#!/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.
"""
# System-level imports
import argparse
import BaseHTTPServer
import json
import logging
import os
import posixpath
import re
import shutil
import socket
import subprocess
import thread
import threading
import time
import urlparse
# Imports from within Skia
import fix_pythonpath # must do this first
from pyutils import gs_utils
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_GM_SUMMARIES_BUCKET = download_actuals.GM_SUMMARIES_BUCKET
DEFAULT_JSON_FILENAME = download_actuals.DEFAULT_JSON_FILENAME
DEFAULT_PORT = 8888
PARENT_DIRECTORY = os.path.dirname(os.path.realpath(__file__))
TRUNK_DIRECTORY = os.path.dirname(os.path.dirname(PARENT_DIRECTORY))
# 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'
GENERATED_IMAGES_SUBDIR = 'generated-images'
GENERATED_JSON_SUBDIR = 'generated-json'
# How often (in seconds) clients should reload while waiting for initial
# results to load.
RELOAD_INTERVAL_UNTIL_READY = 10
SUMMARY_TYPES = [
results_mod.KEY__HEADER__RESULTS_FAILURES,
results_mod.KEY__HEADER__RESULTS_ALL,
]
# If --compare-configs is specified, compare these configs.
CONFIG_PAIRS_TO_COMPARE = [('8888', 'gpu')]
_HTTP_HEADER_CONTENT_LENGTH = 'Content-Length'
_HTTP_HEADER_CONTENT_TYPE = 'Content-Type'
_SERVER = None # This gets filled in by main()
def _run_command(args, directory):
"""Runs a command and returns stdout as a single string.
Args:
args: the command to run, as a list of arguments
directory: directory within which to run the command
Returns: stdout, as a string
Raises an Exception if the command failed (exited with nonzero return code).
"""
logging.debug('_run_command: %s in directory %s' % (args, directory))
proc = subprocess.Popen(args, cwd=directory,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
(stdout, stderr) = proc.communicate()
if proc.returncode is not 0:
raise Exception('command "%s" failed in dir "%s": %s' %
(args, directory, stderr))
return stdout
def _get_routable_ip_address():
"""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_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
similar) to fill in dynamic content.
Args:
file_path: path on local disk to write index to; any directory components
of this path that do not already exist will be created
config_pairs: what pairs of configs (if any) we compare actual results of
"""
dir_path = os.path.dirname(file_path)
if not os.path.isdir(dir_path):
os.makedirs(dir_path)
with open(file_path, 'w') as file_handle:
file_handle.write(
''
'
rebaseline_server'
'')
if SUMMARY_TYPES:
file_handle.write('- Expectations vs Actuals
')
for summary_type in SUMMARY_TYPES:
file_handle.write(
'- '
''
'%s
' % (
STATIC_CONTENTS_SUBDIR, RESULTS_SUBDIR,
summary_type, summary_type))
file_handle.write('
')
if config_pairs:
file_handle.write('- Comparing configs within actual results
')
for config_pair in config_pairs:
file_handle.write('- %s vs %s:' % config_pair)
for summary_type in SUMMARY_TYPES:
file_handle.write(
' %s' % (
STATIC_CONTENTS_SUBDIR, STATIC_CONTENTS_SUBDIR,
GENERATED_JSON_SUBDIR, config_pair[0], config_pair[1],
summary_type, summary_type))
file_handle.write('
')
file_handle.write('
')
file_handle.write('
')
class Server(object):
""" HTTP server for our HTML rebaseline viewer. """
def __init__(self,
actuals_dir=DEFAULT_ACTUALS_DIR,
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
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._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)
# 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):
""" Returns the most recently generated results, or None if we don't have
any valid results (update_results() has not completed yet). """
return self._results
@property
def is_exported(self):
""" Returns true iff HTTP clients on other hosts are allowed to access
this server. """
return self._export
@property
def is_editable(self):
""" Returns true iff HTTP clients are allowed to submit new baselines. """
return self._editable
@property
def reload_seconds(self):
""" Returns the result reload period in seconds, or 0 if we don't reload
results. """
return self._reload_seconds
def update_results(self, invalidate=False):
""" Create or update self._results, based on the latest expectations and
actuals.
We hold self.results_rlock while we do this, to guarantee that no other
thread attempts to update either self._results or the underlying files at
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._gm_summaries_bucket:
logging.info(
'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:
gs_utils.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 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 commands.
