Mercurial > hg
view mercurial/httppeer.py @ 30442:41a8106789ca
util: implement zstd compression engine
Now that zstd is vendored and being built (in some configurations), we
can implement a compression engine for zstd!
The zstd engine is a little different from existing engines. Because
it may not always be present, we have to defer load the module in case
importing it fails. We facilitate this via a cached property that holds
a reference to the module or None. The "available" method is
implemented to reflect reality.
The zstd engine declares its ability to handle bundles using the
"zstd" human name and the "ZS" internal name. The latter was chosen
because internal names are 2 characters (by only convention I think)
and "ZS" seems reasonable.
The engine, like others, supports specifying the compression level.
However, there are no consumers of this API that yet pass in that
argument. I have plans to change that, so stay tuned.
Since all we need to do to support bundle generation with a new
compression engine is implement and register the compression engine,
bundle generation with zstd "just works!" Tests demonstrating this
have been added.
How does performance of zstd for bundle generation compare? On the
mozilla-unified repo, `hg bundle --all -t <engine>-v2` yields the
following on my i7-6700K on Linux:
engine CPU time bundle size vs orig size throughput
none 97.0s 4,054,405,584 100.0% 41.8 MB/s
bzip2 (l=9) 393.6s 975,343,098 24.0% 10.3 MB/s
gzip (l=6) 184.0s 1,140,533,074 28.1% 22.0 MB/s
zstd (l=1) 108.2s 1,119,434,718 27.6% 37.5 MB/s
zstd (l=2) 111.3s 1,078,328,002 26.6% 36.4 MB/s
zstd (l=3) 113.7s 1,011,823,727 25.0% 35.7 MB/s
zstd (l=4) 116.0s 1,008,965,888 24.9% 35.0 MB/s
zstd (l=5) 121.0s 977,203,148 24.1% 33.5 MB/s
zstd (l=6) 131.7s 927,360,198 22.9% 30.8 MB/s
zstd (l=7) 139.0s 912,808,505 22.5% 29.2 MB/s
zstd (l=12) 198.1s 854,527,714 21.1% 20.5 MB/s
zstd (l=18) 681.6s 789,750,690 19.5% 5.9 MB/s
On compression, zstd for bundle generation delivers:
* better compression than gzip with significantly less CPU utilization
* better than bzip2 compression ratios while still being significantly
faster than gzip
* ability to aggressively tune compression level to achieve
significantly smaller bundles
That last point is important. With clone bundles, a server can
pre-generate a bundle file, upload it to a static file server, and
redirect clients to transparently download it during clone. The server
could choose to produce a zstd bundle with the highest compression
settings possible. This would take a very long time - a magnitude
longer than a typical zstd bundle generation - but the result would
be hundreds of megabytes smaller! For the clone volume we do at
Mozilla, this could translate to petabytes of bandwidth savings
per year and faster clones (due to smaller transfer size).
I don't have detailed numbers to report on decompression. However,
zstd decompression is fast: >1 GB/s output throughput on this machine,
even through the Python bindings. And it can do that regardless of the
compression level of the input. By the time you have enough data to
worry about overhead of decompression, you have plenty of other things
to worry about performance wise.
zstd is wins all around. I can't wait to implement support for it
on the wire protocol and in revlogs.
