def choose_best_instance(self, policy, workload, env):
random.seed(self.seed)
available_instances = env.get_available_instances()
value_by_coordinate = {}
for coordinate in Instance.coordinates():
value_by_coordinate[coordinate] = set()
for instance in available_instances:
for coordinate in Instance.coordinates():
value_by_coordinate[coordinate].add(
getattr(instance, coordinate))
print('search space ', value_by_coordinate)
coordinates = list(value_by_coordinate.keys())
random.shuffle(coordinates)
best_coordinates = {}
for coordinate in coordinates:
best_coordinates[coordinate] = None
best_instance = None
for coordinate in coordinates:
print('search for best', coordinate)
instances_with_run_results = []
for coordinate_value in value_by_coordinate[coordinate]:
print('test coordinate_value:', coordinate_value)
best_coordinates[coordinate] = coordinate_value
suitable_instances = find_suitable_instances(
available_instances, best_coordinates)
for suitable_instance in suitable_instances:
print('test suitable_instance', suitable_instance)
results = env.run_workload_on_instance(
workload, suitable_instance, self.runs_per_instance)
# allocation failed
if results.failure and results.mean_cost == 0.0:
continue
instances_with_run_results.append(results)
break
if len(instances_with_run_results) == 0:
print('not found any suitable_instance for coordinate: ',
coordinate)
break
best_instance = policy.choose_best_instance(
instances_with_run_results)
best_coordinates[coordinate] = getattr(best_instance, coordinate)
print('best coordinates is', best_coordinates)
return best_instance
def find_suitable_instances(available_instances, best_coordinates):
for instance in available_instances:
ok = True
for coordinate in Instance.coordinates():
best_value = best_coordinates[coordinate]
if best_value == None:
continue
if best_value != getattr(instance, coordinate):
ok = False
break
if ok:
yield instance