
             raise ex
+        # BONSAI changes end
 
     def __str__(self):
         return repr(self)
@@ -904,12 +977,14 @@ def for_each(self, fn: Callable[[T], U], max_concurrency=1,
             def apply_foreach(it):
                 for item in it:
                     if isinstance(item, _NextValueNotReady):
+                        # BONSAI changes begin - yield item
                         if hasattr(fn, LocalIterator.HANDLE_NEXT_VALUE_NOT_READY_HOOK_NAME):
                             with self._metrics_context():
                                 result = fn._handle_next_value_not_ready(item)
                             yield result
                         else:
                             yield item
+                        # BONSAI changes end
                     else:
                         # Keep retrying the function until it returns a valid
                         # value. This allows for non-blocking functions.
@@ -964,7 +1039,9 @@ def add_wait_hooks(it):
             self.shared_metrics,
             self.local_transforms + [apply_foreach],
             name=self.name + ".for_each()",
+            # BONSAI changes begin - added keyword parameter
             is_infinite_sequence=self.is_infinite_sequence)
+            # BONSAI changes end
 
     def filter(self, fn: Callable[[T], bool]) -> "LocalIterator[T]":
         def apply_filter(it):
@@ -978,7 +1055,9 @@ def apply_filter(it):
             self.shared_metrics,
             self.local_transforms + [apply_filter],
             name=self.name + ".filter()",
+            # BONSAI changes begin - added keyword parameter
             is_infinite_sequence=self.is_infinite_sequence)
+            # BONSAI changes end
 
     def batch(self, n: int) -> "LocalIterator[List[T]]":
         def apply_batch(it):
@@ -999,10 +1078,13 @@ def apply_batch(it):
             self.shared_metrics,
             self.local_transforms + [apply_batch],
             name=self.name + f".batch({n})",
+            # BONSAI changes begin - added keyworkd parameter
             is_infinite_sequence=self.is_infinite_sequence)
+            # BONSAI changes end
 
     def flatten(self) -> "LocalIterator[T[0]]":
         def apply_flatten(it):
+            # BONSAI changes begin - added Exception handling
             try:
                 for item in it:
                     if isinstance(item, _NextValueNotReady):
@@ -1013,13 +1095,16 @@ def apply_flatten(it):
             except (StopIteration, RuntimeError) as ex:
                 if not was_cause_by_stop_iteration(ex):
                     raise ex
+            # BONSAI changes end
 
         return LocalIterator(
             self.base_iterator,
             self.shared_metrics,
             self.local_transforms + [apply_flatten],
             name=self.name + ".flatten()",
+            # BONSAI changes begin - added keyword parameter
             is_infinite_sequence=self.is_infinite_sequence)
+            # BONSAI changes end
 
     def shuffle(self, shuffle_buffer_size: int,
                 seed: int = None) -> "LocalIterator[T]":
@@ -1061,7 +1146,9 @@ def apply_shuffle(it):
             ".shuffle(shuffle_buffer_size={}, seed={})".format(
                 shuffle_buffer_size,
                 str(seed) if seed is not None else "None"),
+            # BONSAI changes begin - added keyword parameter
             is_infinite_sequence=self.is_infinite_sequence)
+            # BONSAI changes end
 
     def combine(self, fn: Callable[[T], List[U]]) -> "LocalIterator[U]":
         it = self.for_each(fn).flatten()
@@ -1135,11 +1222,15 @@ def gen(timeout):
                         if len(queues[i]) == 0:
                             try:
                                 fill_next(timeout)
+                            # BONSAI changes begin
+                            # except StopIteration:
+                            #   return
                             except (StopIteration, RuntimeError) as ex:
                                 if was_cause_by_stop_iteration(ex):
                                     return
                                 else:
                                     raise ex
+                            # BONSAI changes end
                         yield queues[i].popleft()
 
             return gen
@@ -1151,7 +1242,9 @@ def gen(timeout):
                     make_next(i),
                     self.shared_metrics, [],
                     name=self.name + f".duplicate[{i}]",
+                    # BONSAI changes begin - added keyword parameter
                     is_infinite_sequence=self.is_infinite_sequence))
+                    # BONSAI changes end
 
         return iterators
 
@@ -1159,7 +1252,9 @@ def union(self,
               *others: "LocalIterator[T]",
               deterministic: bool = False,
               round_robin_weights: List[float] = None,
+              # BONSAI changes begin - added keyword parameter
               strict=False) -> "LocalIterator[T]":
+              # BONSAI changes end
         """Return an iterator that is the union of this and the others.
 
