from __future__ import print_function import errno import functools import hashlib import json import os import os.path import pickle import re import subprocess import sys import timeit import yaml import collections from functools import wraps from os.path import join REPO_SUFFIX = '.benchrepo' BASEDIR = os.path.abspath(os.path.join( os.path.dirname(__file__), os.path.pardir, os.path.pardir )) STRIP_VARIANTS_PATH = os.path.join(BASEDIR, "partial-sets.yaml") def read_yaml_file(config_path): with open(config_path) as config_file: return yaml.safe_load(config_file.read()) repodir = read_yaml_file(os.path.join(BASEDIR, "config.yaml")).get('repodir') if repodir: REPOS_DIR = os.path.abspath(repodir) else: REPOS_DIR = os.path.join(BASEDIR, "repos") PERFEXTSDIR = os.path.join(BASEDIR, "perfexts") FORMAT_VARIANT_PREFIX = "repo-format" def get_repo_variants(repo_dir): """read all .benchrepo in given directory and computes ASV variants Reads all .benchrepo files in the given directory and returns a tuple of 5 elements: - a list of params list. The first element will be a list of repo ids, the following ones will be the possible values for each repository format key, - a list of params name, the first one will be "repo", - the length of the params name, - a dict of the parsed `.benchrepofile`, the key is the benchrepo file name without the `.benchrepo` extension. - a dict of repo hash, the key is the following tuple (repo_id, ((format_key_1, format_value_1), (format_key_2, format_value_1))) For example, given a directory with two benchrepo files: repo-test-1-XXX.benchrepo: reference-repo: format-info: compression: zlib dotencode: true fncache: true generaldelta: false plain-cl-delta: true sparserevlog: false id: repo-test-1 repo-hash: XXX repo-test-1-YYY.benchrepo: reference-repo: format-info: compression: zlib dotencode: true fncache: true generaldelta: true plain-cl-delta: true sparserevlog: false id: repo-test-1 repo-hash: YYY get_repo_variants will returns: - the params: [['repo-test-1'], ['zlib'], [True], [True], [False, True], [True], [False]] - the params_names: ['repo', 'compression', 'dotencode', 'fncache', 'generaldelta', 'plain-cl-delta', 'sparserevlog'] - the length of params: 7 - the dict of parsed benchrepo data: {'repo-test-1-XXX': {...}, 'repo-test-1-YYY: {...}} - the dict of repo hash: {('repo-test-1', (('compression', 'zlib'), ('dotencode', True), ('fncache', True), ('generaldelta', False), ('plain-cl-delta', True), ('sparserevlog', False))): 'XXX', ('repo-test-1', (('compression', 'zlib'), ('dotencode', True), ('fncache', True), ('generaldelta', True), ('plain-cl-delta', True), ('sparserevlog', False))): 'YYY'}) """ # Find all benchrepos if not os.path.isdir(repo_dir): return ([], [], 0, {}, {}, [], {}) benchrepos_files = sorted( d for d in os.listdir(repo_dir) if d.endswith(REPO_SUFFIX) ) repo_ids = set() format_variants = collections.defaultdict(set) repo_details = {} repo_map = {} partial_sets_set = set() partial_sets_roles = {} # Parse the benchrepos files for benchrepo in benchrepos_files: data = read_yaml_file(join(repo_dir, benchrepo)) reference_repo = data["reference-repo"] repo_id = reference_repo["id"] repo_ids.add(repo_id) repo_name = reference_repo['repo-prefix'] repo_details[repo_name] = data format_info = reference_repo.get("format-info", {}) partial_sets = data.get("partial-sets", {}) if not partial_sets: # compat layer for the level transition. We can drop this later partial_sets = reference_repo.get("partial-sets", {}) partial_sets_set.update(partial_sets) partial_roles = data.get("roles", {}) # We are parsing this kind of data # # roles: # pull: # noop: # same: # source: "reference" # target: "same" for action, subtypes in partial_roles.items(): action_map = partial_sets_roles.setdefault(action, {}) for st, variants in subtypes.items(): subtypes_map = action_map.setdefault(st, {}) for v, pdata in variants.items(): variant_map = subtypes_map.setdefault(v, {}) repo_key = repo_name variant_map[repo_key] = pdata # Add a prefix to format-info keys new_format_info = {} for format_name, format_value in format_info.items(): new_format_name = "{}-{}".format(FORMAT_VARIANT_PREFIX, format_name) new_format_info[new_format_name] = format_value format_variants[new_format_name].add(format_value) repo_map_key = (repo_id, tuple(sorted(new_format_info.items()))) repo_map[repo_map_key] = repo_name