Add performance monitoring back to the AU test.
In CL:544536 the performance monitoring was removed because the device
was restarted before it could retrieve the running test results.
This CL adds the performance monitoring back to the test.
It copies the script to the DUT. During the update the chromite
auto_updater will kick it off and save the results when done.
This CL then pulls the file back from the DUT and records the metrics.
This can be submitted before or after the chromite CL.
BUG=chromium:709710
TEST=autoupdate_EndToEnd passes
Change-Id: I4cb59e30ac68c1e97ca74136f51bac3f76be050f
Reviewed-on: https://chromium-review.googlesource.com/564340
Commit-Ready: David Haddock <dhaddock@chromium.org>
Tested-by: David Haddock <dhaddock@chromium.org>
Reviewed-by: Xixuan Wu <xixuan@chromium.org>
diff --git a/server/site_tests/autoupdate_EndToEndTest/autoupdate_EndToEndTest.py b/server/site_tests/autoupdate_EndToEndTest/autoupdate_EndToEndTest.py
index 2d6299e..6c06637 100755
--- a/server/site_tests/autoupdate_EndToEndTest/autoupdate_EndToEndTest.py
+++ b/server/site_tests/autoupdate_EndToEndTest/autoupdate_EndToEndTest.py
@@ -29,6 +29,9 @@
return ('%s%s\n%s\n%s%s' %
(start, snip[len(start):], text, end, snip[len(end):]))
+UPDATE_ENGINE_PERF_PATH = '/mnt/stateful_partition/unencrypted/preserve'
+UPDATE_ENGINE_PERF_SCRIPT = 'update_engine_performance_monitor.py'
+UPDATE_ENGINE_PERF_RESULTS_FILE = 'perf_data_results.json'
# Update event types.
EVENT_TYPE_DOWNLOAD_COMPLETE = '1'
@@ -449,9 +452,10 @@
raise NotImplementedError
- def stop_update_perf(self):
+ def stop_update_perf(self, resultdir):
"""Stops performance monitoring and returns data (if available).
+ @param resultdir: Directory containing test result files.
@return Dictionary containing performance attributes.
"""
raise NotImplementedError
@@ -712,45 +716,6 @@
client_at.run_test('login_LoginSuccess', tag=tag)
- def _start_perf_mon(self, bindir):
- """Starts monitoring performance and resource usage on a DUT.
-
- Call _stop_perf_mon() with the returned PID to stop monitoring
- and collect the results.
-
- @param bindir: Directoy containing monitoring script.
-
- @return The PID of the newly created DUT monitoring process.
- """
- # We can't assume much about the source image so we copy the
- # performance monitoring script to the DUT directly.
- path = os.path.join(bindir, 'update_engine_performance_monitor.py')
- self._host.send_file(path, '/tmp')
- cmd = 'python /tmp/update_engine_performance_monitor.py --start-bg'
- return int(self._host.run(cmd).stdout)
-
-
- def _stop_perf_mon(self, perf_mon_pid):
- """Stops monitoring performance and resource usage on a DUT.
-
- @param perf_mon_pid: the PID returned from _start_perf_mon().
-
- @return Dictionary containing performance attributes, or None if
- unavailable.
- """
- # Gracefully handle problems with performance monitoring by
- # just returning None.
- try:
- cmd = ('python /tmp/update_engine_performance_monitor.py '
- '--stop-bg=%d') % perf_mon_pid
- perf_json_txt = self._host.run(cmd).stdout
- return json.loads(perf_json_txt)
- except Exception as e:
- logging.warning('Failed to parse output from '
- 'update_engine_performance_monitor.py: %s', e)
- return None
-
-
# Interface overrides.
#
def initialize(self, autotest_devserver, results_dir):
@@ -781,17 +746,27 @@
def start_update_perf(self, bindir):
- if self._perf_mon_pid is None:
- self._perf_mon_pid = self._start_perf_mon(bindir)
+ """Copy performance monitoring script to DUT.
+
+ The updater will kick off the script during the update.
+ """
+ path = os.path.join(bindir, UPDATE_ENGINE_PERF_SCRIPT)
+ self._host.send_file(path, UPDATE_ENGINE_PERF_PATH)
- def stop_update_perf(self):
- perf_data = None
- if self._perf_mon_pid is not None:
- perf_data = self._stop_perf_mon(self._perf_mon_pid)
- self._perf_mon_pid = None
-
- return perf_data
+ def stop_update_perf(self, resultdir):
+ """ Copy the performance metrics back from the DUT."""
+ try:
+ path = os.path.join('/var/log', UPDATE_ENGINE_PERF_RESULTS_FILE)
+ self._host.get_file(path, resultdir)
+ self._host.run('rm %s' % path)
+ script = os.path.join(UPDATE_ENGINE_PERF_PATH,
+ UPDATE_ENGINE_PERF_SCRIPT)
+ self._host.run('rm %s' % script)
+ return os.path.join(resultdir, UPDATE_ENGINE_PERF_RESULTS_FILE)
+ except:
+ logging.debug('Failed to copy performance metrics from DUT.')
+ return None
def trigger_update(self, target_payload_uri):
@@ -849,12 +824,6 @@
e.g. 'localhost:8080/static/my_file.gz'. These are usually
given after staging an artifact using a autotest_devserver
though they can be re-created given enough assumptions.
- *update_url's: Urls refering to the update RPC on a given omaha devserver.
- Since we always use an instantiated omaha devserver to run
- updates, these will always reference an existing instance
- of an omaha devserver that we just created for the purposes
- of updating.
-
"""
version = 1
@@ -918,14 +887,22 @@
pass
- def _report_perf_data(self, perf_data):
+ def _report_perf_data(self, perf_file):
"""Reports performance and resource data.
Currently, performance attributes are expected to include 'rss_peak'
(peak memory usage in bytes).
- @param perf_data: A dictionary containing performance attributes.
+ @param perf_file: A file with performance metrics.
"""
+ logging.debug('Reading perf results from %s.' % perf_file)
+ try:
+ with open(perf_file, 'r') as perf_file_handle:
+ perf_data = json.loads(perf_file_handle.read())
+ except Exception as e:
+ logging.warning('Error while reading the perf data file: %s' % e)
+ return
+
rss_peak = perf_data.get('rss_peak')
if rss_peak:
rss_peak_kib = rss_peak / 1024
@@ -935,8 +912,8 @@
units='KiB',
higher_is_better=False)
else:
- logging.warning('No rss_peak key in JSON returned by '
- 'update_engine_performance_monitor.py')
+ logging.warning('No rss_peak key in JSON returned by %s',
+ UPDATE_ENGINE_PERF_SCRIPT)
def _error_initial_check(self, expected, actual, mismatched_attrs):
@@ -1063,10 +1040,7 @@
source_release = test_conf['source_release']
target_release = test_conf['target_release']
- # TODO(dhaddock): Reuse update_engine_performance_monitor
- # script with chromite autoupdater. Can't use it here anymore because
- # the DUT is restarted a bunch of times during the update and the
- # process is killed before we can get the results back.
+ test_platform.start_update_perf(self.bindir)
try:
# Update the DUT to the target image.
pid = test_platform.trigger_update(test_conf['target_payload_uri'])
@@ -1128,6 +1102,10 @@
logging.fatal('ERROR: Failure occurred during the target update.')
raise
+ perf_file = test_platform.stop_update_perf(self.job.resultdir)
+ if perf_file is not None:
+ self._report_perf_data(perf_file)
+
if test_platform.oobe_triggers_update():
# If DUT automatically checks for update during OOBE,
# checking the post-update CrOS version and slot is sufficient.