         Args:
@@ -1171,17 +1266,20 @@ def union(self,
                 as many items from the first iterator as the second.
                 [2, 1, "*"] will cause as many items to be pulled as possible
                 from the third iterator without blocking. This overrides the
+                # BONSAI changes begin - added documentation
                 deterministic flag. If weights has fixed values, we will Stop
                 the Iteration if any of the iterators with fixed weight stop
                 its iteration before metting the expected count.
+                # BONSAI changes end
         """
 
         for it in others:
             if not isinstance(it, LocalIterator):
                 raise ValueError(
                     f"other must be of type LocalIterator, got {type(it)}")
-
+        # BONSAI changes begin - active = []
         initial_active = []
+        # BONSAI changes end
         parent_iters = [self] + list(others)
         shared_metrics = SharedMetrics(
             parents=[p.shared_metrics for p in parent_iters])
@@ -1198,6 +1296,7 @@ def union(self,
             round_robin_weights = [1] * len(parent_iters)
 
         for i, it in enumerate(parent_iters):
+            # BONSAI changes begin
             initial_active.append(
                 LocalIterator(
                     it.base_iterator,
@@ -1206,8 +1305,10 @@ def union(self,
                     timeout=timeouts[i],
                     is_infinite_sequence=self.is_infinite_sequence))
         initial_active = list(zip(round_robin_weights, initial_active))
+        # BONSAI changes end
 
         def build_union(timeout=None):
+            # BONSAI changes begin
             active = list(initial_active)
             pull_counts = [0] * len(active)
             last_pull_timestamp = [0] * len(active)
@@ -1263,6 +1364,7 @@ def build_union(timeout=None):
                         else:
                             raise ex
                 pull_counts = [c for j, c in enumerate(pull_counts) if j not in removed_iter_indices]
+                # BONSAI changes end
                 if not active:
                     break
 
@@ -1270,7 +1372,9 @@ def build_union(timeout=None):
             build_union,
             shared_metrics, [],
             name=f"LocalUnion[{self}, {', '.join(map(str, others))}]",
+            # BONSAI changes begin - added keyword parameter
             is_infinite_sequence=self.is_infinite_sequence)
+            # BONSAI changes end - added keyword parameter
 
 
 class ParallelIteratorWorker(object):
@@ -1279,7 +1383,9 @@ class ParallelIteratorWorker(object):
     Actors that are passed to iter.from_actors() must subclass this interface.
     """
 
+    # BONSAI changes begin - added is_infinite_sequence keyword parameter
     def __init__(self, item_generator: Any, repeat: bool, is_infinite_sequence: bool = True):
+    # BONSAI changes end
         """Create an iterator worker.
 
         Subclasses must call this init function.
@@ -1290,16 +1396,19 @@ def __init__(self, item_generator: Any, repeat: bool, is_infinite_sequence: bool
                 functions since the generator itself might not be serializable,
                 but a lambda that returns it can be.
             repeat (bool): Whether to loop over the iterator forever.
+            # BONSAI changes begin - added documentation
             is_infinite_sequence (bool): Whether the sequence generated by
             item_generator should be consider an infinite sequence of items.
             For the sake of the parallel iterators, one that hold a infinite
             sequence, could be called again after a stop iteration message.
             In other words, an StopIteration for a infinite sequence must be
             seen as a "no items available" message.
+            # BONSAI changes end
         """
-
+        # BONSAI changes begin - added logging
         logger.info("Creating ParallelIteratorWorker with repeat {} and "
                     "is_infinite_sequence {}".format(repeat, is_infinite_sequence))
+        # BONSAI changes end
 
         def make_iterator():
             if callable(item_generator):
@@ -1308,6 +1417,7 @@ def make_iterator():
                 return item_generator
 
