Teach mojo_benchmark to produce chart_data for the perf dashboard.

mojo/devtools/common/mojo_benchmark

 #!/usr/bin/env python
 # Copyright 2015 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.
 
 """Runner for Mojo application benchmarks."""
 
 import argparse
 import logging
 import sys
 import time
 import os.path
 import re
 
 from devtoolslib import shell_arguments
 from devtoolslib import shell_config
+from devtoolslib import performance_dashboard
 
 
 _DESCRIPTION = """Runner for Mojo application benchmarks.
 
 |benchmark_list_file| has to be a valid Python program that sets a |benchmarks|
 global variable, containing entries of the following form:
 
   {
     'name': '<name of the benchmark>',
     'app': '<url of the app to benchmark>',
@@ skipping 141 matching lines @@
 
 
 def main():
   parser = argparse.ArgumentParser(
       formatter_class=argparse.RawDescriptionHelpFormatter,
       description=_DESCRIPTION)
   parser.add_argument('benchmark_list_file', type=file,
                       help='a file listing benchmarks to run')
   parser.add_argument('--save-traces', action='store_true',
                       help='save the traces produced by benchmarks to disk')
+  parser.add_argument('--chart-data-output-file', type=argparse.FileType('w'),
+                      help='file to which chart data for the performance '

[viettrungluu, 2015/10/15 22:32:56]

Despite what snobs may think, "file to write chart data for the performance
dashboard to" is both shorter and clearer.

[ppi, 2015/10/15 23:13:14]

Done.

+                      'dashboard is to be written')
 
   # Common shell configuration arguments.
   shell_config.add_shell_arguments(parser)
   script_args = parser.parse_args()
   config = shell_config.get_shell_config(script_args)
 
   try:
     shell, common_shell_args = shell_arguments.get_shell(config, [])
   except shell_arguments.ShellConfigurationException as e:
     print e
     return 1
 
   target_os = 'android' if script_args.android else 'linux'
   benchmark_list_params = {"target_os": target_os}
   exec script_args.benchmark_list_file in benchmark_list_params
 
+  chart_data_recorder = None
+  if script_args.chart_data_output_file:
+    chart_data_recorder = performance_dashboard.ChartDataRecorder()
+
   exit_code = 0
   for benchmark_spec in benchmark_list_params['benchmarks']:
     for variant_spec in _generate_benchmark_variants(benchmark_spec):
       name = variant_spec['name']
       app = variant_spec['app']
       duration = variant_spec['duration']
       shell_args = variant_spec.get('shell-args', []) + common_shell_args
       measurements = variant_spec['measurements']
       benchmark_succeeded, benchmark_error, output = _run_benchmark(
           shell, shell_args, name, app, duration, measurements,
           script_args.verbose, script_args.android,
           script_args.save_traces)
 
       print '[ %s ]' % name
 
       some_measurements_failed = False
       if benchmark_succeeded:
         measurement_results = _parse_measurement_results(output)
         # Iterate over the list of specs, not the dictionary, to detect missing
         # results and preserve the required order.
         for measurement_spec in measurements:
           if measurement_spec in measurement_results:
-            print '%s %s' % (measurement_spec,
-                             measurement_results[measurement_spec])
+            result = measurement_results[measurement_spec]
+            print '%s %s' % (measurement_spec, result)
+
+            if chart_data_recorder:
+              measurement_name = measurement_spec.replace('/', '-')

[viettrungluu, 2015/10/15 22:32:56]

Can't you align the way things are named/specified?

[ppi, 2015/10/15 23:13:14]

Mixing undescores with dashes in the spec seems not palatable.

I think we will eventually want to have some better format to spec the
measurements - maybe in the function notation, ie
"time_until(category:event_name)",
time_between(category1:event1,category2:event2),
percentile_duration(category:event_name, percentile)", but  this would perhaps
better be addressed separately.

Wdyt?

[viettrungluu, 2015/10/21 15:45:27]

OK

+              chart_data_recorder.record_scalar(name, measurement_name, 'ms',

[viettrungluu, 2015/10/15 22:32:56]

Are the results always milliseconds?

(Likewise here: It'd be nice if the measurement_results also contained the
units.)

[ppi, 2015/10/15 23:13:14]

For now, yes. I think we will want to have the benchmark app report results
along with the units and propagate this here. I'd suggest to to this when/if we
grow first measurement type that does not return milliseconds.

[viettrungluu, 2015/10/21 15:45:27]

Sure.

+                                                result)
           else:
             print '%s ?' % measurement_spec
             some_measurements_failed = True
 
       if not benchmark_succeeded or some_measurements_failed:
         if not benchmark_succeeded:
           print 'benchmark failed: ' + benchmark_error
         if some_measurements_failed:
           print 'some measurements failed'
         print 'output: '
         print '-' * 72
         print output
         print '-' * 72
         exit_code = 1
 
+  if script_args.chart_data_output_file:
+    script_args.chart_data_output_file.write(chart_data_recorder.get_json())
+
   return exit_code
 
 if __name__ == '__main__':
   sys.exit(main())