mercurial/httppeer.py

wireproto: add streams to frame-based protocol

Previously, the frame-based protocol was just a series of frames,
with each frame associated with a request ID.

In order to scale the protocol, we'll want to enable the use of
compression. While it is possible to enable compression at the
socket/pipe level, this has its disadvantages. The big one is it
undermines the point of frames being standalone, atomic units that
can be read and written: if you add compression above the framing
protocol, you are back to having a stream-based protocol as opposed
to something frame-based.

So in order to preserve frames, compression needs to occur at
the frame payload level.

Compressing each frame's payload individually will limit compression
ratios because the window size of the compressor will be limited
by the max frame size, which is 32-64kb as currently defined. It
will also add CPU overhead, as it is more efficient for compressors
to operate on fewer, larger blocks of data than more, smaller blocks.

So compressing each frame independently is out.

This means we need to compress each frame's payload as if it is part
of a larger stream.

The simplest approach is to have 1 stream per connection. This
could certainly work. However, it has disadvantages (documented below).

We could also have 1 stream per RPC/command invocation. (This is the
model HTTP/2 goes with.) This also has disadvantages.

The main disadvantage to one global stream is that it has the very
real potential to create CPU bottlenecks doing compression. Networks
are only getting faster and the performance of single CPU cores has
been relatively flat. Newer compression formats like zstandard offer
better CPU cycle efficiency than predecessors like zlib. But it still
all too common to saturate your CPU with compression overhead long
before you saturate the network pipe.

The main disadvantage with streams per request is that you can't
reap the benefits of the compression context for multiple requests.
For example, if you send 1000 RPC requests (or HTTP/2 requests for
that matter), the response to each would have its own compression
context. The overall size of the raw responses would be larger because
compression contexts wouldn't be able to reference data from another
request or response.

The approach for streams as implemented in this commit is to support
N streams per connection and for streams to potentially span requests
and responses. As explained by the added internals docs, this
facilitates servers and clients delegating independent streams and
compression to independent threads / CPU cores. This helps alleviate
the CPU bottleneck of compression. This design also allows compression
contexts to be reused across requests/responses. This can result in
improved compression ratios and less overhead for compressors and
decompressors having to build new contexts.

Another feature that was defined was the ability for individual frames
within a stream to declare whether that individual frame's payload
uses the content encoding (read: compression) defined by the stream.
The idea here is that some servers may serve data from a combination
of caches and dynamic resolution. Data coming from caches may be
pre-compressed. We want to facilitate servers being able to essentially
stream bytes from caches to the wire with minimal overhead. Being
able to mix and match with frames are compressed within a stream
enables these types of advanced server functionality.

This commit defines the new streams mechanism. Basic code for
supporting streams in frames has been added. But that code is
seriously lacking and doesn't fully conform to the defined protocol.
For example, we don't close any streams. And support for content
encoding within streams is not yet implemented. The change was
rather invasive and I didn't think it would be reasonable to implement
the entire feature in a single commit.

For the record, I would have loved to reuse an existing multiplexing
protocol to build the new wire protocol on top of. However, I couldn't
find a protocol that offers the performance and scaling characteristics
that I desired. Namely, it should support multiple compression
contexts to facilitate scaling out to multiple CPU cores and
compression contexts should be able to live longer than single RPC
requests. HTTP/2 *almost* fits the bill. But the semantics of HTTP
message exchange state that streams can only live for a single
request-response. We /could/ tunnel on top of HTTP/2 streams and
frames with HEADER and DATA frames. But there's no guarantee that
HTTP/2 libraries and proxies would allow us to use HTTP/2 streams
and frames without the HTTP message exchange semantics defined in
RFC 7540 Section 8. Other RPC protocols like gRPC tunnel are built
on top of HTTP/2 and thus preserve its semantics of stream per
RPC invocation. Even QUIC does this. We could attempt to invent a
higher-level stream that spans HTTP/2 streams. But this would be
violating HTTP/2 because there is no guarantee that HTTP/2 streams
are routed to the same server. The best we can do - which is what
this protocol does - is shoehorn all request and response data into
a single HTTP message and create streams within. At that point, we've
defined a Content-Type in HTTP parlance. It just so happens our
media type can also work as a standalone, stream-based protocol,
without leaning on HTTP or similar protocol.

