[tune][minor] gpu warning (#5948)

* gpu

* formaat

* defaults

* format_and_check

* better registration

* fix

* fix

* trial

* foramt

* tune

python/ray/tune/ray_trial_executor.py

@@ -93,7 +93,7 @@ class RayTrialExecutor(TrialExecutor):
                 memory=trial.resources.memory,
                 object_store_memory=trial.resources.object_store_memory,
                 resources=trial.resources.custom_resources)(
-                    trial._get_trainable_cls())
+                    trial.get_trainable_cls())
 
         trial.init_logger()
         # We checkpoint metadata here to try mitigating logdir duplication
@@ -622,6 +622,11 @@ class RayTrialExecutor(TrialExecutor):
                 trial.runner.export_model.remote(trial.export_formats))
         return {}
 
+    def has_gpus(self):
+        if self._resources_initialized:
+            self._update_avail_resources()
+            return self._avail_resources.gpu > 0
+
 
 def _to_gb(n_bytes):
     return round(n_bytes / (1024**3), 2)

python/ray/tune/registry.py

@@ -7,6 +7,7 @@ from types import FunctionType
 
 import ray
 import ray.cloudpickle as pickle
+
 from ray.experimental.internal_kv import _internal_kv_initialized, \
     _internal_kv_get, _internal_kv_put
 
@@ -23,6 +24,24 @@ KNOWN_CATEGORIES = [
 logger = logging.getLogger(__name__)
 
 
+def has_trainable(trainable_name):
+    return _global_registry.contains(TRAINABLE_CLASS, trainable_name)
+
+
+def get_trainable_cls(trainable_name):
+    validate_trainable(trainable_name)
+    return _global_registry.get(TRAINABLE_CLASS, trainable_name)
+
+
+def validate_trainable(trainable_name):
+    if not has_trainable(trainable_name):
+        # Make sure rllib agents are registered
+        from ray import rllib  # noqa: F401
+        from ray.tune.error import TuneError
+        if not has_trainable(trainable_name):
+            raise TuneError("Unknown trainable: " + trainable_name)
+
+
 def register_trainable(name, trainable):
     """Register a trainable function or class.
 

python/ray/tune/trial.py

@@ -10,13 +10,12 @@ import uuid
 import time
 import tempfile
 import os
-import ray
 from ray.tune import TuneError
 from ray.tune.logger import pretty_print, UnifiedLogger
 # NOTE(rkn): We import ray.tune.registry here instead of importing the names we
 # need because there are cyclic imports that may cause specific names to not
 # have been defined yet. See https://github.com/ray-project/ray/issues/1716.
-import ray.tune.registry
+from ray.tune.registry import get_trainable_cls, validate_trainable
 from ray.tune.result import DEFAULT_RESULTS_DIR, DONE, TRAINING_ITERATION
 from ray.utils import binary_to_hex, hex_to_binary
 from ray.tune.resources import Resources, json_to_resources, resources_to_json
@@ -30,11 +29,6 @@ def date_str():
     return datetime.today().strftime("%Y-%m-%d_%H-%M-%S")
 
 
-def has_trainable(trainable_name):
-    return ray.tune.registry._global_registry.contains(
-        ray.tune.registry.TRAINABLE_CLASS, trainable_name)
-
-
 class Checkpoint(object):
     """Describes a checkpoint of trial state.
 
@@ -126,7 +120,7 @@ class Trial(object):
         in ray.tune.config_parser.
         """
 
-        Trial._registration_check(trainable_name)
+        validate_trainable(trainable_name)
         # Trial config
         self.trainable_name = trainable_name
         self.trial_id = Trial.generate_id() if trial_id is None else trial_id
@@ -136,7 +130,7 @@ class Trial(object):
         #: Parameters that Tune varies across searches.
         self.evaluated_params = evaluated_params or {}
         self.experiment_tag = experiment_tag
-        trainable_cls = self._get_trainable_cls()
+        trainable_cls = self.get_trainable_cls()
         if trainable_cls and hasattr(trainable_cls,
                                      "default_resource_request"):
             default_resources = trainable_cls.default_resource_request(
@@ -202,14 +196,6 @@ class Trial(object):
         if trial_name_creator:
             self.custom_trial_name = trial_name_creator(self)
 