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Fri, 11 Nov 2016 01:10:07 -0800 |
parents | cac4ca036dff |
children | e16e234b9ca3 |
line wrap: on
line source
# httppeer.py - HTTP repository proxy classes for mercurial # # Copyright 2005, 2006 Matt Mackall <mpm@selenic.com> # Copyright 2006 Vadim Gelfer <vadim.gelfer@gmail.com> # # 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 os import socket import tempfile import zlib from .i18n import _ from .node import nullid from . import ( bundle2, error, httpconnection, statichttprepo, url, util, wireproto, ) httplib = util.httplib urlerr = util.urlerr urlreq = util.urlreq def zgenerator(f): zd = zlib.decompressobj() try: for chunk in util.filechunkiter(f): while chunk: yield zd.decompress(chunk, 2**18) chunk = zd.unconsumed_tail except httplib.HTTPException: raise IOError(None, _('connection ended unexpectedly')) yield zd.flush() class httppeer(wireproto.wirepeer): def __init__(self, ui, path): self.path = path self.caps = None self.handler = None self.urlopener = None self.requestbuilder = None 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 self._url, authinfo = u.authinfo() self.ui = ui self.ui.debug('using %s\n' % self._url) self.urlopener = url.opener(ui, authinfo) self.requestbuilder = urlreq.request def __del__(self): urlopener = getattr(self, 'urlopener', None) if urlopener: for h in urlopener.handlers: h.close() getattr(h, "close_all", lambda : None)() def url(self): return self.path # look up capabilities only when needed def _fetchcaps(self): self.caps = set(self._call('capabilities').split()) def _capabilities(self): if self.caps is None: try: self._fetchcaps() except error.RepoError: self.caps = set() self.ui.debug('capabilities: %s\n' % (' '.join(self.caps or ['none']))) return self.caps def lock(self): raise error.Abort(_('operation not supported over http')) def _callstream(self, cmd, **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 # 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 # TODO: support for httppostargs when data is a file-like # object rather than a basestring canmungedata = not data or isinstance(data, basestring) if postargsok and canmungedata: strargs = urlreq.urlencode(sorted(args.items())) if strargs: if not data: data = strargs elif isinstance(data, basestring): data = strargs + data headers['X-HgArgs-Post'] = len(strargs) else: if len(args) > 0: httpheader = self.capable('httpheader') if httpheader: headersize = int(httpheader.split(',', 1)[0]) if headersize > 0: # The headers can typically carry more data than the URL. encargs = urlreq.urlencode(sorted(args.items())) headerfmt = 'X-HgArg-%s' contentlen = headersize - len(headerfmt % '000' + ': \r\n') headernum = 0 varyheaders = [] for i in xrange(0, len(encargs), contentlen): headernum += 1 header = headerfmt % str(headernum) headers[header] = encargs[i:i + contentlen] varyheaders.append(header) headers['Vary'] = ','.join(varyheaders) 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 size and self.ui.configbool('ui', 'usehttp2', False): headers['Expect'] = '100-Continue' headers['X-HgHttp2'] = '1' if data is not None and 'Content-Type' not in headers: headers['Content-Type'] = 'application/mercurial-0.1' req = self.requestbuilder(cu, data, headers) if data is not None: self.ui.debug("sending %s bytes\n" % size) req.add_unredirected_header('Content-Length', '%d' % size) try: resp = self.urlopener.open(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) except IndexError: # this only happens with Python 2.3, later versions raise URLError raise error.Abort(_('http error, possibly caused by proxy setting')) # record the url we got redirected to resp_url = 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 = resp.getheader('content-type') except AttributeError: proto = resp.headers.get('content-type', '') 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)) if version_info > (0, 1): raise error.RepoError(_("'%s' uses newer protocol %s") % (safeurl, version)) 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 = {'Content-Type': '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 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, "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 = {'Content-Type': '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): stream = self._callstream(cmd, **args) return util.chunkbuffer(zgenerator(stream)) def _abort(self, exception): raise exception class httpspeer(httppeer): def __init__(self, ui, path): if not url.has_https: raise error.Abort(_('Python support for SSL and HTTPS ' 'is not installed')) httppeer.__init__(self, ui, path) def instance(ui, path, create): if create: raise error.Abort(_('cannot create new http repository')) try: if path.startswith('https:'): inst = httpspeer(ui, path) else: inst = httppeer(ui, path) try: # Try to do useful work when checking compatibility. # Usually saves a roundtrip since we want the caps anyway. inst._fetchcaps() except error.RepoError: # No luck, try older compatibility check. inst.between([(nullid, nullid)]) 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