Differential Revision: https://phab.mercurial-scm.org/D2907

# httppeer.py - HTTP repository proxy classes for mercurial
#
# Copyright 2005, 2006 Matt Mackall <[e-mail]>
# Copyright 2006 Vadim Gelfer <[e-mail]>
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.

from __future__ import absolute_import

import errno
import io
import os
import socket
import struct
import tempfile

from .i18n import _
from . import (
    bundle2,
    error,
    httpconnection,
    pycompat,
    statichttprepo,
    url as urlmod,
    util,
    wireproto,
)

httplib = util.httplib
urlerr = util.urlerr
urlreq = util.urlreq

def encodevalueinheaders(value, header, limit):
    """Encode a string value into multiple HTTP headers.

    ``value`` will be encoded into 1 or more HTTP headers with the names
    ``header-<N>`` where ``<N>`` is an integer starting at 1. Each header
    name + value will be at most ``limit`` bytes long.

    Returns an iterable of 2-tuples consisting of header names and
    values as native strings.
    """
    # HTTP Headers are ASCII. Python 3 requires them to be unicodes,
    # not bytes. This function always takes bytes in as arguments.
    fmt = pycompat.strurl(header) + r'-%s'
    # Note: it is *NOT* a bug that the last bit here is a bytestring
    # and not a unicode: we're just getting the encoded length anyway,
    # and using an r-string to make it portable between Python 2 and 3
    # doesn't work because then the \r is a literal backslash-r
    # instead of a carriage return.
    valuelen = limit - len(fmt % r'000') - len(': \r\n')
    result = []

    n = 0
    for i in xrange(0, len(value), valuelen):
        n += 1
        result.append((fmt % str(n), pycompat.strurl(value[i:i + valuelen])))

    return result

def _wraphttpresponse(resp):
    """Wrap an HTTPResponse with common error handlers.

    This ensures that any I/O from any consumer raises the appropriate
    error and messaging.
    """
    origread = resp.read

    class readerproxy(resp.__class__):
        def read(self, size=None):
            try:
                return origread(size)
            except httplib.IncompleteRead as e:
                # e.expected is an integer if length known or None otherwise.
                if e.expected:
                    msg = _('HTTP request error (incomplete response; '
                            'expected %d bytes got %d)') % (e.expected,
                                                           len(e.partial))
                else:
                    msg = _('HTTP request error (incomplete response)')

                raise error.PeerTransportError(
                    msg,
                    hint=_('this may be an intermittent network failure; '
                           'if the error persists, consider contacting the '
                           'network or server operator'))
            except httplib.HTTPException as e:
                raise error.PeerTransportError(
                    _('HTTP request error (%s)') % e,
                    hint=_('this may be an intermittent network failure; '
                           'if the error persists, consider contacting the '
                           'network or server operator'))

    resp.__class__ = readerproxy

class _multifile(object):
    def __init__(self, *fileobjs):
        for f in fileobjs:
            if not util.safehasattr(f, 'length'):
                raise ValueError(
                    '_multifile only supports file objects that '
                    'have a length but this one does not:', type(f), f)
        self._fileobjs = fileobjs
        self._index = 0

    @property
    def length(self):
        return sum(f.length for f in self._fileobjs)

    def read(self, amt=None):
        if amt <= 0:
            return ''.join(f.read() for f in self._fileobjs)
        parts = []
        while amt and self._index < len(self._fileobjs):
            parts.append(self._fileobjs[self._index].read(amt))
            got = len(parts[-1])
            if got < amt:
                self._index += 1
            amt -= got
        return ''.join(parts)

