rebaseline_server: download actual-results.json files from GCS instead of SVN

gm/rebaseline_server/server.py

 #!/usr/bin/python
 
 """
 Copyright 2013 Google Inc.
 
 Use of this source code is governed by a BSD-style license that can be
 found in the LICENSE file.
 
 HTTP server for our HTML rebaseline viewer.
 """
@@ skipping 10 matching lines @@
 import socket
 import subprocess
 import sys
 import thread
 import threading
 import time
 import urlparse
 
 # Imports from within Skia
 #
-# We need to add the 'tools' directory for svn.py, and the 'gm' directory for
-# gm_json.py .
-# that directory.
-# Make sure that the 'tools' dir is in the PYTHONPATH, but add it at the *end*
+# We need to add the 'gm' directory for gm_json.py .
+# Make sure this dir is in the PYTHONPATH, but add it at the *end*
 # so any dirs that are already in the PYTHONPATH will be preferred.
 PARENT_DIRECTORY = os.path.dirname(os.path.realpath(__file__))
 GM_DIRECTORY = os.path.dirname(PARENT_DIRECTORY)
 TRUNK_DIRECTORY = os.path.dirname(GM_DIRECTORY)
-TOOLS_DIRECTORY = os.path.join(TRUNK_DIRECTORY, 'tools')
-if TOOLS_DIRECTORY not in sys.path:
-  sys.path.append(TOOLS_DIRECTORY)
-import svn
 if GM_DIRECTORY not in sys.path:
   sys.path.append(GM_DIRECTORY)
 import gm_json
 
 # Imports from local dir
 #
 # Note: we import results under a different name, to avoid confusion with the
 # Server.results() property. See discussion at
 # https://codereview.chromium.org/195943004/diff/1/gm/rebaseline_server/server.py#newcode44
 import compare_configs
 import compare_to_expectations
+import download_actuals
 import imagepairset
 import results as results_mod
 
 PATHSPLIT_RE = re.compile('/([^/]+)/(.+)')
 
 # A simple dictionary of file name extensions to MIME types. The empty string
 # entry is used as the default when no extension was given or if the extension
 # has no entry in this dictionary.
 MIME_TYPE_MAP = {'': 'application/octet-stream',
                  'html': 'text/html',
                  'css': 'text/css',
                  'png': 'image/png',
                  'js': 'application/javascript',
                  'json': 'application/json'
                  }
 
 # Keys that server.py uses to create the toplevel content header.
 # NOTE: Keep these in sync with static/constants.js
 KEY__EDITS__MODIFICATIONS = 'modifications'
 KEY__EDITS__OLD_RESULTS_HASH = 'oldResultsHash'
 KEY__EDITS__OLD_RESULTS_TYPE = 'oldResultsType'
 
 DEFAULT_ACTUALS_DIR = results_mod.DEFAULT_ACTUALS_DIR
-DEFAULT_ACTUALS_REPO_REVISION = 'HEAD'
-DEFAULT_ACTUALS_REPO_URL = 'http://skia-autogen.googlecode.com/svn/gm-actual'
+DEFAULT_GM_SUMMARIES_BUCKET = download_actuals.GM_SUMMARIES_BUCKET
+DEFAULT_JSON_FILENAME = download_actuals.DEFAULT_JSON_FILENAME
 DEFAULT_PORT = 8888
 
 # Directory, relative to PARENT_DIRECTORY, within which the server will serve
 # out live results (not static files).
 RESULTS_SUBDIR = 'results'
 # Directory, relative to PARENT_DIRECTORY, within which the server will serve
 # out static files.
 STATIC_CONTENTS_SUBDIR = 'static'
 # All of the GENERATED_*_SUBDIRS are relative to STATIC_CONTENTS_SUBDIR
 GENERATED_HTML_SUBDIR = 'generated-html'
@@ skipping 43 matching lines @@
   """Returns routable IP address of this host (the IP address of its network
      interface that would be used for most traffic, not its localhost
      interface).  See http://stackoverflow.com/a/166589 """
   sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
   sock.connect(('8.8.8.8', 80))
   host = sock.getsockname()[0]
   sock.close()
   return host
 