#
# TODO(epoger): Fetch latest expectations in some other way.
# Eric points out that our official documentation recommends an
# unmanaged Skia checkout, so "gclient sync" will not bring down updated
# expectations from origin/master-- you'd have to do a "git pull" of
# some sort instead.
# However, the live rebaseline_server at
# http://skia-tree-status.appspot.com/redirect/rebaseline-server (which
# is probably the only user of the --reload flag!) uses a managed
# checkout, so "gclient sync" works in that case.
# Probably the best idea is to avoid all of this nonsense by fetching
# updated expectations into a temp directory, and leaving the rest of
# the checkout alone. This could be done using "git show", or by
# downloading individual expectation JSON files from
# skia.googlesource.com .
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,
GENERATED_IMAGES_SUBDIR),
diff_base_url=posixpath.join(
os.pardir, STATIC_CONTENTS_SUBDIR, GENERATED_IMAGES_SUBDIR),
builder_regex_list=self._builder_regex_list)
json_dir = os.path.join(
PARENT_DIRECTORY, STATIC_CONTENTS_SUBDIR, GENERATED_JSON_SUBDIR)
if not os.path.isdir(json_dir):
os.makedirs(json_dir)
for config_pair in self._config_pairs:
config_comparisons = compare_configs.ConfigComparisons(
configs=config_pair,
actuals_root=self._actuals_dir,
generated_images_root=os.path.join(
PARENT_DIRECTORY, STATIC_CONTENTS_SUBDIR,
GENERATED_IMAGES_SUBDIR),
diff_base_url=posixpath.join(
os.pardir, GENERATED_IMAGES_SUBDIR),
builder_regex_list=self._builder_regex_list)
for summary_type in SUMMARY_TYPES:
gm_json.WriteToFile(
config_comparisons.get_packaged_results_of_type(
results_type=summary_type),
os.path.join(
json_dir, '%s-vs-%s_%s.json' % (
config_pair[0], config_pair[1], summary_type)))
def _result_loader(self, reload_seconds=0):
""" Call self.update_results(), either once or periodically.
Params:
reload_seconds: integer; if nonzero, reload results at this interval
(in which case, this method will never return!)
"""
self.update_results()
logging.info('Initial results loaded. Ready for requests on %s' % self._url)
if reload_seconds:
while True:
time.sleep(reload_seconds)
self.update_results()
def run(self):
arg_tuple = (self._reload_seconds,) # start_new_thread needs a tuple,
# even though it holds just one param
thread.start_new_thread(self._result_loader, arg_tuple)
if self._export:
server_address = ('', self._port)
host = _get_routable_ip_address()
if self._editable:
logging.warning('Running with combination of "export" and "editable" '
'flags. Users on other machines will '
'be able to modify your GM expectations!')
else:
host = '127.0.0.1'
server_address = (host, self._port)
http_server = BaseHTTPServer.HTTPServer(server_address, HTTPRequestHandler)
self._url = 'http://%s:%d' % (host, http_server.server_port)
logging.info('Listening for requests on %s' % self._url)
http_server.serve_forever()
class HTTPRequestHandler(BaseHTTPServer.BaseHTTPRequestHandler):
""" HTTP request handlers for various types of queries this server knows
how to handle (static HTML and Javascript, expected/actual results, etc.)
"""
def do_GET(self):
"""
Handles all GET requests, forwarding them to the appropriate
do_GET_* dispatcher.
If we see any Exceptions, return a 404. This fixes http://skbug.com/2147
"""
try:
logging.debug('do_GET: path="%s"' % self.path)
if self.path == '' or self.path == '/' or self.path == '/index.html' :
self.redirect_to('/%s/%s/index.html' % (
STATIC_CONTENTS_SUBDIR, GENERATED_HTML_SUBDIR))
return
if self.path == '/favicon.ico' :
self.redirect_to('/%s/favicon.ico' % STATIC_CONTENTS_SUBDIR)
return
# All requests must be of this form:
# /dispatcher/remainder
# where 'dispatcher' indicates which do_GET_* dispatcher to run
# and 'remainder' is the remaining path sent to that dispatcher.
normpath = posixpath.normpath(self.path)
(dispatcher_name, remainder) = PATHSPLIT_RE.match(normpath).groups()
dispatchers = {
RESULTS_SUBDIR: self.do_GET_results,
STATIC_CONTENTS_SUBDIR: self.do_GET_static,
}
dispatcher = dispatchers[dispatcher_name]
dispatcher(remainder)
except:
self.send_error(404)
raise
def do_GET_results(self, results_type):
""" Handle a GET request for GM results.
Args:
results_type: string indicating which set of results to return;
must be one of the results_mod.RESULTS_* constants
"""
logging.debug('do_GET_results: sending results of type "%s"' % results_type)
# Since we must make multiple calls to the ExpectationComparisons object,
# grab a reference to it in case it is updated to point at a new
# ExpectationComparisons object within another thread.
#
# TODO(epoger): Rather than using a global variable for the handler
# to refer to the Server object, make Server a subclass of
# HTTPServer, and then it could be available to the handler via
# the handler's .server instance variable.
results_obj = _SERVER.results
if results_obj:
response_dict = results_obj.get_packaged_results_of_type(
results_type=results_type, reload_seconds=_SERVER.reload_seconds,
is_editable=_SERVER.is_editable, is_exported=_SERVER.is_exported)
else:
now = int(time.time())
response_dict = {
imagepairset.KEY__ROOT__HEADER: {
results_mod.KEY__HEADER__SCHEMA_VERSION: (
results_mod.VALUE__HEADER__SCHEMA_VERSION),
results_mod.KEY__HEADER__IS_STILL_LOADING: True,
results_mod.KEY__HEADER__TIME_UPDATED: now,
results_mod.KEY__HEADER__TIME_NEXT_UPDATE_AVAILABLE: (
now + RELOAD_INTERVAL_UNTIL_READY),
},
}
self.send_json_dict(response_dict)
def do_GET_static(self, path):
""" Handle a GET request for a file under STATIC_CONTENTS_SUBDIR .
