##// END OF EJS Templates
wireproto: add streams to frame-based protocol...
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

File last commit:

r37102:f0b6fbea default
r37304:9bfcbe4f default
Show More
repair.py
447 lines | 15.3 KiB | text/x-python | PythonLexer
# repair.py - functions for repository repair for mercurial
#
# Copyright 2005, 2006 Chris Mason <mason@suse.com>
# Copyright 2007 Matt Mackall
#
# 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 hashlib
from .i18n import _
from .node import (
hex,
short,
)
from . import (
bundle2,
changegroup,
discovery,
error,
exchange,
obsolete,
obsutil,
util,
)
from .utils import (
stringutil,
)
def backupbundle(repo, bases, heads, node, suffix, compress=True,
obsolescence=True):
"""create a bundle with the specified revisions as a backup"""
backupdir = "strip-backup"
vfs = repo.vfs
if not vfs.isdir(backupdir):
vfs.mkdir(backupdir)
# Include a hash of all the nodes in the filename for uniqueness
allcommits = repo.set('%ln::%ln', bases, heads)
allhashes = sorted(c.hex() for c in allcommits)
totalhash = hashlib.sha1(''.join(allhashes)).digest()
name = "%s/%s-%s-%s.hg" % (backupdir, short(node),
hex(totalhash[:4]), suffix)
cgversion = changegroup.localversion(repo)
comp = None
if cgversion != '01':
bundletype = "HG20"
if compress:
comp = 'BZ'
elif compress:
bundletype = "HG10BZ"
else:
bundletype = "HG10UN"
outgoing = discovery.outgoing(repo, missingroots=bases, missingheads=heads)
contentopts = {
'cg.version': cgversion,
'obsolescence': obsolescence,
'phases': True,
}
return bundle2.writenewbundle(repo.ui, repo, 'strip', name, bundletype,
outgoing, contentopts, vfs, compression=comp)
def _collectfiles(repo, striprev):
"""find out the filelogs affected by the strip"""
files = set()
for x in xrange(striprev, len(repo)):
files.update(repo[x].files())
return sorted(files)
def _collectrevlog(revlog, striprev):
_, brokenset = revlog.getstrippoint(striprev)
return [revlog.linkrev(r) for r in brokenset]
def _collectmanifest(repo, striprev):
return _collectrevlog(repo.manifestlog._revlog, striprev)
def _collectbrokencsets(repo, files, striprev):
"""return the changesets which will be broken by the truncation"""
s = set()
s.update(_collectmanifest(repo, striprev))
for fname in files:
s.update(_collectrevlog(repo.file(fname), striprev))
return s
def strip(ui, repo, nodelist, backup=True, topic='backup'):
# This function requires the caller to lock the repo, but it operates
# within a transaction of its own, and thus requires there to be no current
# transaction when it is called.
if repo.currenttransaction() is not None:
raise error.ProgrammingError('cannot strip from inside a transaction')
# Simple way to maintain backwards compatibility for this
# argument.
if backup in ['none', 'strip']:
backup = False
repo = repo.unfiltered()
repo.destroying()
cl = repo.changelog
# TODO handle undo of merge sets
if isinstance(nodelist, str):
nodelist = [nodelist]
striplist = [cl.rev(node) for node in nodelist]
striprev = min(striplist)
files = _collectfiles(repo, striprev)
saverevs = _collectbrokencsets(repo, files, striprev)
# Some revisions with rev > striprev may not be descendants of striprev.
# We have to find these revisions and put them in a bundle, so that
# we can restore them after the truncations.
# To create the bundle we use repo.changegroupsubset which requires
# the list of heads and bases of the set of interesting revisions.
# (head = revision in the set that has no descendant in the set;
# base = revision in the set that has no ancestor in the set)
tostrip = set(striplist)
saveheads = set(saverevs)
for r in cl.revs(start=striprev + 1):
if any(p in tostrip for p in cl.parentrevs(r)):
tostrip.add(r)
if r not in tostrip:
saverevs.add(r)
saveheads.difference_update(cl.parentrevs(r))
saveheads.add(r)
saveheads = [cl.node(r) for r in saveheads]
# compute base nodes
if saverevs:
descendants = set(cl.descendants(saverevs))
saverevs.difference_update(descendants)
savebases = [cl.node(r) for r in saverevs]
stripbases = [cl.node(r) for r in tostrip]
stripobsidx = obsmarkers = ()
if repo.ui.configbool('devel', 'strip-obsmarkers'):
obsmarkers = obsutil.exclusivemarkers(repo, stripbases)
if obsmarkers:
stripobsidx = [i for i, m in enumerate(repo.obsstore)
if m in obsmarkers]
# For a set s, max(parents(s) - s) is the same as max(heads(::s - s)), but
# is much faster
newbmtarget = repo.revs('max(parents(%ld) - (%ld))', tostrip, tostrip)
if newbmtarget:
newbmtarget = repo[newbmtarget.first()].node()
else:
newbmtarget = '.'