params = [list(sorted(repo_ids))] params_names = ["repo"] for format_name in sorted(format_variants): params_names.append(format_name) params.append(list(sorted(format_variants[format_name]))) return ( params, params_names, len(params_names), repo_details, repo_map, list(sorted(partial_sets_set)), partial_sets_roles, ) VARIANTS = get_repo_variants(REPOS_DIR) BASE_PARAMS, BASE_PARAMS_NAMES, BASE_PARAMS_LEN = VARIANTS[:3] REPO_DETAILS, REPO_HASH_MAP, STRIP_VARIANTS_LIST = VARIANTS[3:6] ROLES = VARIANTS[6] class SkipResult(Exception): pass # Backward compatibility for python 2.6 if not hasattr(subprocess, 'check_output'): STDOUT = subprocess.STDOUT def check_output(*popenargs, **kwargs): if 'stdout' in kwargs: # pragma: no cover raise ValueError('stdout argument not allowed, ' 'it will be overridden.') process = subprocess.Popen(stdout=subprocess.PIPE, *popenargs, **kwargs) output, _ = process.communicate() retcode = process.poll() if retcode: cmd = kwargs.get("args") if cmd is None: cmd = popenargs[0] raise subprocess.CalledProcessError(retcode, cmd, output=output) return output subprocess.check_output = check_output # overwrite CalledProcessError due to `output` # keyword not being available (in 2.6) class CalledProcessError(Exception): def __init__(self, returncode, cmd, output=None): self.returncode = returncode self.cmd = cmd self.output = output def __str__(self): return "Command '%s' returned non-zero exit status %d\n%s" % ( self.cmd, self.returncode, self.output) subprocess.CalledProcessError = CalledProcessError PERF_RE = re.compile(r'! wall (\d+.\d+) comb (\d+.\d+) user (\d+.\d+) sys (\d+.\d+) \(best of (\d+)\)') # noqa: E501 ### # Base classes for benchmarks ### def median(lst): quotient, remainder = divmod(len(lst), 2) if remainder: return sorted(lst)[quotient] return sum(sorted(lst)[quotient - 1:quotient + 1]) / 2. def params_as_kwargs(f): """Pass in test parameters as keyword arguments. Use as a decorator on BaseTestSuite methods """ @functools.wraps(f) def wrapper(self, *args, **kwargs): names = self.param_names args, values = args[len(names):], args[:len(names)] kwargs.update(zip(names, values)) return f(self, *args, **kwargs) return wrapper def not_compatible_with(revset, filter_fn=None): """Specifies the revset wherein the command is NOT expected to work. Skips the benchmark if the command is not expected to work. The current version is obtained from the environment. """ def decorator(f): @functools.wraps(f) def wrapper(self, *args, **kwargs): current_version = os.environ.get('ASV_COMMIT') if self.should_skip_benchmark(incompatibility_revset=revset, current_version=current_version, filter_fn=filter_fn, test_kwargs=kwargs): message = "{} is not expected to work for hash {}" print( message.format(f.__name__, current_version), file=sys.stderr ) raise NotImplementedError() return f(self, *args, **kwargs) return wrapper return decorator class BaseTestSuite(object): timer = timeit.default_timer repeat = (3, 10, 60.0) number = 1 params = BASE_PARAMS param_names = BASE_PARAMS_NAMES @params_as_kwargs def setup(self, repo, **kwargs): venv = os.path.abspath(os.path.join(os.path.dirname(sys.executable), "..")) self.project_dir = os.path.join(venv, 'project') if os.path.isdir(self.project_dir): # use hg in virtualenv for "asv run" self.hgpath = os.path.join(venv, "bin", "hg") else: # use local hg for "asv dev" self.project_dir = os.path.join(BASEDIR, 'mercurial') sys.path.insert(0, self.project_dir) self.hgpath = os.path.join(os.path.join(self.project_dir, 'hg')) repo_name = self.get_repo_name(repo, **kwargs) if repo_name is None: raise NotImplementedError( "not repository for formats/params combination, skipping." ) self.repo_id = repo self.repo_name = repo_name reference_basename = "{}-reference".format(self.repo_name) self.repo_path = os.path.join(REPOS_DIR, reference_basename) ### Use a clean environ to run command # # (We rely on running from the vevn using an explicite path) self.environ = {'HGRCPATH': ''} # keep some environment variables # SSH_AUTH_SOCK for hg over ssh for key in ('SSH_AUTH_SOCK', 'HGMODULEPOLICY'): if key in os.environ: self.environ[key] = os.environ[key] # define the cache dir for simplicity self._cache_dir = os.getcwd() # Don't run if using Rust on a revision before the Rust modulepolicy # was introduced if