annotate mercurial/httpconnection.py @ 26623:5a95fe44121d

clonebundles: support for seeding clones from pre-generated bundles

Cloning can be an expensive operation for servers because the server
generates a bundle from existing repository data at request time. For
a large repository like mozilla-central, this consumes 4+ minutes
of CPU time on the server. It also results in significant network
utilization. Multiplied by hundreds or even thousands of clients and
the ensuing load can result in difficulties scaling the Mercurial server.

Despite generation of bundles being deterministic until the next
changeset is added, the generation of bundles to service a clone request
is not cached. Each clone thus performs redundant work. This is
wasteful.

This patch introduces the "clonebundles" extension and related
client-side functionality to help alleviate this deficiency. The
client-side feature is behind an experimental flag and is not enabled by
default.

It works as follows:

1) Server operator generates a bundle and makes it available on a
   server (likely HTTP).
2) Server operator defines the URL of a bundle file in a
   .hg/clonebundles.manifest file.
3) Client `hg clone`ing sees the server is advertising bundle URLs.
4) Client fetches and applies the advertised bundle.
5) Client performs equivalent of `hg pull` to fetch changes made since
   the bundle was created.

Essentially, the server performs the expensive work of generating a
bundle once and all subsequent clones fetch a static file from
somewhere. Scaling static file serving is a much more manageable
problem than scaling a Python application like Mercurial. Assuming your
repository grows less than 1% per day, the end result is 99+% of CPU
and network load from clones is eliminated, allowing Mercurial servers
to scale more easily. Serving static files also means data can be
transferred to clients as fast as they can consume it, rather than as
fast as servers can generate it. This makes clones faster.

Mozilla has implemented similar functionality of this patch on
hg.mozilla.org using a custom extension. We are hosting bundle files in
Amazon S3 and CloudFront (a CDN) and have successfully offloaded
>1 TB/day in data transfer from hg.mozilla.org, freeing up significant
bandwidth and CPU resources. The positive impact has been stellar and
I believe it has proved its value to be included in Mercurial core. I
feel it is important for the client-side support to be enabled in core
by default because it means that clients will get faster, more reliable
clones and will enable server operators to reduce load without
requiring any client-side configuration changes (assuming clients are
up to date, of course).

The scope of this feature is narrowly and specifically tailored to
cloning, despite "serve pulls from pre-generated bundles" being a valid
and useful feature. I would eventually like for Mercurial servers to
support transferring *all* repository data via statically hosted files.
You could imagine a server that siphons all pushed data to bundle files
and instructs clients to apply a stream of bundles to reconstruct all
repository data. This feature, while useful and powerful, is
significantly more work to implement because it requires the server
component have awareness of discovery and a mapping of which changesets
are in which files. Full, clone bundles, by contrast, are much simpler.

The wire protocol command is named "clonebundles" instead of something
more generic like "staticbundles" to leave the door open for a new, more
powerful and more generic server-side component with minimal backwards
compatibility implications. The name "bundleclone" is used by Mozilla's
extension and would cause problems since there are subtle differences
in Mozilla's extension.

Mozilla's experience with this idea has taught us that some form of
"content negotiation" is required. Not all clients will support all
bundle formats or even URLs (advanced TLS requirements, etc). To ensure
the highest uptake possible, a server needs to advertise multiple
versions of bundles and clients need to be able to choose the most
appropriate from that list one. The "attributes" in each
server-advertised entry facilitate this filtering and sorting. Their
use will become apparent in subsequent patches.

Initial inspiration and credit for the idea of cloning from static files
belongs to Augie Fackler and his "lookaside clone" extension proof of
concept.

# httpconnection.py - urllib2 handler for new http support
#
# Copyright 2005, 2006, 2007, 2008 Matt Mackall <[e-mail]>
# Copyright 2006, 2007 Alexis S. L. Carvalho <[e-mail]>
# Copyright 2006 Vadim Gelfer <[e-mail]>
# Copyright 2011 Google, Inc.
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.
import logging
import socket
import urllib
import urllib2
import os

from mercurial import httpclient
from mercurial import sslutil
from mercurial import util
from mercurial.i18n import _

# moved here from url.py to avoid a cycle
class httpsendfile(object):
    """This is a wrapper around the objects returned by python's "open".

