Merge branch 'master' into ray_cg_target

.buildkite/rllib.rayci.yml

@@ -51,7 +51,7 @@ steps:
 
   - label: ":brain: rllib: examples"
     tags: rllib
-    parallelism: 5
+    parallelism: 6
     instance_type: large
     commands:
       - bazel run //ci/ray_ci:test_in_docker -- //rllib/... rllib

doc/source/data/api/input_output.rst

@@ -213,6 +213,15 @@ Lance
 
    read_lance
 
+ClickHouse
+----------
+
+.. autosummary::
+   :nosignatures:
+   :toctree: doc/
+
+   read_clickhouse
+
 Dask
 ----
 

doc/source/rllib/rllib-models.rst

@@ -390,29 +390,6 @@ Take a look at this model example that does exactly that:
    :end-before: __sphinx_doc_end__
 
 
-**Using the Trajectory View API: Passing in the last n actions (or rewards or observations) as inputs to a custom Model**
-
-It is sometimes helpful for learning not only to look at the current observation
-in order to calculate the next action, but also at the past n observations.
-In other cases, you may want to provide the most recent rewards or actions to the model as well
-(like our LSTM wrapper does if you specify: ``use_lstm=True`` and ``lstm_use_prev_action/reward=True``).
-All this may even be useful when not working with partially observable environments (PO-MDPs)
-and/or RNN/Attention models, as for example in classic Atari runs, where we usually use framestacking of
-the last four observed images.
-
-The trajectory view API allows your models to specify these more complex "view requirements".
-
-Here is a simple (non-RNN/Attention) example of a Model that takes as input
-the last 3 observations (very similar to the recommended "framestacking" for
-learning in Atari environments):
-
-.. literalinclude:: ../../../rllib/examples/_old_api_stack/models/trajectory_view_utilizing_models.py
-   :language: python
-   :start-after: __sphinx_doc_begin__
-   :end-before: __sphinx_doc_end__
-
-A PyTorch version of the above model is also `given in the same file <https://github.com/ray-project/ray/blob/master/rllib/examples/_old_api_stack/models/trajectory_view_utilizing_models.py>`__.
-
 
 Custom Action Distributions
 ---------------------------

doc/source/train/user-guides/using-gpus.rst

@@ -228,21 +228,6 @@ an actor for the :class:`Trainer <ray.train.trainer.BaseTrainer>` object when
 you call :meth:`Trainer.fit() <ray.train.trainer.BaseTrainer.fit>`.
 
 This object often only manages lightweight communication between the training workers.
-You can still specify its resources, which can be useful if you implemented your own
-Trainer that does heavier processing.
-
-.. testcode::
-
-    from ray.train import ScalingConfig
-
-    scaling_config = ScalingConfig(
-        num_workers=8,
-        trainer_resources={
-            "CPU": 4,
-            "GPU": 1,
-        }
-    )
-
 Per default, a trainer uses 1 CPU. If you have a cluster with 8 CPUs and want
 to start 4 training workers a 2 CPUs, this will not work, as the total number
 of required CPUs will be 9 (4 * 2 + 1). In that case, you can specify the trainer

python/ray/air/config.py

@@ -102,6 +102,8 @@ def equals(value, default_value):
 class ScalingConfig:
     """Configuration for scaling training.
 
+    For more details, see :ref:`train_scaling_config`.
+
     Args:
         trainer_resources: Resources to allocate for the training coordinator.
             The training coordinator launches the worker group and executes
@@ -120,8 +122,9 @@ class ScalingConfig:
             argument.
         resources_per_worker: If specified, the resources
             defined in this Dict is reserved for each worker.
-            Define the ``"CPU"`` and ``"GPU"`` keys (case-sensitive) to
-            override the number of CPU or GPUs used by each worker.
+            Define the ``"CPU"`` key (case-sensitive) to
+            override the number of CPUs used by each worker.
+            This can also be used to request :ref:`custom resources <custom-resources>`.
         placement_strategy: The placement strategy to use for the
             placement group of the Ray actors. See :ref:`Placement Group
             Strategies <pgroup-strategy>` for the possible options.
@@ -142,8 +145,8 @@ class ScalingConfig:
                 num_workers=2,
                 # Turn on/off GPU.
                 use_gpu=True,
-                # Specify resources used for trainer.
-                trainer_resources={"CPU": 1},
+                # Assign extra CPU/GPU/custom resources per worker.
+                resources_per_worker={"GPU": 1, "CPU": 1, "memory": 1e9, "custom": 1.0},
                 # Try to schedule workers on different nodes.
                 placement_strategy="SPREAD",
             )

python/ray/dashboard/client/src/type/serve.ts

@@ -26,6 +26,9 @@ export enum ServeDeploymentStatus {
   UPDATING = "UPDATING",
   HEALTHY = "HEALTHY",
   UNHEALTHY = "UNHEALTHY",
+  DEPLOY_FAILED = "DEPLOY_FAILED",
+  UPSCALING = "UPSCALING",
+  DOWNSCALING = "DOWNSCALING",
 }
 
 export type ServeDeployment = {

python/ray/data/BUILD

@@ -611,6 +611,14 @@ py_test(
     deps = ["//:ray_lib", ":conftest"],
 )
 
+py_test(
+    name = "test_clickhouse",
+    size = "small",
+    srcs = ["tests/test_clickhouse.py"],
+    tags = ["team:data", "exclusive"],
+    deps = ["//:ray_lib", ":conftest"],
+)
+
 py_test(
     name = "test_lance",
     size = "small",

python/ray/data/__init__.py

@@ -44,6 +44,7 @@
     read_avro,
     read_bigquery,
     read_binary_files,
+    read_clickhouse,
     read_csv,
     read_databricks_tables,
     read_datasource,
@@ -137,6 +138,7 @@
     "read_avro",
     "read_text",
     "read_binary_files",
+    "read_clickhouse",
     "read_csv",
     "read_datasource",
     "read_delta_sharing_tables",