         if repeat:
+            # BONSAI changes begin
             class _GeneratorWrapper(Iterator[Any]):
                 def __init__(self):
                     self.inner_iterator = None
@@ -1344,19 +1454,24 @@ def __next__(self) -> Any:
                             raise ex
 
             self.item_generator = _GeneratorWrapper()
+            # BONSAI changes end
         else:
             self.item_generator = make_iterator()
 
         self.transforms = []
         self.local_it = None
         self.next_ith_buffer = None
+        # BONSAI changes begin - added code
         self.is_infinite_sequence = is_infinite_sequence
+        # BONSAI changes end
 
     def par_iter_init(self, transforms):
         """Implements ParallelIterator worker init."""
         it = LocalIterator(lambda timeout: self.item_generator,
                            SharedMetrics(),
+                           # BONSAI changes begin - added keyword parameter
                            is_infinite_sequence=self.is_infinite_sequence)
+                           # BONSAI changes end
         for fn in transforms:
             it = fn(it)
             assert it is not None, fn
@@ -1377,6 +1492,7 @@ def par_iter_next_batch(self, batch_ms: int):
         while time.time() < t_end:
             try:
                 batch.append(self.par_iter_next())
+            # BONSAI changes begin
             except (StopIteration, RuntimeError) as ex:
                 if was_cause_by_stop_iteration(ex):
                     if len(batch) == 0:
@@ -1385,6 +1501,7 @@ def par_iter_next_batch(self, batch_ms: int):
                         pass
                 else:
                     raise ex
+            # BONSAI changes end
         return batch
 
     def par_iter_slice(self, step: int, start: int):
@@ -1402,11 +1519,15 @@ def par_iter_slice(self, step: int, start: int):
                 try:
                     val = next(self.local_it)
                     self.next_ith_buffer[j].append(val)
+                # BONSAI changes begin
+                # except StopIteration:
+                #   pass
                 except (StopIteration, RuntimeError) as ex:
                     if was_cause_by_stop_iteration(ex):
                         pass
                     else:
                         raise ex
+                # BONSAI changes end
 
             if not self.next_ith_buffer[start]:
                 raise StopIteration
@@ -1423,6 +1544,7 @@ def par_iter_slice_batch(self, step: int, start: int, batch_ms: int):
         while time.time() < t_end:
             try:
                 batch.append(self.par_iter_slice(step, start))
+            # BONSAI changes begin
             except (StopIteration, RuntimeError) as ex:
                 if was_cause_by_stop_iteration(ex):
                     if len(batch) == 0:
@@ -1431,6 +1553,7 @@ def par_iter_slice_batch(self, step: int, start: int, batch_ms: int):
                         pass
                 else:
                     raise ex
+            # BONSAI changes end
         return batch
 
 
diff --git a/python/ray/util/sgd/BUILD b/python/ray/util/sgd/BUILD
index dd07856faf4e..a2f325ef379d 100644
--- a/python/ray/util/sgd/BUILD
+++ b/python/ray/util/sgd/BUILD
@@ -20,7 +20,9 @@ py_test(
 
 py_test(
     name = "test_torch",
+    # BONSAI changes begin - size = "large",
     size = "enormous",
+    # BONSAI changes end
     srcs = ["tests/test_torch.py"],
     tags = ["exclusive", "pytorch"],
     deps = [":sgd_lib"],
diff --git a/python/ray/util/sgd/tests/test_torch.py b/python/ray/util/sgd/tests/test_torch.py
index e078ffe428ae..2819a5f1c386 100644
--- a/python/ray/util/sgd/tests/test_torch.py
+++ b/python/ray/util/sgd/tests/test_torch.py
@@ -365,12 +365,16 @@ def test_dataset(ray_start_4_cpus, use_local):
     )
 
     dataset = dataset_creator()
+    # BONSAI changes begin - for i in range(5):
     for i in range(1):
+        # BONSAI changes end
         trainer.train(dataset=dataset, num_steps=100)
 
     x = mlp_identity.to_mat(0.5)
     prediction = float(trainer.get_model()(x)[0][0])
+    # BONSAI changes begin - assert 0.4 <= prediction <= 0.6
     assert 0.1 <= prediction <= 0.6
+    # BONSAI changes end
     trainer.shutdown()
 