    Its purpose is to send file-like objects via HTTP.
    It do however not define a __len__ attribute because the length
    might be more than Py_ssize_t can handle.
    """

    def __init__(self, ui, *args, **kwargs):
        # We can't just "self._data = open(*args, **kwargs)" here because there
        # is an "open" function defined in this module that shadows the global
        # one
        self.ui = ui
        self._data = open(*args, **kwargs)
        self.seek = self._data.seek
        self.close = self._data.close
        self.write = self._data.write
        self.length = os.fstat(self._data.fileno()).st_size
        self._pos = 0
        self._total = self.length // 1024 * 2

    def read(self, *args, **kwargs):
        try:
            ret = self._data.read(*args, **kwargs)
        except EOFError:
            self.ui.progress(_('sending'), None)
        self._pos += len(ret)
        # We pass double the max for total because we currently have
        # to send the bundle twice in the case of a server that
        # requires authentication. Since we can't know until we try
        # once whether authentication will be required, just lie to
        # the user and maybe the push succeeds suddenly at 50%.
        self.ui.progress(_('sending'), self._pos // 1024,
                         unit=_('kb'), total=self._total)
        return ret

# moved here from url.py to avoid a cycle
def readauthforuri(ui, uri, user):
    # Read configuration
    config = dict()
    for key, val in ui.configitems('auth'):
        if '.' not in key:
            ui.warn(_("ignoring invalid [auth] key '%s'\n") % key)
            continue
        group, setting = key.rsplit('.', 1)
        gdict = config.setdefault(group, dict())
        if setting in ('username', 'cert', 'key'):
            val = util.expandpath(val)
        gdict[setting] = val

    # Find the best match
    scheme, hostpath = uri.split('://', 1)
    bestuser = None
    bestlen = 0
    bestauth = None
    for group, auth in config.iteritems():
        if user and user != auth.get('username', user):
            # If a username was set in the URI, the entry username
            # must either match it or be unset
            continue
        prefix = auth.get('prefix')
        if not prefix:
            continue
        p = prefix.split('://', 1)
        if len(p) > 1:
            schemes, prefix = [p[0]], p[1]
        else:
            schemes = (auth.get('schemes') or 'https').split()
        if (prefix == '*' or hostpath.startswith(prefix)) and \
            (len(prefix) > bestlen or (len(prefix) == bestlen and \
                not bestuser and 'username' in auth)) \
             and scheme in schemes:
            bestlen = len(prefix)
            bestauth = group, auth
            bestuser = auth.get('username')
            if user and not bestuser:
                auth['username'] = user
    return bestauth

# Mercurial (at least until we can remove the old codepath) requires
# that the http response object be sufficiently file-like, so we
# provide a close() method here.
class HTTPResponse(httpclient.HTTPResponse):
    def close(self):
        pass

class HTTPConnection(httpclient.HTTPConnection):
    response_class = HTTPResponse
    def request(self, method, uri, body=None, headers=None):
        if headers is None:
            headers = {}
        if isinstance(body, httpsendfile):
            body.seek(0)
        httpclient.HTTPConnection.request(self, method, uri, body=body,
                                          headers=headers)


_configuredlogging = False
LOGFMT = '%(levelname)s:%(name)s:%(lineno)d:%(message)s'
# Subclass BOTH of these because otherwise urllib2 "helpfully"
# reinserts them since it notices we don't include any subclasses of
# them.
class http2handler(urllib2.HTTPHandler, urllib2.HTTPSHandler):
    def __init__(self, ui, pwmgr):
        global _configuredlogging
        urllib2.AbstractHTTPHandler.__init__(self)
        self.ui = ui
        self.pwmgr = pwmgr
        self._connections = {}
        # developer config: ui.http2debuglevel
        loglevel = ui.config('ui', 'http2debuglevel', default=None)
        if loglevel and not _configuredlogging:
            _configuredlogging = True
            logger = logging.getLogger('mercurial.httpclient')
            logger.setLevel(getattr(logging, loglevel.upper()))
            handler = logging.StreamHandler()
            handler.setFormatter(logging.Formatter(LOGFMT))
            logger.addHandler(handler)

    def close_all(self):
        """Close and remove all connection objects being kept for reuse."""
        for openconns in self._connections.values():
            for conn in openconns:
                conn.close()
        self._connections = {}

    # shamelessly borrowed from urllib2.AbstractHTTPHandler
    def do_open(self, http_class, req, use_ssl):
        """Return an addinfourl object for the request, using http_class.

