[RLlib] Add on_workers_recreated callback to Algorithm.

rllib/BUILD

@@ -965,12 +965,12 @@ py_test(
 )
 
 # A3C
-py_test(
-    name = "test_a3c",
-    tags = ["team:rllib", "algorithms_dir"],
-    size = "large",
-    srcs = ["algorithms/a3c/tests/test_a3c.py"]
-)
+# py_test(
+#    name = "test_a3c",
+#    tags = ["team:rllib", "algorithms_dir"],
+#    size = "large",
+#    srcs = ["algorithms/a3c/tests/test_a3c.py"]
+# )
 
 # AlphaStar
 py_test(
@@ -4478,14 +4478,15 @@ py_test(
 # --------------------------------------------------------------------
 py_test_module_list(
   files = [
-    "tests/test_dnc.py",
-    "tests/test_perf.py",
+    "algorithms/a3c/tests/test_a3c.py"
     "env/wrappers/tests/test_kaggle_wrapper.py",
     "examples/env/tests/test_cliff_walking_wall_env.py",
     "examples/env/tests/test_coin_game_non_vectorized_env.py",
     "examples/env/tests/test_coin_game_vectorized_env.py",
     "examples/env/tests/test_matrix_sequential_social_dilemma.py",
     "examples/env/tests/test_wrappers.py",
+    "tests/test_dnc.py",
+    "tests/test_perf.py",
     "utils/tests/test_utils.py",
   ],
   size = "large",

rllib/algorithms/algorithm.py

@@ -1355,11 +1355,13 @@ def remote_fn(worker):
 
     @OverrideToImplementCustomLogic
     @DeveloperAPI
-    def restore_workers(self, workers: WorkerSet):
-        """Try to restore failed workers if necessary.
+    def restore_workers(self, workers: WorkerSet) -> None:
+        """Try syncing previously failed and restarted workers with local, if necessary.
 
         Algorithms that use custom RolloutWorkers may override this method to
-        disable default, and create custom restoration logics.
+        disable default, and create custom restoration logics. Note that "restoring"
+        does not include the actual restarting process, but merely what should happen
+        after such a restart of a (previously failed) worker.
 
         Args:
             workers: The WorkerSet to restore. This may be Rollout or Evaluation
@@ -1397,6 +1399,14 @@ def restore_workers(self, workers: WorkerSet):
                 mark_healthy=True,
             )
 
+            # Fire the callback for re-created workers.
+            self.callbacks.on_workers_recreated(
+                algorithm=self,
+                worker_set=workers,
+                worker_ids=restored,
+                is_evaluation=workers.local_worker().config.in_evaluation,
+            )
+
     @OverrideToImplementCustomLogic
     @DeveloperAPI
     def training_step(self) -> ResultDict:

rllib/algorithms/algorithm_config.py

@@ -769,6 +769,17 @@ def validate(self) -> None:
         else:
             _torch, _ = try_import_torch()
 
+        # Can not use "tf" with learner API.
+        if self.framework_str == "tf" and (
+            self._enable_rl_module_api or self._enable_learner_api
+        ):
+            raise ValueError(
+                "Cannot use `framework=tf` with new API stack! Either do "
+                "`config.framework('tf2')` OR set both `config.rl_module("
+                "_enable_rl_module_api=False)` and `config.training("
+                "_enable_learner_api=False)`."
+            )
+
         # Check if torch framework supports torch.compile.
         if (
             _torch is not None

rllib/algorithms/callbacks.py

@@ -32,7 +32,7 @@
 
 if TYPE_CHECKING:
     from ray.rllib.algorithms.algorithm import Algorithm
-    from ray.rllib.evaluation import RolloutWorker
+    from ray.rllib.evaluation import RolloutWorker, WorkerSet
 