bm = repo._bookmarks
updatebm = []
for m in bm:
rev = repo[bm[m]].rev()
if rev in tostrip:
updatebm.append(m)
# create a changegroup for all the branches we need to keep
backupfile = None
vfs = repo.vfs
node = nodelist[-1]
if backup:
backupfile = backupbundle(repo, stripbases, cl.heads(), node, topic)
repo.ui.status(_("saved backup bundle to %s\n") %
vfs.join(backupfile))
repo.ui.log("backupbundle", "saved backup bundle to %s\n",
vfs.join(backupfile))
tmpbundlefile = None
if saveheads:
# do not compress temporary bundle if we remove it from disk later
#
# We do not include obsolescence, it might re-introduce prune markers
# we are trying to strip. This is harmless since the stripped markers
# are already backed up and we did not touched the markers for the
# saved changesets.
tmpbundlefile = backupbundle(repo, savebases, saveheads, node, 'temp',
compress=False, obsolescence=False)
try:
with repo.transaction("strip") as tr:
offset = len(tr.entries)
tr.startgroup()
cl.strip(striprev, tr)
stripmanifest(repo, striprev, tr, files)
for fn in files:
repo.file(fn).strip(striprev, tr)
tr.endgroup()
for i in xrange(offset, len(tr.entries)):
file, troffset, ignore = tr.entries[i]
with repo.svfs(file, 'a', checkambig=True) as fp:
fp.truncate(troffset)
if troffset == 0:
repo.store.markremoved(file)
deleteobsmarkers(repo.obsstore, stripobsidx)
del repo.obsstore
repo.invalidatevolatilesets()
repo._phasecache.filterunknown(repo)
if tmpbundlefile:
ui.note(_("adding branch\n"))
f = vfs.open(tmpbundlefile, "rb")
gen = exchange.readbundle(ui, f, tmpbundlefile, vfs)
if not repo.ui.verbose:
# silence internal shuffling chatter
repo.ui.pushbuffer()
tmpbundleurl = 'bundle:' + vfs.join(tmpbundlefile)
txnname = 'strip'
if not isinstance(gen, bundle2.unbundle20):
txnname = "strip\n%s" % util.hidepassword(tmpbundleurl)
with repo.transaction(txnname) as tr:
bundle2.applybundle(repo, gen, tr, source='strip',
url=tmpbundleurl)
if not repo.ui.verbose:
repo.ui.popbuffer()
f.close()
with repo.transaction('repair') as tr:
bmchanges = [(m, repo[newbmtarget].node()) for m in updatebm]
bm.applychanges(repo, tr, bmchanges)
# remove undo files
for undovfs, undofile in repo.undofiles():
try:
undovfs.unlink(undofile)
except OSError as e:
if e.errno != errno.ENOENT:
ui.warn(_('error removing %s: %s\n') %
(undovfs.join(undofile),
stringutil.forcebytestr(e)))
except: # re-raises
if backupfile:
ui.warn(_("strip failed, backup bundle stored in '%s'\n")
% vfs.join(backupfile))
if tmpbundlefile:
ui.warn(_("strip failed, unrecovered changes stored in '%s'\n")
% vfs.join(tmpbundlefile))
ui.warn(_("(fix the problem, then recover the changesets with "
"\"hg unbundle '%s'\")\n") % vfs.join(tmpbundlefile))
raise
else:
if tmpbundlefile:
# Remove temporary bundle only if there were no exceptions
vfs.unlink(tmpbundlefile)
repo.destroyed()
# return the backup file path (or None if 'backup' was False) so
# extensions can use it
return backupfile
def safestriproots(ui, repo, nodes):
"""return list of roots of nodes where descendants are covered by nodes"""
torev = repo.unfiltered().changelog.rev
revs = set(torev(n) for n in nodes)
# tostrip = wanted - unsafe = wanted - ancestors(orphaned)
# orphaned = affected - wanted
# affected = descendants(roots(wanted))
# wanted = revs
tostrip = set(repo.revs('%ld-(::((roots(%ld)::)-%ld))', revs, revs, revs))
notstrip = revs - tostrip
if notstrip:
nodestr = ', '.join(sorted(short(repo[n].node()) for n in notstrip))
ui.warn(_('warning: orphaned descendants detected, '
'not stripping %s\n') % nodestr)
return [c.node() for c in repo.set('roots(%ld)', tostrip)]
class stripcallback(object):
"""used as a transaction postclose callback"""
def __init__(self, ui, repo, backup, topic):
self.ui = ui
self.repo = repo
self.backup = backup
self.topic = topic or 'backup'
self.nodelist = []
def addnodes(self, nodes):
self.nodelist.extend(nodes)
def __call__(self, tr):
roots = safestriproots(self.ui, self.repo, self.nodelist)
if roots:
strip(self.ui, self.repo, roots, self.backup, self.topic)
def delayedstrip(ui, repo, nodelist, topic=None):
"""like strip, but works inside transaction and won't strip irreverent revs
nodelist must explicitly contain all descendants. Otherwise a warning will
be printed that some nodes are not stripped.