 
-def _create_svn_checkout(dir_path, repo_url):
-  """Creates local checkout of an SVN repository at the specified directory
-  path, returning an svn.Svn object referring to the local checkout.
-
-  Args:
-    dir_path: path to the local checkout; if this directory does not yet exist,
-              it will be created and the repo will be checked out into it
-    repo_url: URL of SVN repo to check out into dir_path (unless the local
-              checkout already exists)
-  Returns: an svn.Svn object referring to the local checkout.
-  """
-  local_checkout = svn.Svn(dir_path)
-  if not os.path.isdir(dir_path):
-    os.makedirs(dir_path)
-    local_checkout.Checkout(repo_url, '.')
-  return local_checkout
-
-
 def _create_index(file_path, config_pairs):
   """Creates an index file linking to all results available from this server.
 
   Prior to https://codereview.chromium.org/215503002 , we had a static
   index.html within our repo.  But now that the results may or may not include
   config comparisons, index.html needs to be generated differently depending
   on which results are included.
 
   TODO(epoger): Instead of including raw HTML within the Python code,
   consider restoring the index.html file as a template and using django (or
@@ skipping 36 matching lines @@
         file_handle.write('</li>')
       file_handle.write('</ul>')
     file_handle.write('</ul></body></html>')
 
 
 class Server(object):
   """ HTTP server for our HTML rebaseline viewer. """
 
   def __init__(self,
                actuals_dir=DEFAULT_ACTUALS_DIR,
-               actuals_repo_revision=DEFAULT_ACTUALS_REPO_REVISION,
-               actuals_repo_url=DEFAULT_ACTUALS_REPO_URL,
+               json_filename=DEFAULT_JSON_FILENAME,
+               gm_summaries_bucket=DEFAULT_GM_SUMMARIES_BUCKET,
                port=DEFAULT_PORT, export=False, editable=True,
                reload_seconds=0, config_pairs=None, builder_regex_list=None):
     """
     Args:
       actuals_dir: directory under which we will check out the latest actual
           GM results
-      actuals_repo_revision: revision of actual-results.json files to process
-      actuals_repo_url: SVN repo to download actual-results.json files from;
-          if None or '', don't fetch new actual-results files at all,
-          just compare to whatever files are already in actuals_dir
+      json_filename: basename of the JSON summary file to load for each builder
+      gm_summaries_bucket: Google Storage bucket to download json_filename
+          files from; if None or '', don't fetch new actual-results files
+          at all, just compare to whatever files are already in actuals_dir
       port: which TCP port to listen on for HTTP requests
       export: whether to allow HTTP clients on other hosts to access this server
       editable: whether HTTP clients are allowed to submit new baselines
       reload_seconds: polling interval with which to check for new results;
           if 0, don't check for new results at all
       config_pairs: List of (string, string) tuples; for each tuple, compare
           actual results of these two configs.  If None or empty,
           don't compare configs at all.
       builder_regex_list: List of regular expressions specifying which builders
           we will process. If None, process all builders.
     """
     self._actuals_dir = actuals_dir
-    self._actuals_repo_revision = actuals_repo_revision
-    self._actuals_repo_url = actuals_repo_url
+    self._json_filename = json_filename
+    self._gm_summaries_bucket = gm_summaries_bucket
     self._port = port
     self._export = export
     self._editable = editable
     self._reload_seconds = reload_seconds
     self._config_pairs = config_pairs or []
     self._builder_regex_list = builder_regex_list
     _create_index(
         file_path=os.path.join(
             PARENT_DIRECTORY, STATIC_CONTENTS_SUBDIR, GENERATED_HTML_SUBDIR,
             "index.html"),
         config_pairs=config_pairs)
-    # TODO(epoger): Create shareable functions within download_actuals.py that
-    # we can use both there and here to download the actual image results.
-    if actuals_repo_url:
-      self._actuals_repo = _create_svn_checkout(
-          dir_path=actuals_dir, repo_url=actuals_repo_url)
 
     # Reentrant lock that must be held whenever updating EITHER of:
     # 1. self._results
     # 2. the expected or actual results on local disk
     self.results_rlock = threading.RLock()
     # self._results will be filled in by calls to update_results()
     self._results = None
 
   @property
   def results(self):
@@ skipping 27 matching lines @@
     the same time.
 