 
diff --git a/python/ray/util/sgd/tf/examples/tf-example-sgd.yaml b/python/ray/util/sgd/tf/examples/tf-example-sgd.yaml
index fcf31354b70e..78f6b4ce07ce 100644
--- a/python/ray/util/sgd/tf/examples/tf-example-sgd.yaml
+++ b/python/ray/util/sgd/tf/examples/tf-example-sgd.yaml
@@ -44,7 +44,9 @@ worker_nodes:
     #         MarketType: spot
 
 setup_commands:
-    - conda install setuptools=45.1.0=py36_0 wrapt=1.11.2 --yes # workaround to fix wrapt error
+    # BONSAI changes begin - setuptools version changed from 45.1 to 41.0
+    - conda install setuptools=41.0.0=py36_0 wrapt=1.11.2 --yes # workaround to fix wrapt error
+    # BONSAI changes end
     - ray &> /dev/null || pip install -U https://s3-us-west-2.amazonaws.com/ray-wheels/latest/ray-2.0.0.dev0-cp36-cp36m-manylinux2014_x86_64.whl
     - pip install -U ray[tune]
     - pip install tensorflow-gpu==2.1.0
diff --git a/python/ray/util/sgd/torch/examples/dcgan.py b/python/ray/util/sgd/torch/examples/dcgan.py
index 7d9f65338c28..9d5539d2d0d0 100644
--- a/python/ray/util/sgd/torch/examples/dcgan.py
+++ b/python/ray/util/sgd/torch/examples/dcgan.py
@@ -192,7 +192,9 @@ def train_batch(self, batch, batch_info):
         # self.device is set automatically
         real_cpu = batch[0].to(self.device)
         batch_size = real_cpu.size(0)
+        # BONSAI changes begin - label = torch.full((batch_size, ), real_label, device=self.device)
         label = torch.full((batch_size, ), real_label, device=self.device, dtype=torch.float)
+        # BONSAI changes end
         output = discriminator(real_cpu).view(-1)
         errD_real = self.criterion(output, label)
         errD_real.backward()
diff --git a/python/requirements.txt b/python/requirements.txt
index c1b25bd49f11..de2280613df0 100644
--- a/python/requirements.txt
+++ b/python/requirements.txt
@@ -5,8 +5,12 @@
 # In short, if you change it here, PLEASE also change it in setup.py.
 #
 # setup.py install_requires
+# BONSAI changes begin
+# aiohttp==3.7
+# aioredis
 aiohttp==3.7.4
 aioredis==1.3.1
+# BONSAI changes end
 click >= 7.0
 cloudpickle
 colorama
@@ -14,34 +18,58 @@ colorful
 filelock
 gpustat
 grpcio >= 1.28.1
+# BONSAI changes begin - jsonschema pinned
 jsonschema==3.0
+# BONSAI changes end
 msgpack >= 1.0.0, < 2.0.0
+# BONSAI changes begin - numpy >= 1.16
 numpy ~= 1.19.2
+# BONSAI changes end
 opencensus
+# BONSAI changes begin
+# prometheus_client >= 0.7.1
+# protobuf >= 3.8.0
 prometheus_client >= 0.11.0
 protobuf==3.15.3
+# BONSAI changes end
 py-spy >= 0.2.0
 pyyaml
+# BONSAI changes begin - redis >= 3.5.0
 redis==3.5.3
+# BONSAI changes end
 requests
+# BONSAI changes begin - added pinned dependencies
 urllib3==1.26.7
 pickle5==0.0.11
 h5py==3.1.0
+# BONSAI changes end
 
 ## setup.py extras
+# BONSAI changes begin - pinned dependency
 atari_py==0.2.6
+# BONSAI changes end
 dm_tree
 flask
+# BONSAI changes begin - pinned dependencies
 gym[atari]==0.18.0
 lz4==3.1.3
+# BONSAI changes end
+# BONSAI changes begin -added dependency
 opencv-python==4.5.2.52
+# BONSAI changes end
+# BONSAI changes begin
+# opencv-python-headless==4.3.0.36
+# pandas==1.0.5
 opencv-python-headless==4.5.2.52
 pandas==1.1.5
+# BONSAI changes end
 scipy==1.4.1
 tabulate
 tensorboardX
 uvicorn
+# BONSAI changes begin - pinned dependency
 pydantic<1.8.1
+# BONSAI changes end
 dataclasses; python_version < '3.7'
 starlette
 
