feat: add execution time threshold handling.

spot/Spot.py

@@ -37,7 +37,10 @@ def __init__(self, config_dir: str, aws_session):
             self.ctx, aws_session, self.config.function_name
         )
         self.recommendation_engine = RecommendationEngine(
-            function_invoker, self.payload_file_path, self.config.mem_bounds
+            function_invoker,
+            self.payload_file_path,
+            self.config.mem_bounds,
+            self.config.execution_time_threshold,
         )
         self.benchmark_name = os.path.basename(config_dir)
 

spot/benchmark_config.py

@@ -12,6 +12,7 @@ def __init__(self, f=None):
         self.mem_size: int
         self.mem_bounds: list
         self.nickname: str
+        self.execution_time_threshold: float
 
         if f is not None:
             self.deserialize(f)
@@ -29,6 +30,7 @@ def _set_properties(
         region: str,
         random_seed: int,
         nickname: str,
+        execution_time_threshold: float = None,
         mem_bounds: Optional[list] = None,
     ):
         self.function_name = function_name
@@ -37,6 +39,7 @@ def _set_properties(
         self.mem_bounds = DEFAULT_MEM_BOUNDS if mem_bounds is None else mem_bounds
         self.random_seed = random_seed
         self.nickname = nickname
+        self.execution_time_threshold = execution_time_threshold
 
     def deserialize(self, f):
         try:

spot/exceptions/NoMemoryLeft.py

@@ -0,0 +1,2 @@
+class NoMemoryLeft(Exception):
+    pass

spot/invocation/aws_lambda_invoker.py

@@ -2,7 +2,7 @@
 import sys
 import time
 
-import botocore
+import botocore.exceptions
 import re
 import pandas as pd
 from concurrent.futures import ThreadPoolExecutor

spot/main.py

@@ -21,7 +21,7 @@ def main():
         help="Return best memory configuration for lowest cost",
     )
     parser.add_argument(
-        "--force", "-f", action="store_true", help="Ignore the invokation cache"
+        "--force", "-f", action="store_true", help="Ignore the invocation cache"
     )
     parser.add_argument(
         "--invoke", "-i", type=int, help="The number of times you invoke the function"

spot/recommendation_engine/recommendation_engine.py

@@ -1,6 +1,9 @@
+import sys
+
 import numpy as np
 import pandas as pd
 
+from spot.exceptions import NoMemoryLeft
 from spot.recommendation_engine.objectives import *
 from spot.recommendation_engine.utility import Utility
 from spot.constants import *
@@ -14,13 +17,20 @@ def __init__(self, memory, billed_time):
 
 
 class RecommendationEngine:
-    def __init__(self, invocator, payload_path, memory_range):
+    def __init__(
+        self,
+        invocator,
+        payload_path,
+        memory_range,
+        execution_time_threshold: float = None,
+    ):
         self.payload_path = payload_path
         self.function_invocator = invocator
         self.sampled_datapoints = []
         self.fitted_function = None
         self.function_parameters = {}
         self.memory_range = memory_range
+        self.execution_time_threshold = execution_time_threshold
         if OPTIMIZATION_OBJECTIVE == "normal":
             self.objective = NormalObjective(self, self.memory_range)
         elif OPTIMIZATION_OBJECTIVE == "fit_to_real_cost":
@@ -50,15 +60,26 @@ def run(self):
         return self.report()
 
     def report(self):
-        minimum_memory, minimum_cost = Utility.find_minimum_memory_cost(
-            self.fitted_function, self.function_parameters, self.memory_range
-        )
-        result = {
-            "Minimum Cost Memory": [minimum_memory],
-            "Expected Cost": [minimum_cost],
-            "Exploration Cost": [self.exploration_cost],
-        }
-        return pd.DataFrame.from_dict(result)
+        try:
+            minimum_memory, minimum_cost = Utility.find_minimum_memory_cost(
+                self.fitted_function,
+                self.function_parameters,
+                self.memory_range,
+                self.execution_time_threshold,
+            )
+            result = {
+                "Minimum Cost Memory": [minimum_memory],
+                "Expected Cost": [minimum_cost],
+                "Exploration Cost": [self.exploration_cost],
+            }
+            return pd.DataFrame.from_dict(result)
+
+        except NoMemoryLeft:
+            print(
+                "No memory configuration is possible. The execution time threshold is too low!",
+                file=sys.stderr,
+            )
+            exit(1)
 
     def initial_sample(self):
         # TODO: ensure initial memories are in the accepted memory range.

spot/recommendation_engine/utility.py

@@ -2,12 +2,27 @@
 import numpy as np
 from scipy.optimize import curve_fit
 
+from spot.exceptions.NoMemoryLeft import NoMemoryLeft
+
 
 class Utility:
     @staticmethod
-    def find_minimum_memory_cost(f, params, memory_range):
+    def find_minimum_memory_cost(
+        f, params, memory_range, execution_time_threshold: float = None
+    ):
         mems = np.arange(memory_range[0], memory_range[1] + 1, dtype=np.double)
         costs = f(mems, *params)
+
+        # Handling execution threshold
+        if execution_time_threshold is not None:
+            execution_times = costs / mems
+            for i in range(len(execution_times)):
+                if execution_times[i] > execution_time_threshold:
+                    mems = np.delete(mems, i)
+                    costs = np.delete(costs, i)
+            if len(mems) == 0:
+                raise NoMemoryLeft()
+
         min_index = np.argmin(costs)
         return mems[min_index], costs[min_index]