Merge pull request #823 from bouthilx/hotfix/reduce_db_io

Reduce DB I/O during optimization

setup.py

@@ -11,7 +11,7 @@
 repo_root = os.path.dirname(os.path.abspath(__file__))
 
 
-tests_require = ["pytest>=3.0.0", "scikit-learn"]
+tests_require = ["pytest>=3.0.0", "scikit-learn", "ptera>=1.1.0"]
 
 
 packages = [  # Packages must be sorted alphabetically to ease maintenance and merges.

src/orion/client/runner.py

@@ -283,7 +283,7 @@ def run(self):
     def should_sample(self):
         """Check if more trials could be generated"""
 
-        if self.is_broken or self.is_done:
+        if self.free_worker <= 0 or (self.is_broken or self.is_done):
             return 0
 
         pending = len(self.pending_trials) + self.trials
@@ -332,8 +332,9 @@ def gather(self):
 
         to_be_raised = None
         log.debug(f"Gathered new results {len(results)}")
-
         # register the results
+        # NOTE: For Ptera instrumentation
+        trials = 0  # pylint:disable=unused-variable
         for result in results:
             trial = self.pending_trials.pop(result.future)
 
@@ -342,6 +343,8 @@ def gather(self):
                     # NB: observe release the trial already
                     self.client.observe(trial, result.value)
                     self.trials += 1
+                    # NOTE: For Ptera instrumentation
+                    trials = self.trials  # pylint:disable=unused-variable
                 except InvalidResult as exception:
                     # stop the optimization process if we received `InvalidResult`
                     # as all the trials are assumed to be returning those
@@ -413,15 +416,19 @@ def _suggest_trials(self, count):
 
             # non critical errors
             except WaitingForTrials:
+                log.debug("Runner cannot sample because WaitingForTrials")
                 break
 
             except ReservationRaceCondition:
+                log.debug("Runner cannot sample because ReservationRaceCondition")
                 break
 
             except LockAcquisitionTimeout:
+                log.debug("Runner cannot sample because LockAcquisitionTimeout")
                 break
 
             except CompletedExperiment:
+                log.debug("Runner cannot sample because CompletedExperiment")
                 break
 
         return trials