     Args:
       invalidate: if True, invalidate self._results immediately upon entry;
                   otherwise, we will let readers see those results until we
                   replace them
     """
     with self.results_rlock:
       if invalidate:
         self._results = None
-      if self._actuals_repo_url:
+      if self._gm_summaries_bucket:
         logging.info(
-            'Updating actual GM results in %s to revision %s from repo %s ...'
-            % (
-                self._actuals_dir, self._actuals_repo_revision,
-                self._actuals_repo_url))
-        self._actuals_repo.Update(
-            path='.', revision=self._actuals_repo_revision)
+            'Updating GM result summaries in %s from gm_summaries_bucket %s ...'
+            % (self._actuals_dir, self._gm_summaries_bucket))
+
+        # Clean out actuals_dir first, in case some builders have gone away
+        # since we last ran.
+        if os.path.isdir(self._actuals_dir):
+          shutil.rmtree(self._actuals_dir)
+
+        # Get the list of builders we care about.
+        all_builders = download_actuals.get_builders_list(
+            summaries_bucket=self._gm_summaries_bucket)
+        if self._builder_regex_list:
+          matching_builders = []
+          for builder in all_builders:
+            for regex in self._builder_regex_list:
+              if re.match(regex, builder):
+                matching_builders.append(builder)
+                break  # go on to the next builder, no need to try more regexes
+        else:
+          matching_builders = all_builders
+
+        # Download the JSON file for each builder we care about.
+        #
+        # TODO(epoger): When this is a large number of builders, we would be
+        # better off downloading them in parallel!
+        for builder in matching_builders:
+          download_actuals.gcs_download_file(
+              source_bucket=self._gm_summaries_bucket,
+              source_path=posixpath.join(builder, self._json_filename),
+              dest_path=os.path.join(self._actuals_dir, builder,
+                                     self._json_filename),
+              create_subdirs_if_needed=True)
 
       # We only update the expectations dir if the server was run with a
       # nonzero --reload argument; otherwise, we expect the user to maintain
       # her own expectations as she sees fit.
       #
-      # Because the Skia repo is moving from SVN to git, and git does not
+      # Because the Skia repo is hosted using git, and git does not
       # support updating a single directory tree, we have to update the entire
       # repo checkout.
       #
       # Because Skia uses depot_tools, we have to update using "gclient sync"
-      # instead of raw git (or SVN) update.  Happily, this will work whether
-      # the checkout was created using git or SVN.
+      # instead of raw git commands.

[borenet, 2014/06/03 20:27:47]

This is not really true, and the below will not have the intended effect for
unmanaged checkouts.  Would we be better off doing this?

$ git reset --hard HEAD
$ git checkout master -f
$ git fetch
$ git reset --hard origin/master
$ gclient sync

That should ensure that we get exactly where we want to go.  It's destructive
though...

[epoger, 2014/06/03 20:39:04]

Yikes.

[Insert "blah blah what is our officially supported method of checking out Skia
blah blah" here.]

I dunno if that's a good change or not, but I'd definitely like to leave it for
a future CL either way, so we can kill skia-autogen without getting bogged down
in all that.

       if self._reload_seconds:
         logging.info(
             'Updating expected GM results in %s by syncing Skia repo ...' %
             compare_to_expectations.DEFAULT_EXPECTATIONS_DIR)
         _run_command(['gclient', 'sync'], TRUNK_DIRECTORY)
 
