sandwich: Fixes a bug when served from cache request are before.

tools/android/loading/sandwich_prefetch.py

 # Copyright 2016 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 """
 Implements a task builder for benchmarking effects of NoState Prefetch.
 Noticeable steps of the task pipeline:
   * Save a WPR archive
   * Process the WPR archive to make all resources cacheable
   * Process cache archive to patch response headers back to their original
@@ skipping 396 matching lines @@
       runner_output_dir):
     trace_path = os.path.join(repeat_dir, sandwich_runner.TRACE_FILENAME)
 
     logging.info('loading trace: %s', trace_path)
     trace = loading_trace.LoadingTrace.FromJsonFile(trace_path)
 
     logging.info('verifying trace: %s', trace_path)
     run_output_verifier.VerifyTrace(trace)
 
     logging.info('extracting metrics from trace: %s', trace_path)
-    served_from_network_bytes = 0
-    served_from_cache_bytes = 0
-    urls_hitting_network = set()
+
+    # Gather response size per URLs.
     response_sizes = {}
     for request in sandwich_utils.FilterOutDataAndIncompleteRequests(
         trace.request_track.GetEvents()):
       # Ignore requests served from the blink's cache.
       if request.served_from_cache:
         continue
-      urls_hitting_network.add(request.url)
       if request.from_disk_cache:
         if request.url in cached_encoded_data_lengths:
           response_size = cached_encoded_data_lengths[request.url]
         else:
           # Some fat webpages may overflow the Memory cache, and so some
           # requests might be served from disk cache couple of times per page
           # load.
           logging.warning('Looks like could be served from memory cache: %s',
               request.url)
-          response_size = response_sizes[request.url]
-        served_from_cache_bytes += response_size
+          if request.url in response_sizes:
+            response_size = response_sizes[request.url]
       else:
         response_size = request.GetResponseTransportLength()
-        served_from_network_bytes += response_size
       response_sizes[request.url] = response_size
 
+    # Sums the served from cache/network bytes.
+    served_from_network_bytes = 0
+    served_from_cache_bytes = 0
+    urls_hitting_network = set()
+    for request in sandwich_utils.FilterOutDataAndIncompleteRequests(
+        trace.request_track.GetEvents()):
+      # Ignore requests served from the blink's cache.
+      if request.served_from_cache:
+        continue
+      urls_hitting_network.add(request.url)
+      if request.from_disk_cache:
+        served_from_cache_bytes += response_sizes[request.url]
+      else:
+        served_from_network_bytes += response_sizes[request.url]
+
     # Make sure the served from blink's cache requests have at least one
     # corresponding request that was not served from the blink's cache.
     for request in sandwich_utils.FilterOutDataAndIncompleteRequests(
         trace.request_track.GetEvents()):
       assert (request.url in urls_hitting_network or
               not request.served_from_cache)
 
     run_metrics = {
         'url': trace.url,
         'repeat_id': repeat_id,
@@ skipping 204 matching lines @@
       run_metrics_list = _ProcessRunOutputDir(
           cache_validation_result, benchmark_setup, RunBenchmark.path)
       with open(ProcessRunOutputDir.path, 'w') as csv_file:
         writer = csv.DictWriter(csv_file, fieldnames=(additional_column_names +
                                     sandwich_metrics.COMMON_CSV_COLUMN_NAMES))
         writer.writeheader()
         for trace_metrics in run_metrics_list:
           writer.writerow(trace_metrics)
 
     self._common_builder.default_final_tasks.append(ProcessRunOutputDir)