    def seek(self, offset, whence=os.SEEK_SET):
        if whence != os.SEEK_SET:
            raise NotImplementedError(
                '_multifile does not support anything other'
                ' than os.SEEK_SET for whence on seek()')
        if offset != 0:
            raise NotImplementedError(
                '_multifile only supports seeking to start, but that '
                'could be fixed if you need it')
        for f in self._fileobjs:
            f.seek(0)
        self._index = 0

class httppeer(wireproto.wirepeer):
    def __init__(self, ui, path, url, opener):
        self._ui = ui
        self._path = path
        self._url = url
        self._caps = None
        self._urlopener = opener
        # This is an its own attribute to facilitate extensions overriding
        # the default type.
        self._requestbuilder = urlreq.request

    def __del__(self):
        for h in self._urlopener.handlers:
            h.close()
            getattr(h, "close_all", lambda: None)()

    def _openurl(self, req):
        if (self._ui.debugflag
            and self._ui.configbool('devel', 'debug.peer-request')):
            dbg = self._ui.debug
            line = 'devel-peer-request: %s\n'
            dbg(line % '%s %s' % (req.get_method(), req.get_full_url()))
            hgargssize = None

            for header, value in sorted(req.header_items()):
                if header.startswith('X-hgarg-'):
                    if hgargssize is None:
                        hgargssize = 0
                    hgargssize += len(value)
                else:
                    dbg(line % '  %s %s' % (header, value))

            if hgargssize is not None:
                dbg(line % '  %d bytes of commands arguments in headers'
                    % hgargssize)

            if req.has_data():
                data = req.get_data()
                length = getattr(data, 'length', None)
                if length is None:
                    length = len(data)
                dbg(line % '  %d bytes of data' % length)

            start = util.timer()

        ret = self._urlopener.open(req)
        if self._ui.configbool('devel', 'debug.peer-request'):
            dbg(line % '  finished in %.4f seconds (%s)'
                % (util.timer() - start, ret.code))
        return ret

    # Begin of _basepeer interface.

    @util.propertycache
    def ui(self):
        return self._ui

    def url(self):
        return self._path

    def local(self):
        return None

    def peer(self):
        return self

    def canpush(self):
        return True

    def close(self):
        pass

    # End of _basepeer interface.

    # Begin of _basewirepeer interface.

    def capabilities(self):
        # self._fetchcaps() should have been called as part of peer
        # handshake. So self._caps should always be set.
        assert self._caps is not None
        return self._caps

    # End of _basewirepeer interface.

    # look up capabilities only when needed

    def _fetchcaps(self):
        self._caps = set(self._call('capabilities').split())

    def _callstream(self, cmd, _compressible=False, **args):
        args = pycompat.byteskwargs(args)
        if cmd == 'pushkey':
            args['data'] = ''
        data = args.pop('data', None)
        headers = args.pop('headers', {})

        self.ui.debug("sending %s command\n" % cmd)
        q = [('cmd', cmd)]
        headersize = 0
        varyheaders = []
        # Important: don't use self.capable() here or else you end up
        # with infinite recursion when trying to look up capabilities
        # for the first time.
        postargsok = self._caps is not None and 'httppostargs' in self._caps

        # Send arguments via POST.
        if postargsok and args:
            strargs = urlreq.urlencode(sorted(args.items()))
            if not data:
                data = strargs
            else:
                if isinstance(data, bytes):
                    i = io.BytesIO(data)
                    i.length = len(data)
                    data = i
                argsio = io.BytesIO(strargs)
                argsio.length = len(strargs)
                data = _multifile(argsio, data)
            headers[r'X-HgArgs-Post'] = len(strargs)
        elif args:
            # Calling self.capable() can infinite loop if we are calling
            # "capabilities". But that command should never accept wire
            # protocol arguments. So this should never happen.
            assert cmd != 'capabilities'
            httpheader = self.capable('httpheader')
            if httpheader:
                headersize = int(httpheader.split(',', 1)[0])