       self._results = compare_to_expectations.ExpectationComparisons(
           actuals_root=self._actuals_dir,
           generated_images_root=os.path.join(
               PARENT_DIRECTORY, STATIC_CONTENTS_SUBDIR,
@@ skipping 281 matching lines @@
 def main():
   logging.basicConfig(format='%(asctime)s %(levelname)s %(message)s',
                       datefmt='%m/%d/%Y %H:%M:%S',
                       level=logging.INFO)
   parser = argparse.ArgumentParser()
   parser.add_argument('--actuals-dir',
                     help=('Directory into which we will check out the latest '
                           'actual GM results. If this directory does not '
                           'exist, it will be created. Defaults to %(default)s'),
                     default=DEFAULT_ACTUALS_DIR)
-  parser.add_argument('--actuals-repo',
-                    help=('URL of SVN repo to download actual-results.json '
-                          'files from. Defaults to %(default)s ; if set to '
-                          'empty string, just compare to actual-results '
-                          'already found in ACTUALS_DIR.'),
-                    default=DEFAULT_ACTUALS_REPO_URL)
-  parser.add_argument('--actuals-revision',
-                    help=('revision of actual-results.json files to process. '
-                          'Defaults to %(default)s .  Beware of setting this '
-                          'argument in conjunction with --editable; you '
-                          'probably only want to edit results at HEAD.'),
-                    default=DEFAULT_ACTUALS_REPO_REVISION)
+  # TODO(epoger): Before https://codereview.chromium.org/310093003 ,
+  # when this tool downloaded the JSON summaries from skia-autogen,
+  # it had an --actuals-revision the caller could specify to download
+  # actual results as of a specific point in time.  We should add similar
+  # functionality when retrieving the summaries from Google Storage.
   parser.add_argument('--builders', metavar='BUILDER_REGEX', nargs='+',
                       help=('Only process builders matching these regular '
                             'expressions.  If unspecified, process all '
                             'builders.'))
   parser.add_argument('--compare-configs', action='store_true',
                       help=('In addition to generating differences between '
                             'expectations and actuals, also generate '
                             'differences between these config pairs: '
                             + str(CONFIG_PAIRS_TO_COMPARE)))
   parser.add_argument('--editable', action='store_true',
                       help=('Allow HTTP clients to submit new baselines.'))
   parser.add_argument('--export', action='store_true',
                       help=('Instead of only allowing access from HTTP clients '
                             'on localhost, allow HTTP clients on other hosts '
                             'to access this server.  WARNING: doing so will '
                             'allow users on other hosts to modify your '
                             'GM expectations, if combined with --editable.'))
+  parser.add_argument('--gm-summaries-bucket',
+                    help=('Google Cloud Storage bucket to download '
+                          'JSON_FILENAME files from. '
+                          'Defaults to %(default)s ; if set to '
+                          'empty string, just compare to actual-results '
+                          'already found in ACTUALS_DIR.'),
+                    default=DEFAULT_GM_SUMMARIES_BUCKET)
+  parser.add_argument('--json-filename',
+                    help=('JSON summary filename to read for each builder; '
+                          'defaults to %(default)s.'),
+                    default=DEFAULT_JSON_FILENAME)
   parser.add_argument('--port', type=int,
                       help=('Which TCP port to listen on for HTTP requests; '
                             'defaults to %(default)s'),
                       default=DEFAULT_PORT)
   parser.add_argument('--reload', type=int,
                       help=('How often (a period in seconds) to update the '
                             'results.  If specified, both expected and actual '
                             'results will be updated by running "gclient sync" '
                             'on your Skia checkout as a whole.  '
                             'By default, we do not reload at all, and you '
                             'must restart the server to pick up new data.'),
                       default=0)
   args = parser.parse_args()
   if args.compare_configs:
     config_pairs = CONFIG_PAIRS_TO_COMPARE
   else:
     config_pairs = None
 
   global _SERVER
   _SERVER = Server(actuals_dir=args.actuals_dir,
-                   actuals_repo_revision=args.actuals_revision,
-                   actuals_repo_url=args.actuals_repo,
+                   json_filename=args.json_filename,
+                   gm_summaries_bucket=args.gm_summaries_bucket,
                    port=args.port, export=args.export, editable=args.editable,
                    reload_seconds=args.reload, config_pairs=config_pairs,
                    builder_regex_list=args.builders)
   _SERVER.run()
 
 
 if __name__ == '__main__':
   main()