Quelle  run_tests.py   Sprache: Python

#!/usr/bin/env python

# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at http://mozilla.org/MPL/2.0/.
import copy
import os
import pathlib
import sys
import time
import traceback

import mozinfo
import mozversion
import six
from mozgeckoprofiler import view_gecko_profile
from mozlog import get_proxy_logger
from wptserve import server

from talos import utils
from talos.config import ConfigurationError, get_configs
from talos.results import TalosResults
from talos.ttest import TTest
from talos.utils import TalosError, TalosRegression

# directory of this file
here = os.path.dirname(os.path.realpath(__file__))
LOG = get_proxy_logger()


def useBaseTestDefaults(base, tests):
    for test in tests:
        for item in base:
            if item not in test:
                test[item] = base[item]
                if test[item] is None:
                    test[item] = ""
    return tests


def set_tp_preferences(test, browser_config):
    # sanity check pageloader values
    # mandatory options: tpmanifest, tpcycles
    if test["tpcycles"] not in six.moves.range(1, 1000):
        raise TalosError("pageloader cycles must be int 1 to 1,000")
    if "tpmanifest" not in test:
        raise TalosError("tpmanifest not found in test: %s" % test)

    # if profiling is on, override tppagecycles to prevent test hanging
    if test["gecko_profile"]:
        LOG.info(
            "Gecko profiling is enabled so talos is reducing the number "
            "of cycles, please disregard reported numbers"
        )
        for cycle_var in ["tppagecycles", "tpcycles", "cycles"]:
            if test["name"] == "tp5o_scroll" and test[cycle_var] > 1:
                test[cycle_var] = 1
            elif test[cycle_var] > 2:
                test[cycle_var] = 2

    CLI_bool_options = [
        "tpchrome",
        "tphero",
        "tpmozafterpaint",
        "tploadnocache",
        "tpscrolltest",
        "fnbpaint",
        "pdfpaint",
        "a11y",
    ]
    CLI_options = ["tpcycles", "tppagecycles", "tptimeout", "tpmanifest"]
    for key in CLI_bool_options:
        _pref_name = "talos.%s" % key
        if key in test:
            test["preferences"][_pref_name] = test.get(key)
        else:
            # current test doesn't use this setting, remove it from our prefs
            if _pref_name in test["preferences"]:
                del test["preferences"][_pref_name]

    for key in CLI_options:
        value = test.get(key)
        _pref_name = "talos.%s" % key
        if value:
            test["preferences"][_pref_name] = value
        else:
            # current test doesn't use this setting, remove it from our prefs
            if _pref_name in test["preferences"]:
                del test["preferences"][_pref_name]


def setup_webserver(webserver):
    """Set up a new web server with wptserve."""
    LOG.info("starting webserver on %r" % webserver)

    host, port = webserver.split(":")
    httpd = server.WebTestHttpd(host=host, port=int(port), doc_root=here)

    return httpd


def skip_test(test_instance_dict, config):
    # Determines if a test should be skipped, and returns
    # a message with a reason why or None if it doesn't need
    # to be skipped
    if not test_instance_dict.get("pine", True) and config.get(
        "project", ""
    ).startswith("pine"):
        return "Broken on the pine branch"
    return None


def run_tests(config, browser_config):
    """Runs the talos tests on the given configuration and generates a report."""
    # get the test data
    tests = config["tests"]
    tests = useBaseTestDefaults(config.get("basetest", {}), tests)
    paths = ["profile_path", "tpmanifest", "extensions", "setup", "cleanup"]

    for test_index, test in enumerate(tests):
        if config.get("suite", False):
            test["suite"] = config["suite"]
        if test_index == 0:
            test["is_first_test"] = True
        # Check for profile_path, tpmanifest and interpolate based on Talos
        # root https://bugzilla.mozilla.org/show_bug.cgi?id=727711
        # Build command line from config
        for path in paths:
            if test.get(path):
                if path == "extensions":
                    for _index, _ext in enumerate(test["extensions"]):
                        test["extensions"][_index] = utils.interpolate(_ext)
                else:
                    test[path] = utils.interpolate(test[path])
        if test.get("tpmanifest"):
            test["tpmanifest"] = os.path.normpath(
                "file:/%s"
                % (six.moves.urllib.parse.quote(test["tpmanifest"], "/\\t:\\"))
            )
            test["preferences"]["talos.tpmanifest"] = test["tpmanifest"]

        # if using firstNonBlankPaint, set test preference for it
        # so that the browser pref will be turned on (in ffsetup)
        if test.get("fnbpaint", False):
            LOG.info("Test is using firstNonBlankPaint, browser pref will be turned on")
            test["preferences"][
                "dom.performance.time_to_non_blank_paint.enabled"
            ] = True

        test["setup"] = utils.interpolate(test["setup"])
        test["cleanup"] = utils.interpolate(test["cleanup"])

        if not test.get("profile", False):
            test["profile"] = config.get("profile")

    if mozinfo.os == "win":
        browser_config["extra_args"] = ["-wait-for-browser", "-no-deelevate"]
    else:
        browser_config["extra_args"] = []