        http_class must implement the HTTPConnection API from httplib.
        The addinfourl return value is a file-like object.  It also
        has methods and attributes including:
            - info(): return a mimetools.Message object for the headers
            - geturl(): return the original request URL
            - code: HTTP status code
        """
        # If using a proxy, the host returned by get_host() is
        # actually the proxy. On Python 2.6.1, the real destination
        # hostname is encoded in the URI in the urllib2 request
        # object. On Python 2.6.5, it's stored in the _tunnel_host
        # attribute which has no accessor.
        tunhost = getattr(req, '_tunnel_host', None)
        host = req.get_host()
        if tunhost:
            proxyhost = host
            host = tunhost
        elif req.has_proxy():
            proxyhost = req.get_host()
            host = req.get_selector().split('://', 1)[1].split('/', 1)[0]
        else:
            proxyhost = None

        if proxyhost:
            if ':' in proxyhost:
                # Note: this means we'll explode if we try and use an
                # IPv6 http proxy. This isn't a regression, so we
                # won't worry about it for now.
                proxyhost, proxyport = proxyhost.rsplit(':', 1)
            else:
                proxyport = 3128 # squid default
            proxy = (proxyhost, proxyport)
        else:
            proxy = None

        if not host:
            raise urllib2.URLError('no host given')

        connkey = use_ssl, host, proxy
        allconns = self._connections.get(connkey, [])
        conns = [c for c in allconns if not c.busy()]
        if conns:
            h = conns[0]
        else:
            if allconns:
                self.ui.debug('all connections for %s busy, making a new '
                              'one\n' % host)
            timeout = None
            if req.timeout is not socket._GLOBAL_DEFAULT_TIMEOUT:
                timeout = req.timeout
            h = http_class(host, timeout=timeout, proxy_hostport=proxy)
            self._connections.setdefault(connkey, []).append(h)

        headers = dict(req.headers)
        headers.update(req.unredirected_hdrs)
        headers = dict(
            (name.title(), val) for name, val in headers.items())
        try:
            path = req.get_selector()
            if '://' in path:
                path = path.split('://', 1)[1].split('/', 1)[1]
            if path[0] != '/':
                path = '/' + path
            h.request(req.get_method(), path, req.data, headers)
            r = h.getresponse()
        except socket.error as err: # XXX what error?
            raise urllib2.URLError(err)

        # Pick apart the HTTPResponse object to get the addinfourl
        # object initialized properly.
        r.recv = r.read

        resp = urllib.addinfourl(r, r.headers, req.get_full_url())
        resp.code = r.status
        resp.msg = r.reason
        return resp

    # httplib always uses the given host/port as the socket connect
    # target, and then allows full URIs in the request path, which it
    # then observes and treats as a signal to do proxying instead.
    def http_open(self, req):
        if req.get_full_url().startswith('https'):
            return self.https_open(req)
        def makehttpcon(*args, **kwargs):
            k2 = dict(kwargs)
            k2['use_ssl'] = False
            return HTTPConnection(*args, **k2)
        return self.do_open(makehttpcon, req, False)

    def https_open(self, req):
        # req.get_full_url() does not contain credentials and we may
        # need them to match the certificates.
        url = req.get_full_url()
        user, password = self.pwmgr.find_stored_password(url)
        res = readauthforuri(self.ui, url, user)
        if res:
            group, auth = res
            self.auth = auth
            self.ui.debug("using auth.%s.* for authentication\n" % group)
        else:
            self.auth = None
        return self.do_open(self._makesslconnection, req, True)

    def _makesslconnection(self, host, port=443, *args, **kwargs):
        keyfile = None
        certfile = None

        if args: # key_file
            keyfile = args.pop(0)
        if args: # cert_file
            certfile = args.pop(0)

        # if the user has specified different key/cert files in
        # hgrc, we prefer these
        if self.auth and 'key' in self.auth and 'cert' in self.auth:
            keyfile = self.auth['key']
            certfile = self.auth['cert']

        # let host port take precedence
        if ':' in host and '[' not in host or ']:' in host:
            host, port = host.rsplit(':', 1)
            port = int(port)
            if '[' in host:
                host = host[1:-1]

        kwargs['keyfile'] = keyfile
        kwargs['certfile'] = certfile

        kwargs.update(sslutil.sslkwargs(self.ui, host))

        con = HTTPConnection(host, port, use_ssl=True,
                             ssl_wrap_socket=sslutil.wrapsocket,
                             ssl_validator=sslutil.validator(self.ui, host),
                             **kwargs)
        return con