[tune] Reporter crash fix (#5426)

Co-authored-by: Richard Liaw <rliaw@berkeley.edu>

python/ray/tune/tests/test_trial_runner.py

@@ -459,6 +459,15 @@ class TrainableFunctionApiTest(unittest.TestCase):
         self.assertEqual(trial.status, Trial.TERMINATED)
         self.assertEqual(trial.last_result[TIMESTEPS_TOTAL], 100)
 
+    def testReporterNoUsage(self):
+        def run_task(config, reporter):
+            print("hello")
+
+        experiment = Experiment(run=run_task, name="ray_crash_repro")
+        [trial] = ray.tune.run(experiment).trials
+        print(trial.last_result)
+        self.assertEqual(trial.last_result[DONE], True)
+
     def testErrorReturn(self):
         def train(config, reporter):
             raise Exception("uh oh")

python/ray/tune/trial.py

@@ -321,7 +321,7 @@ class Trial(object):
                     location_string(
                         self.last_result.get(HOSTNAME),
                         self.last_result.get(PID))), "{} s".format(
-                            int(self.last_result.get(TIME_TOTAL_S)))
+                            int(self.last_result.get(TIME_TOTAL_S, 0)))
         ]
 
         if self.last_result.get(TRAINING_ITERATION) is not None:

python/ray/tune/trial_runner.py

@@ -506,7 +506,7 @@ class TrialRunner(object):
                 result = trial.last_result
                 result.update(done=True)
 
-            self._total_time += result[TIME_THIS_ITER_S]
+            self._total_time += result.get(TIME_THIS_ITER_S, 0)
 
             flat_result = flatten_dict(result)
             if trial.should_stop(flat_result):

rllib/evaluation/postprocessing.py

@@ -28,7 +28,7 @@ def compute_advantages(rollout, last_r, gamma=0.9, lambda_=1.0, use_gae=True):
         last_r (float): Value estimation for last observation
         gamma (float): Discount factor.
         lambda_ (float): Parameter for GAE
-        use_gae (bool): Using Generalized Advantage Estamation
+        use_gae (bool): Using Generalized Advantage Estimation
 
     Returns:
         SampleBatch (SampleBatch): Object with experience from rollout and