            # Send arguments via HTTP headers.
            if headersize > 0:
                # The headers can typically carry more data than the URL.
                encargs = urlreq.urlencode(sorted(args.items()))
                for header, value in encodevalueinheaders(encargs, 'X-HgArg',
                                                          headersize):
                    headers[header] = value
                    varyheaders.append(header)
            # Send arguments via query string (Mercurial <1.9).
            else:
                q += sorted(args.items())

        qs = '?%s' % urlreq.urlencode(q)
        cu = "%s%s" % (self._url, qs)
        size = 0
        if util.safehasattr(data, 'length'):
            size = data.length
        elif data is not None:
            size = len(data)
        if data is not None and r'Content-Type' not in headers:
            headers[r'Content-Type'] = r'application/mercurial-0.1'

        # Tell the server we accept application/mercurial-0.2 and multiple
        # compression formats if the server is capable of emitting those
        # payloads.
        protoparams = []

        mediatypes = set()
        if self._caps is not None:
            mt = self.capable('httpmediatype')
            if mt:
                protoparams.append('0.1')
                mediatypes = set(mt.split(','))

        if '0.2tx' in mediatypes:
            protoparams.append('0.2')

        if '0.2tx' in mediatypes and self.capable('compression'):
            # We /could/ compare supported compression formats and prune
            # non-mutually supported or error if nothing is mutually supported.
            # For now, send the full list to the server and have it error.
            comps = [e.wireprotosupport().name for e in
                     util.compengines.supportedwireengines(util.CLIENTROLE)]
            protoparams.append('comp=%s' % ','.join(comps))

        if protoparams:
            protoheaders = encodevalueinheaders(' '.join(protoparams),
                                                'X-HgProto',
                                                headersize or 1024)
            for header, value in protoheaders:
                headers[header] = value
                varyheaders.append(header)

        if varyheaders:
            headers[r'Vary'] = r','.join(varyheaders)

        req = self._requestbuilder(pycompat.strurl(cu), data, headers)

        if data is not None:
            self.ui.debug("sending %d bytes\n" % size)
            req.add_unredirected_header(r'Content-Length', r'%d' % size)
        try:
            resp = self._openurl(req)
        except urlerr.httperror as inst:
            if inst.code == 401:
                raise error.Abort(_('authorization failed'))
            raise
        except httplib.HTTPException as inst:
            self.ui.debug('http error while sending %s command\n' % cmd)
            self.ui.traceback()
            raise IOError(None, inst)

        # Insert error handlers for common I/O failures.
        _wraphttpresponse(resp)

        # record the url we got redirected to
        resp_url = pycompat.bytesurl(resp.geturl())
        if resp_url.endswith(qs):
            resp_url = resp_url[:-len(qs)]
        if self._url.rstrip('/') != resp_url.rstrip('/'):
            if not self.ui.quiet:
                self.ui.warn(_('real URL is %s\n') % resp_url)
        self._url = resp_url
        try:
            proto = pycompat.bytesurl(resp.getheader(r'content-type', r''))
        except AttributeError:
            proto = pycompat.bytesurl(resp.headers.get(r'content-type', r''))

        safeurl = util.hidepassword(self._url)
        if proto.startswith('application/hg-error'):
            raise error.OutOfBandError(resp.read())
        # accept old "text/plain" and "application/hg-changegroup" for now
        if not (proto.startswith('application/mercurial-') or
                (proto.startswith('text/plain')
                 and not resp.headers.get('content-length')) or
                proto.startswith('application/hg-changegroup')):
            self.ui.debug("requested URL: '%s'\n" % util.hidepassword(cu))
            raise error.RepoError(
                _("'%s' does not appear to be an hg repository:\n"
                  "---%%<--- (%s)\n%s\n---%%<---\n")
                % (safeurl, proto or 'no content-type', resp.read(1024)))

        if proto.startswith('application/mercurial-'):
            try:
                version = proto.split('-', 1)[1]
                version_info = tuple([int(n) for n in version.split('.')])
            except ValueError:
                raise error.RepoError(_("'%s' sent a broken Content-Type "
                                        "header (%s)") % (safeurl, proto))

