[XB1] Refactor test retry loop to rerun net_args (#1647)

b/299672207

Change-Id: Ic90a8805fa9bc6ec7a99ffe55c318f44b82f85cb

starboard/xb1/tools/xb1_launcher.py

@@ -95,6 +95,14 @@
 _XB1_NET_LOG_PORT = 49353
 _XB1_NET_ARG_PORT = 49355
 
+# Number of times a test will try or retry.
+_TEST_MAX_TRIES = 4
+# Seconds to wait between retries (scales with backoff factor).
+_TEST_RETRY_WAIT = 8
+# Amount to multiply retry time with each failed attempt (i.e. 2 doubles the
+# amount of time to wait between retries).
+_TEST_RETRY_BACKOFF_FACTOR = 2
+
 _PROCESS_TIMEOUT = 60 * 5.0
 _PROCESS_KILL_TIMEOUT_SECONDS = 5.0
 
@@ -448,6 +456,31 @@ def CheckPackageIsDeployed(self):
       return False
     return package_list[package_index:].split('\n')[0].strip()
 
+  def RunTest(self, appx_name: str):
+    self.net_args_thread = None
+    attempt_num = 0
+    retry_wait_s = _TEST_RETRY_WAIT
+    while attempt_num < _TEST_MAX_TRIES:
+      if not self.net_args_thread or not self.net_args_thread.is_alive():
+        # This thread must start before the app executes or else it is possible
+        # the app will hang at _network_api.ExecuteBinary()
+        self.net_args_thread = net_args.NetArgsThread(self.device_id,
+                                                      _XB1_NET_ARG_PORT,
+                                                      self._target_args)
+      if self._network_api.ExecuteBinary(_DEFAULT_PACKAGE_NAME, appx_name):
+        break
+
+      if not self.net_args_thread.ArgsSent():
+        self._LogLn(
+            'Net Args were not sent to the test! This will likely cause '
+            'the test to fail!')
+      attempt_num += 1
+      self._LogLn(f'Retry attempt {attempt_num}.')
+      time.sleep(retry_wait_s)
+      retry_wait_s *= _TEST_RETRY_BACKOFF_FACTOR
+      if hasattr(self, 'net_args_thread'):
+        self.net_args_thread.join()
+
   def Run(self):
     # Only upload and install Appx on the first run.
     if FirstRun():
@@ -474,20 +507,14 @@ def Run(self):
 
     try:
       self.Kill()  # Kill existing running app.
-
-      # These threads must start before the app executes or else it is possible
-      # the app will hang at _network_api.ExecuteBinary()
-      self.net_args_thread = net_args.NetArgsThread(self.device_id,
-                                                    _XB1_NET_ARG_PORT,
-                                                    self._target_args)
       # While binary is running, extract the net log and stream it to
       # the output.
       self.net_log_thread = net_log.NetLogThread(self.device_id,
                                                  _XB1_NET_LOG_PORT)
 
       appx_name = ToAppxFriendlyName(self.target_name)
 
-      self._network_api.ExecuteBinary(_DEFAULT_PACKAGE_NAME, appx_name)
+      self.RunTest(appx_name)
 
       while self._network_api.IsBinaryRunning(self.target_name):
         self._Log(self.net_log_thread.GetLog())

starboard/xb1/tools/xb1_network_api.py

@@ -421,7 +421,8 @@ def FetchPackageFile(self,
         return None
     return None
 
-  def ExecuteBinary(self, partial_package_name, app_alias_name):
+  def ExecuteBinary(self, partial_package_name: str,
+                    app_alias_name: str) -> bool:
     default_relative_name = self._GetDefaultRelativeId(partial_package_name)
     if not default_relative_name or not '!' in default_relative_name:
       raise IOError('Could not resolve package name "' + partial_package_name +
@@ -432,33 +433,25 @@ def ExecuteBinary(self, partial_package_name, app_alias_name):
     appid_64 = base64.b64encode(package_relative_id.encode('UTF-8'))
     package_64 = base64.b64encode(default_relative_name.encode('UTF-8'))
 
-    retry_count = 4
-    # Time to wait between tries.
-    retry_wait_s = 8
     try:
-      while retry_count > 0:
-        self.LogLn('Executing: ' + package_relative_id)
-        response = self._DoJsonRequest(
-            'POST',
-            _TASKMANAGER_ENDPOINT,
-            params={
-                'appid': appid_64,
-                'package': package_64
-            },
-            raise_on_failure=False)
-        if not response or response == requests.codes.OK:
-          self.LogLn('Execution successful')
-          break
-        self.LogLn('Execution not successful: ' + str(response))
-        self.LogLn('Retrying with ' + str(retry_count) + ' attempts remaining.')
-        time.sleep(retry_wait_s)
-        retry_count -= 1
-        # Double the wait time until the next attempt.
-        retry_wait_s *= 2
+      self.LogLn('Executing: ' + package_relative_id)
+      response = self._DoJsonRequest(
+          'POST',
+          _TASKMANAGER_ENDPOINT,
+          params={
+              'appid': appid_64,
+              'package': package_64
+          },
+          raise_on_failure=False)
+      if not response or response == requests.codes.OK:
+        self.LogLn('Execution successful')
+        return True
+      self.LogLn('Execution not successful: ' + str(response))
     except Exception as err:
       err_msg = '\n  Failed to run:\n  ' + package_relative_id + \
                 '\n  because of:\n' + str(err)
       raise IOError(err_msg) from err
+    return False
 
   # Given a package name, return all files + directories.
   # Throws IOError if the app is locked.