@@ -53,36 +81,52 @@ msrestazure
 boto3
 cython==0.29.0
 dataclasses; python_version < '3.7'
+# BONSAI changes begin - pinned dependency
 dask[complete]==2021.06.0
+# BONSAI changes end
 feather-format
+# BONSAI changes begin - pinned dependency
 gym==0.18.0
+# BONSAI changes end
 gym-minigrid
 kubernetes
 lxml
+# BONSAI changes begin - pinned dependencies
 moto==1.3.8
 mypy==0.921
 networkx==2.4.0
+# BONSAI changes end
 numba
 # higher version of llvmlite breaks windows
+# BONSAI changes begin - llvmlite==0.34.0
 llvmlite==0.36.0
+# BONSAI changes end
 openpyxl
 pexpect
 Pillow; platform_system != "Windows"
 pygments
+# BONSAI changes begin - pytest==5.4.3
 pytest==6.1.0
+# BONSAI changes end
 pytest-asyncio
 pytest-rerunfailures
 pytest-sugar
 pytest-timeout
 scikit-learn==0.22.2
+# BONSAI changes begin - pinned dependencies
 tensorflow==2.5.0
-tensorflow_probability==0.11.1
 testfixtures==6.18.5
+# BONSAI changes end
+# BONSAI changes begin - added dependency
+tensorflow_probability==0.11.1
+# BONSAI changes end
 werkzeug
 xlrd
 starlette
+# BONSAI changes begin - removed fastapi
+# BONSAI changes end
 smart_open[s3]
-
+# BONSAI changes begin - added dependencies
 # Moab contrib
 pyrr==0.10.3
 
@@ -92,3 +136,6 @@ idna==2.8.0
 six==1.15.0
 typing-extensions==3.7.4
 wrapt==1.12.1
+# BONSAI changes end
+# BONSAI changes begin - removed tqdm
+# BONSAI changes end
diff --git a/python/requirements_linters.txt b/python/requirements_linters.txt
index d9d2b2552810..1e7406e5b8b1 100644
--- a/python/requirements_linters.txt
+++ b/python/requirements_linters.txt
@@ -1,5 +1,7 @@
 flake8==3.7.7 
 flake8-comprehensions
 flake8-quotes==2.0.0
+# BONSAI changes begin - mypy==0.782
 mypy==0.921
+# BONSAI changes end
 yapf==0.23.0
diff --git a/python/requirements_ml_docker.txt b/python/requirements_ml_docker.txt
index 8a6050433ca6..5455aed8e5b0 100644
--- a/python/requirements_ml_docker.txt
+++ b/python/requirements_ml_docker.txt
@@ -1,7 +1,13 @@
 ipython
+# BONSAI changes begin - tensorflow-gpu>=2.4.0
 tensorflow-gpu==2.5.0
+# BONSAI changes end
 -f https://download.pytorch.org/whl/torch_stable.html
-torch==1.8.1+cu110 
+# BONSAI changes begin - torch==1.7.1+cu110
+torch==1.8.1+cu110
+# BONSAI changes end
 -f https://download.pytorch.org/whl/torch_stable.html
-torchvision==0.9.1+cu110 
+# BONSAI changes begin - torchvision==0.8.2+cu110
+torchvision==0.9.1+cu110
+# BONSAI changes end
 pip; python_version > "3.7"
diff --git a/python/requirements_rllib.txt b/python/requirements_rllib.txt
index 20ff12cb9b7d..d480a2968fed 100644
--- a/python/requirements_rllib.txt
+++ b/python/requirements_rllib.txt
@@ -1,20 +1,35 @@
+# BONSAI changes begin - added dependency
 tensorflow==2.5.0
+# BONSAI changes end
+# BONSAI changes begin - pinned dependency
 tensorflow_probability==0.11.1
-
+# BONSAI changes end
+# BONSAI changes begin - pinned dependency
 gast==0.4.0
+# BONSAI changes end
 # Version requirement to match Tune
+# BONSAI changes begin - torch>=1.6.0
 torch==1.8.1
+# BONSAI changes end
 # Version requirement to match Tune
+# BONSAI changes begin - torchvision>=0.6.0
 torchvision==0.9.1
+# BONSAI changes end
 # For auto-generating a rendering Window.
 pyglet
+# BONSAI changes begin - pinned dependency
 smart_open==5.0.0
+# BONSAI changes end
+# BONSAI changes begin - added dependencies
 gym==0.18.0
 msal==1.11.0
+# BONSAI changes end
 # For testing in MuJoCo-like envs (in PyBullet).
 pybullet
 # For tests on PettingZoo's multi-agent envs.
+# BONSAI changes begin - pettingzoo>=1.4.0
 pettingzoo==1.8.1
+# BONSAI changes end
 # For tests on RecSim and Kaggle envs.
 recsim
 kaggle_environments
@@ -25,6 +40,7 @@ higher
 # Unity3D testing
 mlagents
 mlagents_envs
+# BONSAI changes begin - added dependencies
 cryptography==3.4.7
 idna==2.8.0
 jsonschema==3.0
@@ -32,6 +48,7 @@ jsonschema==3.0
 six==1.15.0
 typing-extensions==3.7.4
 wrapt==1.12.1
+# BONSAI changes end
 