    # Pass subtests filter argument via a preference
    if browser_config["subtests"]:
        browser_config["preferences"]["talos.subtests"] = browser_config["subtests"]

    if not browser_config.get("fission", True):
        browser_config["preferences"]["fission.autostart"] = False

    browser_config["preferences"]["network.proxy.type"] = 2
    browser_config["preferences"]["network.proxy.autoconfig_url"] = (
        """data:text/plain,
function FindProxyForURL(url, host) {
  if (url.startsWith('http')) {
   return 'PROXY %s';
  }
  return 'DIRECT';
}"""
        % browser_config["webserver"]
    )

    # If --code-coverage files are expected, set flag in browser config so ffsetup knows
    # that it needs to delete any ccov files resulting from browser initialization
    # NOTE: This is only supported in production; local setup of ccov folders and
    # data collection not supported yet, so if attempting to run with --code-coverage
    # flag locally, that is not supported yet
    if config.get("code_coverage", False):
        if browser_config["develop"]:
            raise TalosError(
                "Aborting: talos --code-coverage flag is only "
                "supported in production"
            )
        else:
            browser_config["code_coverage"] = True

    # set defaults
    testdate = config.get("testdate", "")

    # get the process name from the path to the browser
    if not browser_config["process"]:
        browser_config["process"] = os.path.basename(browser_config["browser_path"])

    # fix paths to substitute
    # `os.path.dirname(os.path.abspath(__file__))` for ${talos}
    # https://bugzilla.mozilla.org/show_bug.cgi?id=705809
    browser_config["extensions"] = [
        utils.interpolate(i) for i in browser_config["extensions"]
    ]
    browser_config["bcontroller_config"] = utils.interpolate(
        browser_config["bcontroller_config"]
    )

    # normalize browser path to work across platforms
    browser_config["browser_path"] = os.path.normpath(browser_config["browser_path"])

    binary = browser_config["browser_path"]
    version_info = mozversion.get_version(binary=binary)
    browser_config["browser_name"] = version_info["application_name"]
    browser_config["browser_version"] = version_info["application_version"]
    browser_config["buildid"] = version_info["application_buildid"]
    try:
        browser_config["repository"] = version_info["application_repository"]
        browser_config["sourcestamp"] = version_info["application_changeset"]
    except KeyError:
        if not browser_config["develop"]:
            print("Abort: unable to find changeset or repository: %s" % version_info)
            sys.exit(1)
        else:
            browser_config["repository"] = "develop"
            browser_config["sourcestamp"] = "develop"

    # get test date in seconds since epoch
    if testdate:
        date = int(time.mktime(time.strptime(testdate, "%a, %d %b %Y %H:%M:%S GMT")))
    else:
        date = int(time.time())
    LOG.debug("using testdate: %d" % date)
    LOG.debug("actual date: %d" % int(time.time()))

    # results container
    talos_results = TalosResults()

    # results links
    if not browser_config["develop"] and not config["gecko_profile"]:
        results_urls = dict(
            # another hack; datazilla stands for Perfherder
            # and do not require url, but a non empty dict is required...
            output_urls=["local.json"],
        )
    else:
        # local mode, output to files
        results_urls = dict(output_urls=[os.path.abspath("local.json")])

    httpd = setup_webserver(browser_config["webserver"])
    httpd.start()

    # legacy still required for perfherder data
    talos_results.add_extra_option("e10s")
    talos_results.add_extra_option("stylo")

    if config["gecko_profile"]:
        talos_results.add_extra_option("gecko-profile")

    # differentiate fission vs non-fission results in perfherder
    if browser_config.get("fission", True):
        talos_results.add_extra_option("fission")

    # differentiate webrender from non-webrender results
    if browser_config["preferences"].get("gfx.webrender.software", False):
        talos_results.add_extra_option("webrender-sw")
    else:
        # we need to add 'webrender' so reported data is consistent
        talos_results.add_extra_option("webrender")

    testname = None

    # run the tests
    timer = utils.Timer()
    LOG.suite_start(tests=[test["name"] for test in tests])
    try:
        for test in tests:
            testname = test["name"]
            LOG.test_start(testname)

