Enable Deep Memory Profiler in perf_endure.py.

chrome/test/functional/perf_endure.py

Index: chrome/test/functional/perf_endure.py
diff --git a/chrome/test/functional/perf_endure.py b/chrome/test/functional/perf_endure.py
old mode 100644
new mode 100755
index df80e73fdedb323f19f132d46f88c635cc189edb..cbacc6c30b0810be24aa31b1ead54252432ad14d
--- a/chrome/test/functional/perf_endure.py
+++ b/chrome/test/functional/perf_endure.py
@@ -14,6 +14,8 @@ This module accepts the following environment variable inputs:
 import logging
 import os
 import re
+import shutil
+import tempfile
 import time
 
 import perf
@@ -23,6 +25,24 @@ import remote_inspector_client
 import selenium.common.exceptions
 from selenium.webdriver.support.ui import WebDriverWait
 
+DMPROF_DIR_PATH = os.path.join(os.path.dirname(__file__),

[Nirnimesh, 2012/04/09 18:46:44]

Unless you want these vars to be accessible from outside this file, prefix _ to
them. Better stil, move them inside the class below.

[Dai Mikurube (NOT FULLTIME), 2012/04/13 11:16:49]

Done.

+                               os.pardir,
+                               os.pardir,
+                               os.pardir,
+                               'tools',
+                               'deep_memory_profiler')
+
+DMPROF_PATH = os.path.join(DMPROF_DIR_PATH,
+                           'dmprof.py')
+
+# TODO(dmikurube): Need to find an actual running chrome.
+CHROME_BIN_PATH = os.path.join(os.path.dirname(__file__),
+                               os.pardir,
+                               os.pardir,
+                               os.pardir,
+                               'out',
+                               'Debug',
+                               'chrome')

[Nirnimesh, 2012/04/09 18:46:44]

What about non-linux?

[Dai Mikurube (NOT FULLTIME), 2012/04/13 11:16:49]

Done.

 
 class ChromeEndureBaseTest(perf.BasePerfTest):
   """Implements common functionality for all Chrome Endure tests.
@@ -122,6 +142,9 @@ class ChromeEndureBaseTest(perf.BasePerfTest):
         logging.info('Chrome interaction #%d. Time remaining in test: %d sec.' %
                      (self._iteration_num, remaining_time))
 
+      if self._iteration_num % 20 == 5:

[Dai Mikurube (NOT FULLTIME), 2012/04/09 09:10:40]

Temporary trigger.

[Nirnimesh, 2012/04/09 18:46:44]

Add comments / TODO notes.

[dennis_jeffrey, 2012/04/09 19:12:38]

We'll likely want to be able to turn profiling on/off and specify the interval
in which we dump profile data (or analyze the dumped data)

[Dai Mikurube (NOT FULLTIME), 2012/04/11 08:44:02]

Ah, exactly.  Such an option is required.  How can I give an option to Endure
from external?

[dennis_jeffrey, 2012/04/12 00:51:52]

Right now it's done through environment variables.  See lines 60-63 above.

[Dai Mikurube (NOT FULLTIME), 2012/04/13 11:16:49]

Using an environment variable DEEP_MEMORY_PROFILE_INTERVAL.  Thanks.

+        self.HeapProfilerDump('In ChromeEndureControlTest')
+
       do_scenario()
 
     self._GetPerformanceStats(
@@ -178,6 +201,8 @@ class ChromeEndureBaseTest(perf.BasePerfTest):
     return {
       'browser_private_mem': browser_proc_info['working_set_mem']['priv'],
       'tab_private_mem': tab_proc_info['working_set_mem']['priv'],
+      'browser_pid': browser_proc_info['pid'],

[dennis_jeffrey, 2012/04/09 19:12:38]

is this value used anywhere?

[Dai Mikurube (NOT FULLTIME), 2012/04/13 11:16:49]

No for now.  Removed it.

+      'tab_pid': tab_proc_info['pid'],
     }
 
   def _GetPerformanceStats(self, webapp_name, test_description,
@@ -223,6 +248,35 @@ class ChromeEndureBaseTest(perf.BasePerfTest):
     logging.info('  V8 memory used: %f KB' % v8_mem_used)
     self._v8_mem_used_results.append((elapsed_time, v8_mem_used))
 
+    # TODO(dmikurube): Add --single in dmprof.py?

[dennis_jeffrey, 2012/04/09 19:12:38]

So in general, a test will periodically call HeapProfilerDump(), and also
periodically, we'll analyze the dumped data.  What are the requirements for how
often each of these things should occur?  What are the tradeoffs involved in
dumping more frequently, or analyzing the dumped data more frequently?

[Dai Mikurube (NOT FULLTIME), 2012/04/11 08:44:02]

I'm in tries-and-errors about that.  Dumping may take a little long
stop-the-world (2-3 sec), and dumped data are large (3-4 MB or more for one
dump).

Frequent analysis has less negative impact.  But, of course, more frequent
analysis than dumping is just waste.  (Yes, the current version does it.)

+    first_dump = ''
+    print '^%s/endure.%05d.\d+.heap$' % (

[Nirnimesh, 2012/04/09 18:46:44]

Use the right logging call instead of print.

[Dai Mikurube (NOT FULLTIME), 2012/04/16 09:43:20]

Done.

