Revert "[Serve] [doc] Improve runtime env doc (#18782)" (#18935)

This reverts commit d4d71985d5.

.buildkite/pipeline.macos.yml

@@ -64,7 +64,7 @@ steps:
   commands:
     - *prelude_commands
     - TORCH_VERSION=1.6 ./ci/travis/install-dependencies.sh
-    - bazel test --config=ci --test_env=CI $(./scripts/bazel_export_options) --build_tests_only --test_tag_filters=-flaky,-flaky-mac,-post_wheel_build --
+    - bazel test --config=ci --test_env=CI $(./scripts/bazel_export_options) --build_tests_only --test_tag_filters=-flaky,-flaky-mac --
       //:all python/ray/serve/... python/ray/dashboard/... -rllib/... -core_worker_test
     - *epilogue_commands
 

.buildkite/pipeline.yml

@@ -182,9 +182,7 @@
     - TORCH_VERSION=1.6 ./ci/travis/install-dependencies.sh
     - ./dashboard/tests/run_ui_tests.sh
     - bazel test --config=ci $(./scripts/bazel_export_options) python/ray/dashboard/...
-    - bazel test --config=ci $(./scripts/bazel_export_options)
+    - bazel test --config=ci $(./scripts/bazel_export_options) python/ray/serve/...
-      --test_tag_filters=-post_wheel_build
-      python/ray/serve/...
 
 - label: ":python: Minimal install"
   conditions: ["RAY_CI_PYTHON_AFFECTED"]

doc/source/serve/core-apis.rst

@@ -238,31 +238,27 @@ Ray Serve supports serving deployments with different (possibly conflicting)
 Python dependencies.  For example, you can simultaneously serve one deployment
 that uses legacy Tensorflow 1 and another that uses Tensorflow 2.
 
-This is supported on Mac OS and Linux using Ray's :ref:`runtime-environments` feature.
+Currently this is supported on Mac OS and Linux using `conda <https://docs.conda.io/en/latest/>`_
-As with all other Ray actor options, pass the runtime environment in via ``ray_actor_options`` in
+via Ray's built-in ``runtime_env`` option for actors.
-your deployment.  Be sure to first run ``pip install "ray[default]"`` to ensure the
+As with all other actor options, pass these in via ``ray_actor_options`` in
-Runtime Environments feature is installed.
+your deployment.
+You must have a conda environment set up for each set of
+dependencies you want to isolate.  If using a multi-node cluster, the
+desired conda environment must be present on all nodes. Also, the Python patch version
+(e.g. 3.8.10) must be identical on all nodes (this is a requirement for any Ray cluster).
+See :ref:`runtime-environments` for details.
 
-Example:
+Here's an example script.  For it to work, first create a conda
+environment named ``ray-tf1`` with Ray Serve and Tensorflow 1 installed,
+and another named ``ray-tf2`` with Ray Serve and Tensorflow 2.  The Ray and
+Python versions must be the same in both environments.
 
 .. literalinclude:: ../../../python/ray/serve/examples/doc/conda_env.py
 
-.. note::
-  When using a Ray library (for example, Ray Serve) in a runtime environment, it must
-  explicitly be included in the dependencies, as in the above example.  This is not
-  required when just using Ray Core.
-
-.. tip::
-  Avoid dynamically installing packages that install from source: these can be slow and
-  use up all resources while installing, leading to problems with the Ray cluster.  Consider
-  precompiling such packages in a private repository or Docker image.
-
 The dependencies required in the deployment may be different than
 the dependencies installed in the driver program (the one running Serve API
 calls). In this case, you should use a delayed import within the class to avoid
-importing unavailable packages in the driver.  This applies even when not
+importing unavailable packages in the driver.
-using runtime environments.
-
 Example:
 
 .. literalinclude:: ../../../python/ray/serve/examples/doc/imported_backend.py

python/ray/serve/BUILD

@@ -338,11 +338,3 @@ py_test(
     tags = ["exclusive", "team:serve"],
     deps = [":serve_lib"]
 )
-
-py_test(
-    name = "conda_env",
-    size = "medium",
-    srcs = glob(["examples/doc/*.py"]),
-    tags = ["exclusive", "post_wheel_build", "team:serve"],
-    deps = [":serve_lib"]
-)

python/ray/serve/backend_state.py

@@ -1073,9 +1073,8 @@ class BackendState:
                 f"Deployment '{self._name}' has "
                 f"{len(slow_start_replicas)} replicas that have taken "
                 f"more than {SLOW_STARTUP_WARNING_S}s to start up. This "
-                "may be caused by waiting for the cluster to auto-scale, "
+                "may be caused by waiting for the cluster to auto-scale "
-                "waiting for a runtime environment to install, or a slow "
+                "or because the constructor is slow. Resources required "
-                "constructor. Resources required "
                 f"for each replica: {required}, resources available: "
                 f"{available}. component=serve deployment={self._name}")
 

python/ray/serve/examples/doc/conda_env.py

@@ -1,28 +1,27 @@
 import requests
 from ray import serve
+import tensorflow as tf
 
 serve.start()
 
 
 @serve.deployment
-def requests_version(request):
+def tf_version(request):
-    return requests.__version__
+    return ("Tensorflow " + tf.__version__)
 
 
-requests_version.options(
+tf_version.options(
-    name="25",
+    name="tf1", ray_actor_options={
-    ray_actor_options={
         "runtime_env": {
-            "pip": ["ray[serve]", "requests==2.25.1"]
+            "conda": "ray-tf1"
         }
     }).deploy()
-requests_version.options(
+tf_version.options(
-    name="26",
+    name="tf2", ray_actor_options={
-    ray_actor_options={
         "runtime_env": {
-            "pip": ["ray[serve]", "requests==2.26.0"]
+            "conda": "ray-tf2"
         }
     }).deploy()
 
-assert requests.get("http://127.0.0.1:8000/25").text == "2.25.1"
+print(requests.get("http://127.0.0.1:8000/tf1").text)  # Tensorflow 1.15.0
-assert requests.get("http://127.0.0.1:8000/26").text == "2.26.0"
+print(requests.get("http://127.0.0.1:8000/tf2").text)  # Tensorflow 2.3.0