Changeset - 6fb68819e58e
[Not reviewed]
default
0 1 0
Eivind Tagseth - 8 years ago 2017-07-01 21:47:30
eivindt@gmail.com
git: improve performance working with git changesets

GitRepository._repo instantiates a new dulwich.repo.Repo on every usage,
rather than once at initialization time of GitRepository. As this involves a
lot of filesystem access, this is a costly operation.

Instead, let GitRepository.__init__ instantiate a dulwich.repo.Repo once,
and let GitRepository._repo just return it.

This improves performance significantly.
On test_graphmod_git, performance improves from 6.29 seconds median to 3.06
seconds median.

[Thomas De Schampheleire: extend improvement to _all_ usage of
GitRepository._repo instead of only some. To limit the delta, retain the
_repo property but simply return self.repo.]
1 file changed with 4 insertions and 4 deletions:
0 comments (0 inline, 0 general)
kallithea/lib/vcs/backends/git/repository.py
Show inline comments
 
@@ -14,110 +14,110 @@ import re
 
import time
 
import errno
 
import urllib
 
import urllib2
 
import logging
 
import posixpath
 

	
 
from dulwich.objects import Tag
 
from dulwich.repo import Repo, NotGitRepository
 
from dulwich.config import ConfigFile
 

	
 
from kallithea.lib.vcs import subprocessio
 
from kallithea.lib.vcs.backends.base import BaseRepository, CollectionGenerator
 
from kallithea.lib.vcs.conf import settings
 

	
 
from kallithea.lib.vcs.exceptions import (
 
    BranchDoesNotExistError, ChangesetDoesNotExistError, EmptyRepositoryError,
 
    RepositoryError, TagAlreadyExistError, TagDoesNotExistError
 
)
 
from kallithea.lib.vcs.utils import safe_str, safe_unicode, makedate, date_fromtimestamp
 
from kallithea.lib.vcs.utils.lazy import LazyProperty
 
from kallithea.lib.vcs.utils.ordered_dict import OrderedDict
 
from kallithea.lib.vcs.utils.paths import abspath, get_user_home
 

	
 
from kallithea.lib.vcs.utils.hgcompat import (
 
    hg_url, httpbasicauthhandler, httpdigestauthhandler
 
)
 

	
 
from .changeset import GitChangeset
 
from .inmemory import GitInMemoryChangeset
 
from .workdir import GitWorkdir
 

	
 
SHA_PATTERN = re.compile(r'^[[0-9a-fA-F]{12}|[0-9a-fA-F]{40}]$')
 

	
 
log = logging.getLogger(__name__)
 

	
 

	
 
class GitRepository(BaseRepository):
 
    """
 
    Git repository backend.
 
    """
 
    DEFAULT_BRANCH_NAME = 'master'
 
    scm = 'git'
 

	
 
    def __init__(self, repo_path, create=False, src_url=None,
 
                 update_after_clone=False, bare=False):
 

	
 
        self.path = safe_unicode(abspath(repo_path))
 
        repo = self._get_repo(create, src_url, update_after_clone, bare)
 
        self.bare = repo.bare
 
        self.repo = self._get_repo(create, src_url, update_after_clone, bare)
 
        self.bare = self.repo.bare
 

	
 
    @property
 
    def _config_files(self):
 
        return [
 
            self.bare and abspath(self.path, 'config')
 
                      or abspath(self.path, '.git', 'config'),
 
             abspath(get_user_home(), '.gitconfig'),
 
         ]
 

	
 
    @property
 
    def _repo(self):
 
        return Repo(self.path)
 
        return self.repo
 

	
 
    @property
 
    def head(self):
 
        try:
 
            return self._repo.head()
 
        except KeyError:
 
            return None
 

	
 
    @property
 
    def _empty(self):
 
        """
 
        Checks if repository is empty ie. without any changesets
 
        """
 

	
 
        try:
 
            self.revisions[0]
 
        except (KeyError, IndexError):
 
            return True
 
        return False
 

	
 
    @LazyProperty
 
    def revisions(self):
 
        """
 
        Returns list of revisions' ids, in ascending order.  Being lazy
 
        attribute allows external tools to inject shas from cache.
 
        """
 
        return self._get_all_revisions()
 

	
 
    @classmethod
 
    def _run_git_command(cls, cmd, **opts):
 
        """
 
        Runs given ``cmd`` as git command and returns tuple
 
        (stdout, stderr).
 