+          self._deep_tempdir, proc_info['tab_pid'])
+    for filename in sorted(os.listdir(self._deep_tempdir)):
+      if re.match('^%s/endure.%05d.\d+.heap$' % (

[Nirnimesh, 2012/04/09 18:46:44]

use os.path.join() instead of hardcoding /

[Dai Mikurube (NOT FULLTIME), 2012/04/16 09:43:20]

Done.

+          self._deep_tempdir, proc_info['tab_pid']), filename):
+        first_dump = filename
+        break
+
+    # Test printing.
+    if first_dump:

[dennis_jeffrey, 2012/04/09 19:12:38]

What does it mean for a file to be the "first dump", and why do we only need to
use the first dump file?

[Dai Mikurube (NOT FULLTIME), 2012/04/11 08:44:02]

This is just a test printing, and will be removed eventually.

The current deep memory profiler receives the first dump file, and finds next
dump files by itself.  Analyzing the same data many times is wasteful, but it's
the easiest way to store the result in one file.  (Re-analyzing doesn't cost so
much.)

I'm trying analyzing a single file, and storing its result into one file in
dmprof.py.

+      print 'First Dump: %s' % first_dump

[Nirnimesh, 2012/04/09 18:46:44]

logging.info()?

[Dai Mikurube (NOT FULLTIME), 2012/04/16 09:43:20]

Done.

+    else:
+      print 'No Dump!'
+    """
+      with open('%s/endure.%05d.json' % (self._deep_tempdir,

[Nirnimesh, 2012/04/09 18:46:44]

use os.path.join

[Dai Mikurube (NOT FULLTIME), 2012/04/13 11:16:49]

Done.

+                                         proc_info['tab_pid']), 'w+') as json_f:
+        p = subprocess.Popen(

[Nirnimesh, 2012/04/09 18:46:44]

use better varname instead of |p|

[Dai Mikurube (NOT FULLTIME), 2012/04/11 08:44:02]

Good catch.  Thanks.

+            '%s --json %s %s %s' % (DMPROF_PATH,
+                                    CHROME_BIN_PATH,
+                                    os.path.join(DMPROF_DIR_PATH, 'dmpolicy'),
+                                    @@@SNAPSHOT@@@),
+            shell=True, stdout=json_f)
+        p.wait()

[Nirnimesh, 2012/04/09 18:46:44]

you won't need this if you use subprocess.call() in the previous line.

[Dai Mikurube (NOT FULLTIME), 2012/04/16 09:43:20]

Finally, this script is required to run on background.  (dmprof may take long
time.)

The latest version supports background execution with Popen.  Please look at
that.

+        """
+
+    # 2) Parse .json (including only the latest snapshot) file here.
+
     # Output the results seen so far, to be graphed.
     self._OutputPerfGraphValue(
         'TotalDOMNodeCount', self._dom_node_count_results, 'nodes',
@@ -244,6 +298,7 @@ class ChromeEndureBaseTest(perf.BasePerfTest):
         'V8MemoryUsed', self._v8_mem_used_results, 'KB',
         graph_name='%s%s-V8MemUsed' % (webapp_name, test_description),
         units_x='seconds')
+    # 3) Do self._OutputPerfGraphValue for multiple lines, here.
 
   def _GetElement(self, find_by, value):
     """Gets a WebDriver element object from the webpage DOM.
@@ -310,6 +365,19 @@ class ChromeEndureControlTest(ChromeEndureBaseTest):
   _webapp_name = 'Control'
   _tab_title_substring = 'Chrome Endure Control Test'
 
+  def setUp(self):

[dennis_jeffrey, 2012/04/09 19:12:38]

This only enables profiling in the two control tests, since it's inside of class
ChromeEndureControlTest.  Did you mean to enable profiling in all Chrome Endure
tests?

[Dai Mikurube (NOT FULLTIME), 2012/04/11 08:44:02]

Finally, yes.  Is there an efficient way to enable it for all Endure tests?

[dennis_jeffrey, 2012/04/12 00:51:52]

Yes - if you enable profiling above in class ChromeEndureBaseTest (see setUp
function in line 57), then it will be enabled for all Chrome Endure tests, since
all Chrome Endure tests execute that setUp function.

[Dai Mikurube (NOT FULLTIME), 2012/04/13 11:16:49]

I see.  Thank you.
I did it in the uploaded code.

+    # TODO(dmikurube): Have to specify "--no-sandbox" for Chrome.

[Dai Mikurube (NOT FULLTIME), 2012/04/09 09:20:27]

Note: ExtraChromeFlags would work.

+    self._deep_tempdir = tempfile.mkdtemp()
+    os.environ['HEAPPROFILE'] = '%s/endure' % self._deep_tempdir

[Nirnimesh, 2012/04/09 18:46:44]

os.path.join

[Dai Mikurube (NOT FULLTIME), 2012/04/13 11:16:49]

Done.

+    os.environ['DEEP_HEAP_PROFILE'] = 'True'
+    ChromeEndureBaseTest.setUp(self)
+
+  def tearDown(self):
+    del os.environ['DEEP_HEAP_PROFILE']
+    del os.environ['HEAPPROFILE']
+    ChromeEndureBaseTest.tearDown(self)
+    # shutil.rmtree(self._deep_tempdir)
+
   def testControlAttachDetachDOMTree(self):
     """Continually attach and detach a DOM tree from a basic document."""
     test_description = 'AttachDetachDOMTree'