Always do a backup. The last non-None "topic" will be used as the backup
topic name. The default backup topic name is "backup".
"""
tr = repo.currenttransaction()
if not tr:
nodes = safestriproots(ui, repo, nodelist)
return strip(ui, repo, nodes, True, topic)
# transaction postclose callbacks are called in alphabet order.
# use '\xff' as prefix so we are likely to be called last.
callback = tr.getpostclose('\xffstrip')
if callback is None:
callback = stripcallback(ui, repo, True, topic)
tr.addpostclose('\xffstrip', callback)
if topic:
callback.topic = topic
callback.addnodes(nodelist)
def stripmanifest(repo, striprev, tr, files):
revlog = repo.manifestlog._revlog
revlog.strip(striprev, tr)
striptrees(repo, tr, striprev, files)
def striptrees(repo, tr, striprev, files):
if 'treemanifest' in repo.requirements: # safe but unnecessary
# otherwise
for unencoded, encoded, size in repo.store.datafiles():
if (unencoded.startswith('meta/') and
unencoded.endswith('00manifest.i')):
dir = unencoded[5:-12]
repo.manifestlog._revlog.dirlog(dir).strip(striprev, tr)
def rebuildfncache(ui, repo):
"""Rebuilds the fncache file from repo history.
Missing entries will be added. Extra entries will be removed.
"""
repo = repo.unfiltered()
if 'fncache' not in repo.requirements:
ui.warn(_('(not rebuilding fncache because repository does not '
'support fncache)\n'))
return
with repo.lock():
fnc = repo.store.fncache
# Trigger load of fncache.
if 'irrelevant' in fnc:
pass
oldentries = set(fnc.entries)
newentries = set()
seenfiles = set()
repolen = len(repo)
for rev in repo:
ui.progress(_('rebuilding'), rev, total=repolen,
unit=_('changesets'))
ctx = repo[rev]
for f in ctx.files():
# This is to minimize I/O.
if f in seenfiles:
continue
seenfiles.add(f)
i = 'data/%s.i' % f
d = 'data/%s.d' % f
if repo.store._exists(i):
newentries.add(i)
if repo.store._exists(d):
newentries.add(d)
ui.progress(_('rebuilding'), None)
if 'treemanifest' in repo.requirements: # safe but unnecessary otherwise
for dir in util.dirs(seenfiles):
i = 'meta/%s/00manifest.i' % dir
d = 'meta/%s/00manifest.d' % dir
if repo.store._exists(i):
newentries.add(i)
if repo.store._exists(d):
newentries.add(d)
addcount = len(newentries - oldentries)
removecount = len(oldentries - newentries)
for p in sorted(oldentries - newentries):
ui.write(_('removing %s\n') % p)
for p in sorted(newentries - oldentries):
ui.write(_('adding %s\n') % p)
if addcount or removecount:
ui.write(_('%d items added, %d removed from fncache\n') %
(addcount, removecount))
fnc.entries = newentries
fnc._dirty = True
with repo.transaction('fncache') as tr:
fnc.write(tr)
else:
ui.write(_('fncache already up to date\n'))
def stripbmrevset(repo, mark):
"""
The revset to strip when strip is called with -B mark
Needs to live here so extensions can use it and wrap it even when strip is
not enabled or not present on a box.
"""
return repo.revs("ancestors(bookmark(%s)) - "
"ancestors(head() and not bookmark(%s)) - "
"ancestors(bookmark() and not bookmark(%s))",
mark, mark, mark)
def deleteobsmarkers(obsstore, indices):
"""Delete some obsmarkers from obsstore and return how many were deleted
'indices' is a list of ints which are the indices
of the markers to be deleted.
Every invocation of this function completely rewrites the obsstore file,
skipping the markers we want to be removed. The new temporary file is
created, remaining markers are written there and on .close() this file
gets atomically renamed to obsstore, thus guaranteeing consistency."""
if not indices:
# we don't want to rewrite the obsstore with the same content
return
left = []
current = obsstore._all
n = 0
for i, m in enumerate(current):
if i in indices:
n += 1
continue
left.append(m)
newobsstorefile = obsstore.svfs('obsstore', 'w', atomictemp=True)
for bytes in obsolete.encodemarkers(left, True, obsstore._version):
newobsstorefile.write(bytes)
newobsstorefile.close()
return n