	
 
        :param cmd: git command to be executed
 
        :param opts: env options to pass into Subprocess command
 
        """
 

	
 
        if '_bare' in opts:
 
            _copts = []
 
            del opts['_bare']
 
        else:
 
            _copts = ['-c', 'core.quotepath=false', ]
 
        safe_call = False
 
        if '_safe' in opts:
 
            #no exc on failure
 
            del opts['_safe']
 
            safe_call = True
 
@@ -194,97 +194,97 @@ class GitRepository(BaseRepository):
 
            passmgr.add_password(*authinfo)
 

	
 
            handlers.extend((httpbasicauthhandler(passmgr),
 
                             httpdigestauthhandler(passmgr)))
 

	
 
        o = urllib2.build_opener(*handlers)
 
        o.addheaders = [('User-Agent', 'git/1.7.8.0')]  # fake some git
 

	
 
        q = {"service": 'git-upload-pack'}
 
        qs = '?%s' % urllib.urlencode(q)
 
        cu = "%s%s" % (test_uri, qs)
 
        req = urllib2.Request(cu, None, {})
 

	
 
        try:
 
            resp = o.open(req)
 
            if resp.code != 200:
 
                raise Exception('Return Code is not 200')
 
        except Exception as e:
 
            # means it cannot be cloned
 
            raise urllib2.URLError("[%s] org_exc: %s" % (cleaned_uri, e))
 

	
 
        # now detect if it's proper git repo
 
        gitdata = resp.read()
 
        if not 'service=git-upload-pack' in gitdata:
 
            raise urllib2.URLError(
 
                "url [%s] does not look like an git" % (cleaned_uri))
 

	
 
        return True
 

	
 
    def _get_repo(self, create, src_url=None, update_after_clone=False,
 
                  bare=False):
 
        if create and os.path.exists(self.path):
 
            raise RepositoryError("Location already exist")
 
        if src_url and not create:
 
            raise RepositoryError("Create should be set to True if src_url is "
 
                                  "given (clone operation creates repository)")
 
        try:
 
            if create and src_url:
 
                GitRepository._check_url(src_url)
 
                self.clone(src_url, update_after_clone, bare)
 
                return Repo(self.path)
 
            elif create:
 
                os.makedirs(self.path)
 
                if bare:
 
                    return Repo.init_bare(self.path)
 
                else:
 
                    return Repo.init(self.path)
 
            else:
 
                return self._repo
 
                return Repo(self.path)
 
        except (NotGitRepository, OSError) as err:
 
            raise RepositoryError(err)
 

	
 
    def _get_all_revisions(self):
 
        # we must check if this repo is not empty, since later command
 
        # fails if it is. And it's cheaper to ask than throw the subprocess
 
        # errors
 
        try:
 
            self._repo.head()
 
        except KeyError:
 
            return []
 

	
 
        rev_filter = settings.GIT_REV_FILTER
 
        cmd = ['rev-list', rev_filter, '--reverse', '--date-order']
 
        try:
 
            so, se = self.run_git_command(cmd)
 
        except RepositoryError:
 
            # Can be raised for empty repositories
 
            return []
 
        return so.splitlines()
 

	
 
    def _get_all_revisions2(self):
 
        #alternate implementation using dulwich
 
        includes = [x[1][0] for x in self._parsed_refs.iteritems()
 
                    if x[1][1] != 'T']
 
        return [c.commit.id for c in self._repo.get_walker(include=includes)]
 

	
 
    def _get_revision(self, revision):
 
        """
 
        For git backend we always return integer here. This way we ensure
 
        that changeset's revision attribute would become integer.
 
        """
 

	
 
        is_null = lambda o: len(o) == revision.count('0')
 

	
 
        if self._empty:
 
            raise EmptyRepositoryError("There are no changesets yet")
 

	
 
        if revision in (None, '', 'tip', 'HEAD', 'head', -1):
 
            return self.revisions[-1]
 

	
 
        is_bstr = isinstance(revision, (str, unicode))
 
        if ((is_bstr and revision.isdigit() and len(revision) < 12)
 
            or isinstance(revision, int) or is_null(revision)):
 
            try:
 
                revision = self.revisions[int(revision)]
 
            except IndexError:
 
                msg = ("Revision %s does not exist for %s" % (revision, self))
0 comments (0 inline, 0 general)