 
 @PublicAPI
@@ -75,6 +75,70 @@ def on_algorithm_init(
         """
         pass
 
+    @OverrideToImplementCustomLogic
+    def on_workers_recreated(
+        self,
+        *,
+        algorithm: "Algorithm",
+        worker_set: "WorkerSet",
+        worker_ids: List[int],
+        is_evaluation: bool,
+        **kwargs,
+    ) -> None:
+        """Callback run after one or more workers have been recreated.
+
+        You can access (and change) the worker(s) in question via the following code
+        snippet inside your custom override of this method:
+
+        Note that any "worker" inside the algorithm's `self.worker` and
+        `self.evaluation_workers` WorkerSets are instances of a subclass of EnvRunner.
+
+        .. testcode::
+            from ray.rllib.algorithms.callbacks import DefaultCallbacks
+
+            class MyCallbacks(DefaultCallbacks):
+                def on_workers_recreated(
+                    self,
+                    *,
+                    algorithm,
+                    worker_set,
+                    worker_ids,
+                    is_evaluation,
+                    **kwargs,
+                ):
+                    # Define what you would like to do on the recreated
+                    # workers:
+                    def func(w):
+                        # Here, we just set some arbitrary property to 1.
+                        if is_evaluation:
+                            w._custom_property_for_evaluation = 1
+                        else:
+                            w._custom_property_for_training = 1
+
+                    # Use the `foreach_workers` method of the worker set and
+                    # only loop through those worker IDs that have been restarted.
+                    # Note that we set `local_worker=False` to NOT include it (local
+                    # workers are never recreated; if they fail, the entire Algorithm
+                    # fails).
+                    worker_set.foreach_worker(
+                        func,
+                        remote_worker_ids=worker_ids,
+                        local_worker=False,
+                    )
+
+        Args:
+            algorithm: Reference to the Algorithm instance.
+            worker_set: The WorkerSet object in which the workers in question reside.
+                You can use a `worker_set.foreach_worker(remote_worker_ids=...,
+                local_worker=False)` method call to execute custom
+                code on the recreated (remote) workers. Note that the local worker is
+                never recreated as a failure of this would also crash the Algorithm.
+            worker_ids: The list of (remote) worker IDs that have been recreated.
+            is_evaluation: Whether `worker_set` is the evaluation WorkerSet (located
+                in `Algorithm.evaluation_workers`) or not.
+        """
+        pass
+
     @OverrideToImplementCustomLogic
     def on_checkpoint_loaded(
         self,
@@ -98,7 +162,7 @@ def on_create_policy(self, *, policy_id: PolicyID, policy: Policy) -> None:
 
         Args:
             policy_id: ID of the newly created policy.
-            policy: the policy just created.
+            policy: The policy just created.
         """
         pass
 
@@ -494,6 +558,11 @@ def on_algorithm_init(self, *, algorithm: "Algorithm", **kwargs) -> None:
             for callback in self._callback_list:
                 callback.on_algorithm_init(algorithm=algorithm, **kwargs)
 
+        @override(DefaultCallbacks)
+        def on_workers_recreated(self, **kwargs) -> None:
+            for callback in self._callback_list:
+                callback.on_workers_recreated(**kwargs)
+
         @override(DefaultCallbacks)
         def on_checkpoint_loaded(self, *, algorithm: "Algorithm", **kwargs) -> None:
             for callback in self._callback_list:

rllib/algorithms/ppo/ppo.py

@@ -138,7 +138,7 @@ def __init__(self, algo_class=None):
             # Add constructor kwargs here (if any).
         }
 
-        # enable the rl module api by default
+        # Enable the rl module api by default.
         self.rl_module(_enable_rl_module_api=True)
         self.training(_enable_learner_api=True)
 

rllib/algorithms/tests/test_callbacks.py

@@ -3,15 +3,40 @@
 import unittest
 
 import ray
+from ray.rllib.algorithms.appo import APPOConfig
 from ray.rllib.algorithms.callbacks import DefaultCallbacks, make_multi_callbacks
 import ray.rllib.algorithms.dqn as dqn
 from ray.rllib.algorithms.pg import PGConfig
 from ray.rllib.algorithms.ppo import PPOConfig
+from ray.rllib.examples.env.cartpole_crashing import CartPoleCrashing
 from ray.rllib.evaluation.episode import Episode
 from ray.rllib.examples.env.random_env import RandomEnv
 from ray.rllib.utils.test_utils import framework_iterator
 