os.environ.get("HGMODULEPOLICY").startswith("rust"): # Don't use `subprocess.PIPE` as it can deadlock (see docs). # We capture `stderr` because the error message from `hg log` would # make it seem like there was a bug. currentrev = self.get_asv_rev() revset = "94167e701e125dce1788e19b1e1489958235e40c::" if not self._matchrevset(revset, currentrev): raise NotImplementedError( "Rust modulepolicy was not implemented before " "revision 94167e701e12" ", skipping." ) def _matchrevset(self, revset, target_rev): """return True if <target-rev> is within <revset>""" tmpl = "({}) and ({})" full_revset = tmpl.format(revset, target_rev) match_pattern = 'ASV-REVSET-MATCH' # TODO use `hg test` if/when it is stable command = [ self.hgpath, '--cwd', os.path.join(BASEDIR, "mercurial"), 'log', '-r', full_revset, '--template', match_pattern ] try: output = self.check_output(*command, env=self.environ) except subprocess.CalledProcessError as exc: if exc.returncode == 255: return False else: raise return match_pattern in output def should_skip_benchmark(self, incompatibility_revset, current_version, filter_fn, test_kwargs): """Determines whether the benchmark should be run given an exclusion revset, the current mercurial version and an optional filter function. The filter function can be used for deciding more precisely (most likely based on test_kwargs) if the benchmark should be skipped. It must return a boolean, `True` if it should indeed be skipped, `False` otherwise. """ in_revset = True in_filter = False if incompatibility_revset is not None and current_version: in_revset = self._matchrevset(incompatibility_revset, current_version) in_filter = True if filter_fn is not None: in_filter = bool(filter_fn(test_kwargs, current_version)) return in_revset and in_filter @staticmethod def get_skip(): with open(os.path.join(REPOS_DIR, 'skip.json'), 'r') as f: return json.load(f) def get_repo_name(self, repo, **kwargs): # Old way if BASE_PARAMS_LEN == 1: return repo # Filter the format variants variants = {} for key, value in kwargs.items(): if not key.startswith(FORMAT_VARIANT_PREFIX): continue variants[key] = value repo_hash_key = (repo, tuple(sorted(variants.items()))) repo_name = REPO_HASH_MAP.get(repo_hash_key) return repo_name def get_asv_rev(self): '''Return currently benchmarked mercurial revision''' return self.hg('log', '--config', 'experimental.evolution=all', '--cwd', self.project_dir, '--template', '{node|short}', '-r', '.').strip() def check_output(self, *args, **kwargs): """Helper to run commands Run given command in a subprocess Optional expected_return_code (default 0) is used to control whenever we expect the command should exit. If the command succeeded with expected_return_code = 0, return the output If the command succeeded with expected_return_code != 0, raise RuntimeError If the command fail with expected_return_code, return None, else raise original subprocess.CalledProcessError exception. """ env = kwargs.pop('env', self.environ) expected_return_code = kwargs.pop('expected_return_code', 0) cmd = list(args) try: output = subprocess.check_output(cmd, env=env, **kwargs) except subprocess.CalledProcessError as exc: if exc.returncode == expected_return_code: # failed as we expected return None raise else: if expected_return_code != 0: raise RuntimeError('unexpected return code 0 for {}'.format(cmd)) return output def hg(self, *args, **kwargs): """Run given command arguments with hg When there is no '--cwd' in arguments, use the benchmarked repo with 'hg --cwd /path/to/repo' """ args = list(args) # disabled multi worker because out current test setup is bad with multi CPU args = ["--config", "worker.enabled=no"] + args if '--cwd' not in args: # use self.repo_path as repo args = ['--cwd', self.repo_path] + args cmd = [self.hgpath] + list(args) return self.check_output(*cmd, **kwargs) def safe_hg(self, command, *args, **kwargs): """Run given command argument with hg and ignore unknown commands This is to be used for commands that may not exist in earlier mercurial versions. When hg exit code is 255, test the command existence with 'hg help CMD' to test command existence, if it also return 255, raise SkipResult exception. We expect `command` to be the hg command we want to run, either a list of arguments required to run 'hg help' (for extensions etc). """ if isinstance(command, (list, tuple)): cmd = command else: cmd = [command] try: return self.hg(*(cmd + list(args)), **kwargs) except subprocess.CalledProcessError as exc: if exc.returncode == 255: # test if it return 255 because the command does not exist # or if it's another issue try: self.hg(*(['help'] + command)) except subprocess.CalledProcessError as exc: if exc.returncode == 255: # command does not exist in this version of mercurial raise SkipResult() raise raise def _getperfext(self, version): """given a mercurial-core node (or None) return a path to perfext If a mecurial-core node is provided, the path will point to a file with the perf.py content at that revision. """ if version is None: return os.path.join(self.project_dir, 'contrib', 'perf.py') filename = 'perf-%s.py' % version extpath = os.path.join(PERFEXTSDIR, filename) if not os.path.exists(extpath): # XXX this is a bit fragile because if anything goes wrong, we create and empty file with open(extpath, 'wb') as f: subprocess.check_call( [ 'hg', '--cwd', os.path.join(BASEDIR, 'mercurial'), 'cat', '--rev', version, 'contrib/perf.py', ], stdout=f ) return extpath def _perfext(self, command, *args, **kwargs): """Use contrib/perf.py extension from mercurial to get data""" kwargs = kwargs.copy() perfextversion = kwargs.pop('perfextversion', None) perfpath = self._getperfext(perfextversion) kwargs.setdefault('stderr', subprocess.STDOUT) try: return self.safe_hg( [command, '--config', 'extensions.perf={0}'.format(perfpath)], *args, **kwargs) except SkipResult: # command does not exist for this version of perf.py # return NaN which is a "n/a" status for asv return None except subprocess.CalledProcessError as exc: msg = "calling perf command failed [%d]: %s %s" % (exc.returncode, command, args) print(msg, file=sys.stderr) print(exc.output, file=sys.stderr) def perfext(self, command, *args, **kwargs): """Use contrib/perf.py extension from mercurial to get a benchmark result""" output = self._perfext(command, *args, **kwargs) if output is None: return float('nan') match = PERF_RE.search(output) if not match: raise ValueError("Invalid output {0}".format(output)) return float(match.group(1)) def perfextjson(self, command, *args, **kwargs): """Use contrib/perf.py extension from mercurial to get a benchmark result""" args = ['--template', 'json', '--config', 'perf.all-timing=yes'] + list(args) output = self._perfext(command, *args, **kwargs) if output is None: return None try: data = json.loads(output) except ValueError: return None result = {} for item in data: title = item.get("title") if title in result: # multiple conflicting entry, skipping this one print("ignoring duplicated entry in json output:", title, file=sys.stderr) continue result[title] = localdata = { "count": item.get("count"), "minimum": item.get("wall"), } if 'median.wall' in item: localdata['median'] = item['median.wall'] if 'avg.wall' in item: localdata['average'] = item['avg.wall'] if 'max.wall' in item: localdata['maximum'] = item['max.wall'] return result def getperfdata(self, key, cmd): """Retrieve performance data from running cmd The result is cached to avoid running the same benchmark multiple time. Use this to handle commands from the perf extensions that returns multiple values.""" key = (repr(self.__class__), self.get_asv_rev(), self.repo_name) + key try: return self._get_cache(key) except KeyError: data = self.perfextjson(*cmd) self._set_cache(key, data) return data def _cache_file(self, key): filename = hashlib.sha256(repr(key)).hexdigest() return os.path.join(self._cache_dir, 'perf-cache', filename) def _set_cache(self, key, data): cachefile = self._cache_file(key) cachedir = os.path.dirname(cachefile) if not os.path.isdir(cachedir): os.mkdir(cachedir) with open(cachefile, 'wb') as f: pickle.dump(data, f) def _get_cache(self, key): cachefile = self._cache_file(key) try: with open(cachefile, 'rb') as f: return pickle.load(f) except IOError as exc: if not exc.errno == errno.ENOENT: raise raise KeyError(key) def setup_cache(self): self._cache_dir = os.getcwd() def teardown(self, *args, **kwargs): # only here for consintency and ease; you can use super().teardown() # in subclasses to mirror super().setup(). pass class BaseNChangesetsTestSuite(BaseTestSuite): params = BaseTestSuite.params + [[10, 100, 1000, 10000]] param_names = BaseTestSuite.param_names + ["changesets"]