            # TODO consider switching to a decompression reader that uses
            # generators.
            if version_info == (0, 1):
                if _compressible:
                    return util.compengines['zlib'].decompressorreader(resp)
                return resp
            elif version_info == (0, 2):
                # application/mercurial-0.2 always identifies the compression
                # engine in the payload header.
                elen = struct.unpack('B', resp.read(1))[0]
                ename = resp.read(elen)
                engine = util.compengines.forwiretype(ename)
                return engine.decompressorreader(resp)
            else:
                raise error.RepoError(_("'%s' uses newer protocol %s") %
                                      (safeurl, version))

        if _compressible:
            return util.compengines['zlib'].decompressorreader(resp)

        return resp

    def _call(self, cmd, **args):
        fp = self._callstream(cmd, **args)
        try:
            return fp.read()
        finally:
            # if using keepalive, allow connection to be reused
            fp.close()

    def _callpush(self, cmd, cg, **args):
        # have to stream bundle to a temp file because we do not have
        # http 1.1 chunked transfer.

        types = self.capable('unbundle')
        try:
            types = types.split(',')
        except AttributeError:
            # servers older than d1b16a746db6 will send 'unbundle' as a
            # boolean capability. They only support headerless/uncompressed
            # bundles.
            types = [""]
        for x in types:
            if x in bundle2.bundletypes:
                type = x
                break

        tempname = bundle2.writebundle(self.ui, cg, None, type)
        fp = httpconnection.httpsendfile(self.ui, tempname, "rb")
        headers = {r'Content-Type': r'application/mercurial-0.1'}

        try:
            r = self._call(cmd, data=fp, headers=headers, **args)
            vals = r.split('\n', 1)
            if len(vals) < 2:
                raise error.ResponseError(_("unexpected response:"), r)
            return vals
        except urlerr.httperror:
            # Catch and re-raise these so we don't try and treat them
            # like generic socket errors. They lack any values in
            # .args on Python 3 which breaks our socket.error block.
            raise
        except socket.error as err:
            if err.args[0] in (errno.ECONNRESET, errno.EPIPE):
                raise error.Abort(_('push failed: %s') % err.args[1])
            raise error.Abort(err.args[1])
        finally:
            fp.close()
            os.unlink(tempname)

    def _calltwowaystream(self, cmd, fp, **args):
        fh = None
        fp_ = None
        filename = None
        try:
            # dump bundle to disk
            fd, filename = tempfile.mkstemp(prefix="hg-bundle-", suffix=".hg")
            fh = os.fdopen(fd, r"wb")
            d = fp.read(4096)
            while d:
                fh.write(d)
                d = fp.read(4096)
            fh.close()
            # start http push
            fp_ = httpconnection.httpsendfile(self.ui, filename, "rb")
            headers = {r'Content-Type': r'application/mercurial-0.1'}
            return self._callstream(cmd, data=fp_, headers=headers, **args)
        finally:
            if fp_ is not None:
                fp_.close()
            if fh is not None:
                fh.close()
                os.unlink(filename)

    def _callcompressable(self, cmd, **args):
        return self._callstream(cmd, _compressible=True, **args)

    def _abort(self, exception):
        raise exception

def makepeer(ui, path):
    u = util.url(path)
    if u.query or u.fragment:
        raise error.Abort(_('unsupported URL component: "%s"') %
                          (u.query or u.fragment))

    # urllib cannot handle URLs with embedded user or passwd.
    url, authinfo = u.authinfo()
    ui.debug('using %s\n' % url)

    opener = urlmod.opener(ui, authinfo)

    return httppeer(ui, path, url, opener)

def instance(ui, path, create):
    if create:
        raise error.Abort(_('cannot create new http repository'))
    try:
        if path.startswith('https:') and not urlmod.has_https:
            raise error.Abort(_('Python support for SSL and HTTPS '
                                'is not installed'))

        inst = makepeer(ui, path)
        inst._fetchcaps()

        return inst
    except error.RepoError as httpexception:
        try:
            r = statichttprepo.instance(ui, "static-" + path, create)
            ui.note(_('(falling back to static-http)\n'))
            return r
        except error.RepoError:
            raise httpexception # use the original http RepoError instead