            # Skip test if necessary
            skip_reason = skip_test(test, config)
            if skip_reason is not None and skip_reason != "":
                LOG.info("Skipping %s, reason: %s" % (testname, skip_reason))
                LOG.test_end(
                    testname,
                    status="SKIP",
                    message="Test skipped: %s" % skip_reason,
                )
                continue

            if not test.get("url"):
                # set browser prefs for pageloader test setings (doesn't use cmd line args / url)
                test["url"] = None
                set_tp_preferences(test, browser_config)

            mytest = TTest()

            utility_path = None
            if config.get("screenshot_on_failure"):
                obj_dir = os.environ.get("MOZ_DEVELOPER_OBJ_DIR", None)
                if obj_dir is None:
                    build_dir = pathlib.Path(os.environ.get("MOZ_UPLOAD_DIR")).parent
                    utility_path = pathlib.Path(build_dir, "tests", "bin")
                else:
                    utility_path = os.path.join(obj_dir, "dist", "bin")

            # some tests like ts_paint return multiple results in a single iteration
            if test.get("firstpaint", False) or test.get("userready", None):
                # we need a 'testeventmap' to tell us which tests each event should map to
                multi_value_result = None
                separate_results_list = []

                test_event_map = test.get("testeventmap", None)
                if test_event_map is None:
                    raise TalosError(
                        "Need 'testeventmap' in test.py for %s" % test.get("name")
                    )

                # run the test
                multi_value_result = mytest.runTest(browser_config, test)
                if multi_value_result is None:
                    raise TalosError(
                        "Abort: no results returned for %s" % test.get("name")
                    )

                # parse out the multi-value results, and 'fake it' to appear like separate tests
                separate_results_list = convert_to_separate_test_results(
                    multi_value_result, test_event_map
                )

                # now we have three separate test results, store them
                for test_result in separate_results_list:
                    talos_results.add(test_result)

            # some tests like bloom_basic run two separate tests and then compare those values
            # we want the results in perfherder to only be the actual difference between those
            # and store the base and reference test replicates in results.json for upload
            elif test.get("base_vs_ref", False):
                # run the test, results will be reported for each page like two tests in the suite
                base_and_reference_results = mytest.runTest(
                    browser_config, test, utility_path=utility_path
                )
                # now compare each test, and create a new test object for the comparison
                talos_results.add(make_comparison_result(base_and_reference_results))
            else:
                # just expecting regular test - one result value per iteration
                talos_results.add(
                    mytest.runTest(browser_config, test, utility_path=utility_path)
                )
            LOG.test_end(testname, status="OK")

    except TalosRegression as exc:
        LOG.error("Detected a regression for %s" % testname)
        # by returning 1, we report an orange to buildbot
        # http://docs.buildbot.net/latest/developer/results.html
        LOG.test_end(
            testname, status="FAIL", message=str(exc), stack=traceback.format_exc()
        )
        return 1
    except Exception as exc:
        # NOTE: if we get into this condition, talos has an internal
        # problem and cannot continue
        #       this will prevent future tests from running
        LOG.test_end(
            testname, status="ERROR", message=str(exc), stack=traceback.format_exc()
        )
        # indicate a failure to buildbot, turn the job red
        return 2
    finally:
        LOG.suite_end()
        httpd.stop()

    LOG.info("Completed test suite (%s)" % timer.elapsed())

    if talos_results.has_results():
        # output results
        if results_urls and not browser_config["no_upload_results"]:
            talos_results.output(results_urls)
            if browser_config["develop"] or config["gecko_profile"]:
                print(
                    "Thanks for running Talos locally. Results are in %s"
                    % (results_urls["output_urls"])
                )

        # when running talos locally with gecko profiling on, use the view-gecko-profile
        # tool to automatically load the latest gecko profile in profiler.firefox.com
        if config["gecko_profile"] and browser_config["develop"]:
            if os.environ.get("DISABLE_PROFILE_LAUNCH", "0") == "1":
                LOG.info(
                    "Not launching profiler.firefox.com because DISABLE_PROFILE_LAUNCH=1"
                )
            else:
                view_gecko_profile_from_talos()
    else:
        LOG.error("No tests ran")
        return 2

    # we will stop running tests on a failed test, or we will return 0 for
    # green
    return 0


def view_gecko_profile_from_talos():
    profile_zip_path = os.environ.get("TALOS_LATEST_GECKO_PROFILE_ARCHIVE", None)
    if profile_zip_path is None or not os.path.exists(profile_zip_path):
        LOG.info(
            "No local talos gecko profiles were found so not launching profiler.firefox.com"
        )
        return

    LOG.info("Profile saved locally to: %s" % profile_zip_path)
    view_gecko_profile(profile_zip_path)


def make_comparison_result(base_and_reference_results):
    """Receive a test result object meant to be used as a base vs reference test. The result
    object will have one test with two subtests; instead of traditional subtests we want to
    treat them as separate tests, comparing them together and reporting the comparison results.