-    @classmethod
-    def _registration_check(cls, trainable_name):
-        if not has_trainable(trainable_name):
-            # Make sure rllib agents are registered
-            from ray import rllib  # noqa: F401
-            if not has_trainable(trainable_name):
-                raise TuneError("Unknown trainable: " + trainable_name)
-
     @classmethod
     def generate_id(cls):
         return str(uuid.uuid1().hex)[:8]
@@ -363,9 +349,8 @@ class Trial(object):
             return True
         return False
 
-    def _get_trainable_cls(self):
+    def get_trainable_cls(self):
-        return ray.tune.registry._global_registry.get(
+        return get_trainable_cls(self.trainable_name)
-            ray.tune.registry.TRAINABLE_CLASS, self.trainable_name)
 
     def set_verbose(self, verbose):
         self.verbose = verbose
@@ -430,6 +415,6 @@ class Trial(object):
             state[key] = cloudpickle.loads(hex_to_binary(state[key]))
 
         self.__dict__.update(state)
-        Trial._registration_check(self.trainable_name)
+        validate_trainable(self.trainable_name)
         if logger_started:
             self.init_logger()

python/ray/tune/trial_executor.py

@@ -226,3 +226,7 @@ class TrialExecutor(object):
         """
         raise NotImplementedError("Subclasses of TrialExecutor must provide "
                                   "export_trial_if_needed() method")
+
+    def has_gpus(self):
+        """Returns True if GPUs are detected on the cluster."""
+        return None

python/ray/tune/trial_runner.py

@@ -346,7 +346,7 @@ class TrialRunner(object):
                              "up. {}").format(
                                  trial.resources.summary_string(),
                                  self.trial_executor.resource_string(),
-                                 trial._get_trainable_cls().resource_help(
+                                 trial.get_trainable_cls().resource_help(
                                      trial.config)))
                 elif trial.status == Trial.PAUSED:
                     raise TuneError(

python/ray/tune/tune.py

@@ -10,7 +10,9 @@ from ray.tune.experiment import convert_to_experiment_list, Experiment
 from ray.tune.analysis import ExperimentAnalysis
 from ray.tune.suggest import BasicVariantGenerator
 from ray.tune.trial import Trial, DEBUG_PRINT_INTERVAL
+from ray.tune.trainable import Trainable
 from ray.tune.ray_trial_executor import RayTrialExecutor
+from ray.tune.registry import get_trainable_cls
 from ray.tune.syncer import wait_for_sync
 from ray.tune.trial_runner import TrialRunner
 from ray.tune.progress_reporter import CLIReporter, JupyterNotebookReporter
@@ -42,6 +44,13 @@ def _make_scheduler(args):
             args.scheduler, _SCHEDULERS.keys()))
 
 
+def _check_default_resources_override(run_identifier):
+    trainable_cls = get_trainable_cls(run_identifier)
+    return hasattr(trainable_cls, "default_resource_request") and (
+        trainable_cls.default_resource_request.__code__ !=
+        Trainable.default_resource_request.__code__)
+
+
 def run(run_or_experiment,
         name=None,
         stop=None,
@@ -250,6 +259,24 @@ def run(run_or_experiment,
     else:
         reporter = CLIReporter()
 
+    # User Warning for GPUs
+    if trial_executor.has_gpus():
+        if isinstance(resources_per_trial,
+                      dict) and "gpu" in resources_per_trial:
+            # "gpu" is manually set.
+            pass
+        elif _check_default_resources_override(run_identifier):
+            # "default_resources" is manually overriden.
+            pass
+        else:
+            logger.warning("Tune detects GPUs, but no trials are using GPUs. "
+                           "To enable trials to use GPUs, set "
+                           "tune.run(resources_per_trial={'gpu': 1}...) "
+                           "which allows Tune to expose 1 GPU to each trial. "
+                           "You can also override "
+                           "`Trainable.default_resource_request` if using the "
+                           "Trainable API.")
+
     last_debug = 0
     while not runner.is_finished():
         runner.step()