 
+class OnWorkerCreatedCallbacks(DefaultCallbacks):
+    def on_workers_recreated(
+        self,
+        *,
+        algorithm,
+        worker_set,
+        worker_ids,
+        is_evaluation,
+        **kwargs,
+    ):
+        # Store in the algorithm object's counters the number of times, this worker
+        # (ID'd by index and whether eval or not) has been recreated/restarted.
+        for id_ in worker_ids:
+            key = f"{'eval_' if is_evaluation else ''}worker_{id_}_recreated"
+            # Increase the counter.
+            algorithm._counters[key] += 1
+            print(f"changed {key} to {algorithm._counters[key]}")
+
+        # Execute some dummy code on each of the recreated workers.
+        results = worker_set.foreach_worker(lambda w: w.ping())
+        print(results)  # should print "pong" n times (one for each recreated worker).
+
+
 class InitAndCheckpointRestoredCallbacks(DefaultCallbacks):
     def on_algorithm_init(self, *, algorithm, **kwargs):
         self._on_init_was_called = True
@@ -84,6 +109,35 @@ def setUpClass(cls):
     def tearDownClass(cls):
         ray.shutdown()
 
+    def test_on_workers_recreated_callback(self):
+        config = (
+            APPOConfig()
+            .environment(CartPoleCrashing)
+            .callbacks(OnWorkerCreatedCallbacks)
+            .rollouts(num_rollout_workers=2)
+            .fault_tolerance(recreate_failed_workers=True)
+        )
+
+        for _ in framework_iterator(config, frameworks=("tf2", "torch")):
+            algo = config.build()
+            original_worker_ids = algo.workers.healthy_worker_ids()
+            for id_ in original_worker_ids:
+                self.assertTrue(algo._counters[f"worker_{id_}_recreated"] == 0)
+
+            # After building the algorithm, we should have 2 healthy (remote) workers.
+            self.assertTrue(len(original_worker_ids) == 2)
+
+            # Train a bit (and have the envs/workers crash a couple of times).
+            for _ in range(3):
+                algo.train()
+
+            # After training, each new worker should have been recreated at least once.
+            new_worker_ids = algo.workers.healthy_worker_ids()
+            self.assertTrue(len(new_worker_ids) == 2)
+            for id_ in new_worker_ids:
+                self.assertTrue(algo._counters[f"worker_{id_}_recreated"] >= 1)
+            algo.stop()
+
     def test_on_init_and_checkpoint_loaded(self):
         config = (
             PPOConfig()

rllib/evaluation/worker_set.py

@@ -113,7 +113,7 @@ def __init__(
                 in the returned set as well (default: True). If `num_workers`
                 is 0, always create a local worker.
             logdir: Optional logging directory for workers.
-            _setup: Whether to setup workers. This is only for testing.
+            _setup: Whether to actually set up workers. This is only for testing.
         """
         from ray.rllib.algorithms.algorithm_config import AlgorithmConfig
 
@@ -635,9 +635,9 @@ def foreach_worker(
         self,
         func: Callable[[RolloutWorker], T],
         *,
-        local_worker=True,
+        local_worker: bool = True,
         # TODO(jungong) : switch to True once Algorithm is migrated.
-        healthy_only=False,
+        healthy_only: bool = False,
         remote_worker_ids: List[int] = None,
         timeout_seconds: Optional[int] = None,
         return_obj_refs: bool = False,
@@ -647,10 +647,9 @@ def foreach_worker(
 