    Results with multiple pages used as subtests would look like this normally, with the overall
    result value being the mean of the pages/subtests:

    PERFHERDER_DATA: {"framework": {"name": "talos"}, "suites": [{"extraOptions": ["e10s"],
    "name": "bloom_basic", "lowerIsBetter": true, "alertThreshold": 5.0, "value": 594.81,
    "subtests": [{"name": ".html", "lowerIsBetter": true, "alertThreshold": 5.0, "replicates":
    [586.52, ...], "value": 586.52], "unit": "ms"}, {"name": "-ref.html", "lowerIsBetter": true,
    "alertThreshold": 5.0, "replicates": [603.225, ...], "value": 603.225, "unit": "ms"}]}]}

    We want to compare the subtests against eachother (base vs ref) and create a new single test
    results object with the comparison results, that will look like traditional single test results
    like this:

    PERFHERDER_DATA: {"framework": {"name": "talos"}, "suites": [{"lowerIsBetter": true,
    "subtests": [{"name": "", "lowerIsBetter": true, "alertThreshold": 5.0, "replicates":
    [16.705, ...], "value": 16.705, "unit": "ms"}], "extraOptions": ["e10s"], "name":
    "bloom_basic", "alertThreshold": 5.0}]}
    """
    # create a new results object for the comparison result; keep replicates from both pages
    comparison_result = copy.deepcopy(base_and_reference_results)

    # remove original results from our copy as they will be replaced by one comparison result
    comparison_result.results[0].results = []
    comp_results = comparison_result.results[0].results

    # zero-based count of how many base vs reftest sets we have
    subtest_index = 0

    # each set of two results is actually a base test followed by the
    # reference test; we want to go through each set of base vs reference
    for x in range(0, len(base_and_reference_results.results[0].results), 2):
        # separate the 'base' and 'reference' result run values
        results = base_and_reference_results.results[0].results
        base_result_runs = results[x]["runs"]
        ref_result_runs = results[x + 1]["runs"]

        # the test/subtest result is the difference between the base vs reference test page
        # values; for this result use the base test page name for the subtest/test name
        sub_test_name = base_and_reference_results.results[0].results[x]["page"]

        # populate our new comparison result with 'base' and 'ref' replicates
        comp_results.append(
            {
                "index": 0,
                "runs": [],
                "page": sub_test_name,
                "base_runs": base_result_runs,
                "ref_runs": ref_result_runs,
            }
        )

        # now step thru each result, compare 'base' vs 'ref', and store the difference in 'runs'
        _index = 0
        for next_ref in comp_results[subtest_index]["ref_runs"]:
            diff = abs(next_ref - comp_results[subtest_index]["base_runs"][_index])
            # pylint: disable=W1633
            comp_results[subtest_index]["runs"].append(float(round(diff, 3)))
            _index += 1

        # increment our base vs reference subtest index
        subtest_index += 1

    return comparison_result


def convert_to_separate_test_results(multi_value_result, test_event_map):
    """Receive a test result that actually contains multiple values in a single iteration, and
    parse it out in order to 'fake' three seprate test results.

    Incoming result looks like this:

    [{'index': 0, 'runs': {'event_1': [1338, ...], 'event_2': [1438, ...], 'event_3':
    [1538, ...]}, 'page': 'NULL'}]

    We want to parse it out such that we have 'faked' three separate tests, setting test names
    and taking the run values for each. End goal is to have results reported as three separate
    tests, like this:

    PERFHERDER_DATA: {"framework": {"name": "talos"}, "suites": [{"subtests": [{"replicates":
    [1338, ...], "name": "ts_paint", "value": 1338}], "extraOptions": ["e10s"], "name":
    "ts_paint"}, {"subtests": [{"replicates": [1438, ...], "name": "ts_first_paint", "value":
    1438}], "extraOptions": ["e10s"], "name": "ts_first_paint"}, {"subtests": [{"replicates":
    [1538, ...], "name": "ts_user_ready", "value": 1538}], "extraOptions": ["e10s"], "name":
    "ts_user_ready"}]}
    """
    list_of_separate_tests = []

    for next_test in test_event_map:
        # copy the original test result that has multiple values per iteration
        separate_test = copy.deepcopy(multi_value_result)
        # set the name of the new 'faked' test
        separate_test.test_config["name"] = next_test["name"]
        # set the run values for the new test
        for x in separate_test.results:
            for item in x.results:
                all_runs = item["runs"]
                item["runs"] = all_runs[next_test["label"]]
        # add it to our list of results to return
        list_of_separate_tests.append(separate_test)

    return list_of_separate_tests


def main(args=sys.argv[1:]):
    try:
        config, browser_config = get_configs()
    except ConfigurationError as exc:
        sys.exit("ERROR: %s" % exc)
    sys.exit(run_tests(config, browser_config))


if __name__ == "__main__":
    main()