[Android] Add ability to generate test trace json for perf tests runs.

build/android/pylib/local/device/local_device_perf_test_run.py

Index: build/android/pylib/local/device/local_device_perf_test_run.py
diff --git a/build/android/pylib/local/device/local_device_perf_test_run.py b/build/android/pylib/local/device/local_device_perf_test_run.py
index efb00017c2eb61e1619a062266cf28d2ecf9e573..e8e612c2f7914b3a7898c83c115b50a07425e476 100644
--- a/build/android/pylib/local/device/local_device_perf_test_run.py
+++ b/build/android/pylib/local/device/local_device_perf_test_run.py
@@ -29,6 +29,7 @@ from pylib.base import base_test_result
 from pylib.constants import host_paths
 from pylib.local.device import local_device_environment
 from pylib.local.device import local_device_test_run
+from py_trace_event import trace_event

[rnephew (Reviews Here), 2016/12/15 19:24:58]

I keep getting:
** Presubmit Warnings **
Pylint (128 files using ['--disable=cyclic-import'] on 12 cores) (4.40s) failed
************* Module pylib.local.device.local_device_perf_test_run
F: 32, 0: Unable to import 'py_trace_event' (import-error)

Even though it works locally. Do I need to add it to another __init__ file
somewhere so that pylint is appeased? 

[mikecase (-- gone --), 2016/12/15 19:46:47]

Here I think.

https://cs.chromium.org/chromium/src/build/android/PRESUBMIT.py?q=PRESUBMIT+f...

[rnephew (Reviews Here), 2016/12/15 19:55:13]

Thanks.

 
 
 class HeartBeat(object):
@@ -96,8 +97,12 @@ class TestShard(object):
 
     try:
       start_time = time.time()
+      if self._test_instance.trace_output:
+        trace_event.trace_begin(test)
       exit_code, output = cmd_helper.GetCmdStatusAndOutputWithTimeout(
           cmd, timeout, cwd=cwd, shell=True)
+      if self._test_instance.trace_output:
+        trace_event.trace_end(test)
       end_time = time.time()
       json_output = self._test_instance.ReadChartjsonOutput(self._output_dir)
       if exit_code == 0:
@@ -410,6 +415,10 @@ class LocalDevicePerfTestRun(local_device_test_run.LocalDeviceTestRun):
   #override
   def RunTests(self):
     # Affinitize the tests.
+    if self._test_instance.trace_output:
+      assert not trace_event.trace_is_enabled(), 'Tracing already running.'
+      trace_event.trace_enable(self._test_instance.trace_output)
+      assert trace_event.trace_is_enabled(), 'Tracing didn\'t enable properly.'

[mikecase (-- gone --), 2016/12/15 19:56:02]

nit: Maybe just didn't -> "did not enable" or just "not enabled" to remove need
to escape single-quote.


Also nit:

Are the...

assert trace_event.trace_is_enabled(), 'Tracing didn\'t enable properly.'

and final...

assert not trace_event.trace_is_enabled(), 'Tracing not disabled.'

really neccesarry. I get the other two, something else could be enabling tracing
which could screw us up. And somethinhg could disable tracing half-way into run.
Other two asserts dont seem neccessary.

[rnephew (Reviews Here), 2016/12/15 20:02:10]

Thats how I do it in telemetry when using trace_events. I dont recall why
exactly, but its probably safe to get rid of them.

     self._SplitTestsByAffinity()
     if not self._test_buckets and not self._no_device_tests:
       raise local_device_test_run.NoTestsError()
@@ -447,6 +456,10 @@ class LocalDevicePerfTestRun(local_device_test_run.LocalDeviceTestRun):
 
     host_test_results, device_test_results = reraiser_thread.RunAsync(
         [run_no_devices_tests, run_devices_tests])
+    if self._test_instance.trace_output:
+      assert trace_event.trace_is_enabled(), 'Tracing not running.'
+      trace_event.trace_disable()
+      assert not trace_event.trace_is_enabled(), 'Tracing not disabled.'
     return host_test_results + device_test_results
 
   # override