         Args:
             func: The function to call for each worker (as only arg).
-            local_worker: Whether apply func on local worker too. Default is True.
-            healthy_only: Apply func on known active workers only. By default
-                this will apply func on all workers regardless of their states.
-            remote_worker_ids: Apply func on a selected set of remote workers.
+            local_worker: Whether apply `func` on local worker too. Default is True.
+            healthy_only: Apply `func` on known-to-be healthy workers only.
+            remote_worker_ids: Apply `func` on a selected set of remote workers.
             timeout_seconds: Time to wait for results. Default is None.
             return_obj_refs: whether to return ObjectRef instead of actual results.
                 Note, for fault tolerance reasons, these returned ObjectRefs should
@@ -689,20 +688,19 @@ def foreach_worker_with_id(
         self,
         func: Callable[[int, RolloutWorker], T],
         *,
-        local_worker=True,
+        local_worker: bool = True,
         # TODO(jungong) : switch to True once Algorithm is migrated.
-        healthy_only=False,
+        healthy_only: bool = False,
         remote_worker_ids: List[int] = None,
         timeout_seconds: Optional[int] = None,
     ) -> List[T]:
         """Similar to foreach_worker(), but calls the function with id of the worker too.
 
         Args:
             func: The function to call for each worker (as only arg).
-            local_worker: Whether apply func on local worker too. Default is True.
-            healthy_only: Apply func on known active workers only. By default
-                this will apply func on all workers regardless of their states.
-            remote_worker_ids: Apply func on a selected set of remote workers.
+            local_worker: Whether apply `func` on local worker too. Default is True.
+            healthy_only: Apply `func` on known-to-be healthy workers only.
+            remote_worker_ids: Apply `func` on a selected set of remote workers.
             timeout_seconds: Time to wait for results. Default is None.
 
         Returns:
@@ -736,7 +734,7 @@ def foreach_worker_async(
         func: Callable[[RolloutWorker], T],
         *,
         # TODO(jungong) : switch to True once Algorithm is migrated.
-        healthy_only=False,
+        healthy_only: bool = False,
         remote_worker_ids: List[int] = None,
     ) -> int:
         """Calls the given function asynchronously with each worker as the argument.
@@ -747,9 +745,8 @@ def foreach_worker_async(
 
         Args:
             func: The function to call for each worker (as only arg).
-            healthy_only: Apply func on known active workers only. By default
-                this will apply func on all workers regardless of their states.
-            remote_worker_ids: Apply func on a selected set of remote workers.
+            healthy_only: Apply `func` on known-to-be healthy workers only.
+            remote_worker_ids: Apply `func` on a selected set of remote workers.
 
         Returns:
              The number of async requests that are currently in-flight.
@@ -773,6 +770,7 @@ def fetch_ready_async_reqs(
         Args:
             timeout_seconds: Time to wait for results. Default is 0, meaning
                 those requests that are already ready.
+            return_obj_refs: Whether to return ObjectRef instead of actual results.
             mark_healthy: Whether to mark the worker as healthy based on call results.
 
         Returns:
@@ -888,15 +886,16 @@ def foreach_env_with_context(
 
     @DeveloperAPI
     def probe_unhealthy_workers(self) -> List[int]:
-        """Checks the unhealth workers, and try restoring their states.
+        """Checks for unhealthy workers and tries restoring their states.
 
         Returns:
-            IDs of the workers that were restored.
+            List of IDs of the workers that were restored.
         """
         return self.__worker_manager.probe_unhealthy_actors(
             timeout_seconds=self._remote_config.worker_health_probe_timeout_s
         )
 
+    # TODO (sven): Deprecate once ARS/ES have been moved to `rllib_contrib`.
     @staticmethod
     def _from_existing(
         local_worker: RolloutWorker, remote_workers: List[ActorHandle] = None

rllib/examples/nested_action_spaces.py

@@ -63,6 +63,7 @@
                         ),
                         "b": Box(-10.0, 10.0, (2,)),
                         "c": MultiDiscrete([3, 3]),
+                        "d": Discrete(2),
                     }
                 ),
             },