 
 # Ray Serve example
diff --git a/python/setup.py b/python/setup.py
index 691ad48ee5ea..e0f82da5664e 100644
--- a/python/setup.py
+++ b/python/setup.py
@@ -28,8 +28,9 @@
 # manually.
 
 SUPPORTED_PYTHONS = [(3, 6), (3, 7), (3, 8)]
+# BONSAI changes begin - SUPPORTED_BAZEL = (3, 2, 0)
 SUPPORTED_BAZEL = (3, 4, 1)
-
+# BONSAI changes end
 ROOT_DIR = os.path.dirname(__file__)
 BUILD_JAVA = os.getenv("RAY_INSTALL_JAVA") == "1"
 
@@ -106,13 +107,19 @@
     "dm_tree",
     "gym[atari]",
     "lz4",
+    # BONSAI changes begin - added dependency
     "opencv-python==4.5.2.52",
+    # BONSAI changes end
+    # BONSAI changes begin - "opencv-python-headless<=4.3.0.36",
     "opencv-python-headless==4.5.2.52",
+    # BONSAI changes end
     "pyyaml",
     "scipy",
+    # BONSAI changes begin - added dependencies
     "pyrr==0.10.3",
     "tensorflow==2.5.0",
     "tensorflow_probability==0.11.1",
+    # BONSAI changes end
 ]
 
 extras["all"] = list(set(chain.from_iterable(extras.values())))
@@ -124,29 +131,43 @@
     # TODO(alex) Pin the version once this PR is
     # included in the stable release.
     # https://github.com/aio-libs/aiohttp/pull/4556#issuecomment-679228562
+    # BONSAI changes begin - pinned dependency
     "aiohttp == 3.7.4",
+    # BONSAI changes end
     "aiohttp_cors",
+    # BONSAI changes begin - pinned dependency
     "aioredis == 1.3.1",
+    # BONSAI changes begin
     "click >= 7.0",
     "colorama",
     "dataclasses; python_version < '3.7'",
     "filelock",
     "gpustat",
     "grpcio >= 1.28.1",
+    # BONSAI changes begin - pinned dependency
     "jsonschema==3.0",
+    # BONSAI changes end
     "msgpack >= 1.0.0, < 2.0.0",
+    # BONSAI changes begin
+    # "numpy >= 1.16",
+    # "protobuf >= 3.15.3",
     "numpy ~= 1.19.2",
     "protobuf==3.15.3",
+    # BONSAI changes end
     "py-spy >= 0.2.0",
     "pyyaml",
     "requests",
     "redis >= 3.5.0",
     "opencensus",
+    # BONSAI changes begin -"prometheus_client >= 0.7.1",
     "prometheus_client==0.11.0",
+    # BONSAI changes end
+    # BONSAI changes begin - added dependencies
     "idna==2.8.0",
     "six==1.15.0",
     "typing-extensions==3.7.4",
     "wrapt==1.12.1",
+    # BONSAI changes end
 ]