Only allow serving of files within STATIC_CONTENTS_SUBDIR that is a
filesystem sibling of this script.
Args:
path: path to file (within STATIC_CONTENTS_SUBDIR) to retrieve
"""
# Strip arguments ('?resultsToLoad=all') from the path
path = urlparse.urlparse(path).path
logging.debug('do_GET_static: sending file "%s"' % path)
static_dir = os.path.realpath(os.path.join(
PARENT_DIRECTORY, STATIC_CONTENTS_SUBDIR))
full_path = os.path.realpath(os.path.join(static_dir, path))
if full_path.startswith(static_dir):
self.send_file(full_path)
else:
logging.error(
'Attempted do_GET_static() of path [%s] outside of static dir [%s]'
% (full_path, static_dir))
self.send_error(404)
def do_POST(self):
""" Handles all POST requests, forwarding them to the appropriate
do_POST_* dispatcher. """
# All requests must be of this form:
# /dispatcher
# where 'dispatcher' indicates which do_POST_* dispatcher to run.
logging.debug('do_POST: path="%s"' % self.path)
normpath = posixpath.normpath(self.path)
dispatchers = {
'/edits': self.do_POST_edits,
}
try:
dispatcher = dispatchers[normpath]
dispatcher()
self.send_response(200)
except:
self.send_error(404)
raise
def do_POST_edits(self):
""" Handle a POST request with modifications to GM expectations, in this
format:
{
KEY__EDITS__OLD_RESULTS_TYPE: 'all', # type of results that the client
# loaded and then made
# modifications to
KEY__EDITS__OLD_RESULTS_HASH: 39850913, # hash of results when the client
# loaded them (ensures that the
# client and server apply
# modifications to the same base)
KEY__EDITS__MODIFICATIONS: [
# as needed by compare_to_expectations.edit_expectations()
...
],
}
Raises an Exception if there were any problems.
"""
if not _SERVER.is_editable:
raise Exception('this server is not running in --editable mode')
content_type = self.headers[_HTTP_HEADER_CONTENT_TYPE]
if content_type != 'application/json;charset=UTF-8':
raise Exception('unsupported %s [%s]' % (
_HTTP_HEADER_CONTENT_TYPE, content_type))
content_length = int(self.headers[_HTTP_HEADER_CONTENT_LENGTH])
json_data = self.rfile.read(content_length)
data = json.loads(json_data)
logging.debug('do_POST_edits: received new GM expectations data [%s]' %
data)
# Update the results on disk with the information we received from the
# client.
# We must hold _SERVER.results_rlock while we do this, to guarantee that
# no other thread updates expectations (from the Skia repo) while we are
# updating them (using the info we received from the client).
with _SERVER.results_rlock:
oldResultsType = data[KEY__EDITS__OLD_RESULTS_TYPE]
oldResults = _SERVER.results.get_results_of_type(oldResultsType)
oldResultsHash = str(hash(repr(
oldResults[imagepairset.KEY__ROOT__IMAGEPAIRS])))
if oldResultsHash != data[KEY__EDITS__OLD_RESULTS_HASH]:
raise Exception('results of type "%s" changed while the client was '
'making modifications. The client should reload the '
'results and submit the modifications again.' %
oldResultsType)
_SERVER.results.edit_expectations(data[KEY__EDITS__MODIFICATIONS])
# Read the updated results back from disk.
# We can do this in a separate thread; we should return our success message
# to the UI as soon as possible.
thread.start_new_thread(_SERVER.update_results, (True,))
def redirect_to(self, url):
""" Redirect the HTTP client to a different url.
Args:
url: URL to redirect the HTTP client to
"""
self.send_response(301)
self.send_header('Location', url)
self.end_headers()
def send_file(self, path):
""" Send the contents of the file at this path, with a mimetype based
on the filename extension.
Args:
path: path of file whose contents to send to the HTTP client
"""
# Grab the extension if there is one
extension = os.path.splitext(path)[1]
if len(extension) >= 1:
extension = extension[1:]
# Determine the MIME type of the file from its extension
mime_type = MIME_TYPE_MAP.get(extension, MIME_TYPE_MAP[''])
# Open the file and send it over HTTP
if os.path.isfile(path):
with open(path, 'rb') as sending_file:
self.send_response(200)
self.send_header('Content-type', mime_type)
self.end_headers()
self.wfile.write(sending_file.read())
else:
self.send_error(404)
def send_json_dict(self, json_dict):
""" Send the contents of this dictionary in JSON format, with a JSON
mimetype.
Args:
json_dict: dictionary to send
"""
self.send_response(200)
self.send_header('Content-type', 'application/json')
self.end_headers()
json.dump(json_dict, self.wfile)
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)
# 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,
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()