Dump data for stacked graphs in Chrome Endure.

chrome/test/functional/perf.py

Index: chrome/test/functional/perf.py
diff --git a/chrome/test/functional/perf.py b/chrome/test/functional/perf.py
index a9b5ca3c6566790fbf9e0e542776f9a13e7320ed..5f52ce2a13c991be8713ea683639a1211b7c62e0 100755
--- a/chrome/test/functional/perf.py
+++ b/chrome/test/functional/perf.py
@@ -256,7 +256,7 @@ class BasePerfTest(pyauto.PyUITest):
     return timer.timeit(number=1) * 1000  # Convert seconds to milliseconds.
 
   def _OutputPerfForStandaloneGraphing(self, graph_name, description, value,
-                                       units, units_x):
+                                       units, units_x, is_stacked):
     """Outputs perf measurement data to a local folder to be graphed.
 
     This function only applies to Chrome desktop, and assumes that environment
@@ -280,6 +280,8 @@ class BasePerfTest(pyauto.PyUITest):
           with the performance measurements, such as 'iteration' if the x values
           are iteration numbers.  If this argument is specified, then the
           |value| argument must be a list of (x, y) tuples.
+      is_stacked: True to draw a "stacked" graph.  First-come values are
+          stacked at bottom by default.
     """
     revision_num_file = os.path.join(self._local_perf_dir, 'last_revision.dat')
     if os.path.exists(revision_num_file):
@@ -346,6 +348,13 @@ class BasePerfTest(pyauto.PyUITest):
         'traces': { description: [str(value), str(0.0)] }
       }
 
+    if is_stacked:
+      new_line['stacked'] = True
+      if 'default_stacking_order' not in new_line:

[dennis_jeffrey, 2012/09/05 17:34:12]

should we change the name to something shorter like "stack_order"?  We'll just
assume that's the default and can be overridden elsewhere if needed.

[Dai Mikurube (NOT FULLTIME), 2012/09/06 05:30:10]

Renamed:
'stacked' => 'stack'
'default_stacking_order' => 'stack_order'

+        new_line['default_stacking_order'] = []
+      if description not in new_line['default_stacking_order']:
+        new_line['default_stacking_order'].append(description)
+
     if seen_key in self._seen_graph_lines:
       # Update results for the most recent revision.
       existing_lines[0] = new_line
@@ -363,7 +372,7 @@ class BasePerfTest(pyauto.PyUITest):
       f.write(str(revision))
 
   def _OutputPerfGraphValue(self, description, value, units,

[dennis_jeffrey, 2012/09/05 17:34:12]

Note: there are actually 2 ways in which perf results are outputted in this
file:

1) By calling pyauto_utils.PrintPerfResult.  This prints the results to stdout
using a specially-formatted output line.  These output lines are what are parsed
in the live Chrome Endure bots running on the chromium.pyauto waterfall, and
those results get written into text files that are displayed by the live chrome
endure graphs.

2) By calling self._OutputPerfForStandaloneGraphing.  These results are directly
written to text data files to be displayed on local perf graphs only (not on the
live Chrome Endure graphs).

Currently, the changes in this file handle (2) but not (1).  We'll need to also
figure out how best to modify the format of the specially-formatted output lines
to output information about graphs that should be stacked.  Note also that the
code that actually parses these stdout lines and creates the data files for the
live Chrome Endure graphs is also not checked in yet, but that will have to be
modified along with the changes here to get the stacked graphs to work on the
live graphs.  So, I will work on checking in that other code that parses the
stdout of the live Chrome Endure bots to display on the live graphs.  It's just
a python script that runs every 2 hours on my local machine to update the data
that's displayed on the live chrome endure graphs.

[Dai Mikurube (NOT FULLTIME), 2012/09/06 05:30:10]

Thanks for your effort to commit the "RESULT" parsing code!  I was wondering
about that.

For now, I added a TODO in _OutputPerfGraphValue.

[cmp_google, 2012/09/14 04:32:36]

Dennis and Dai, please let me know if/when there's something to review re: the
change to the master-parsing code to handle the endure and also the stacked
graphs.  I'm interested to see how that code would change to support this (and
of course to maintain backwards compatibility with the previous format). 
Thanks.

[Dai Mikurube (NOT FULLTIME), 2012/09/14 10:19:07]

Ah, ok, I'll Cc: you for such changes!

[dennis_jeffrey, 2012/09/14 16:33:31]

Yes, actually I was planning to make you our main point of contact (and code
reviewer) once we start modifying the master parsing code and the original
chrome perf graphing code.

-                            graph_name, units_x=None):
+                            graph_name, units_x=None, is_stacked=False):
     """Outputs a performance value to have it graphed on the performance bots.
 
     The output format differs, depending on whether the current platform is
@@ -394,6 +403,8 @@ class BasePerfTest(pyauto.PyUITest):
           with the performance measurements, such as 'iteration' if the x values
           are iteration numbers.  If this argument is specified, then the
           |value| argument must be a list of (x, y) tuples.
+      is_stacked: True to draw a "stacked" graph.  First-come values are
+          stacked at bottom by default.
     """
     if (isinstance(value, list) and value[0] is not None and
         isinstance(value[0], tuple)):
@@ -433,7 +444,7 @@ class BasePerfTest(pyauto.PyUITest):
 
       if self._local_perf_dir:
         self._OutputPerfForStandaloneGraphing(
-            graph_name, description, value, units, units_x)
+            graph_name, description, value, units, units_x, is_stacked)
 
   def _OutputEventForStandaloneGraphing(self, description, event_list):
     """Outputs event information to a local folder to be graphed.