##// 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:

r36753:ef345f9e default
r37304:9bfcbe4f default
Show More
transaction.py
639 lines | 23.1 KiB | text/x-python | PythonLexer
# transaction.py - simple journaling scheme for mercurial
#
# This transaction scheme is intended to gracefully handle program
# errors and interruptions. More serious failures like system crashes
# can be recovered with an fsck-like tool. As the whole repository is
# effectively log-structured, this should amount to simply truncating
# anything that isn't referenced in the changelog.
#
# Copyright 2005, 2006 Matt Mackall <mpm@selenic.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
from .i18n import _
from . import (
error,
pycompat,
util,
)
version = 2
# These are the file generators that should only be executed after the
# finalizers are done, since they rely on the output of the finalizers (like
# the changelog having been written).
postfinalizegenerators = {
'bookmarks',
'dirstate'
}
gengroupall='all'
gengroupprefinalize='prefinalize'
gengrouppostfinalize='postfinalize'
def active(func):
def _active(self, *args, **kwds):
if self.count == 0:
raise error.Abort(_(
'cannot use transaction when it is already committed/aborted'))
return func(self, *args, **kwds)
return _active
def _playback(journal, report, opener, vfsmap, entries, backupentries,
unlink=True, checkambigfiles=None):
for f, o, _ignore in entries:
if o or not unlink:
checkambig = checkambigfiles and (f, '') in checkambigfiles
try:
fp = opener(f, 'a', checkambig=checkambig)
fp.truncate(o)
fp.close()
except IOError:
report(_("failed to truncate %s\n") % f)
raise
else:
try:
opener.unlink(f)
except (IOError, OSError) as inst:
if inst.errno != errno.ENOENT:
raise
backupfiles = []
for l, f, b, c in backupentries:
if l not in vfsmap and c:
report("couldn't handle %s: unknown cache location %s\n"
% (b, l))
vfs = vfsmap[l]
try:
if f and b:
filepath = vfs.join(f)
backuppath = vfs.join(b)
checkambig = checkambigfiles and (f, l) in checkambigfiles
try:
util.copyfile(backuppath, filepath, checkambig=checkambig)
backupfiles.append(b)
except IOError:
report(_("failed to recover %s\n") % f)
else:
target = f or b
try:
vfs.unlink(target)
except (IOError, OSError) as inst:
if inst.errno != errno.ENOENT:
raise
except (IOError, OSError, error.Abort) as inst:
if not c:
raise
backuppath = "%s.backupfiles" % journal
if opener.exists(backuppath):
opener.unlink(backuppath)
opener.unlink(journal)
try:
for f in backupfiles:
if opener.exists(f):
opener.unlink(f)
except (IOError, OSError, error.Abort) as inst:
# only pure backup file remains, it is sage to ignore any error
pass
class transaction(util.transactional):
def __init__(self, report, opener, vfsmap, journalname, undoname=None,
after=None, createmode=None, validator=None, releasefn=None,
checkambigfiles=None, name=r'<unnamed>'):
"""Begin a new transaction
Begins a new transaction that allows rolling back writes in the event of
an exception.
* `after`: called after the transaction has been committed
* `createmode`: the mode of the journal file that will be created
* `releasefn`: called after releasing (with transaction and result)
`checkambigfiles` is a set of (path, vfs-location) tuples,
which determine whether file stat ambiguity should be avoided
for corresponded files.
"""
self.count = 1
self.usages = 1
self.report = report
# a vfs to the store content
self.opener = opener
# a map to access file in various {location -> vfs}
vfsmap = vfsmap.copy()
vfsmap[''] = opener # set default value
self._vfsmap = vfsmap
self.after = after
self.entries = []
self.map = {}
self.journal = journalname
self.undoname = undoname
self._queue = []
# A callback to validate transaction content before closing it.
# should raise exception is anything is wrong.
# target user is repository hooks.
if validator is None:
validator = lambda tr: None
self.validator = validator
# A callback to do something just after releasing transaction.
if releasefn is None:
releasefn = lambda tr, success: None
self.releasefn = releasefn
self.checkambigfiles = set()
if checkambigfiles:
self.checkambigfiles.update(checkambigfiles)
self.names = [name]
# A dict dedicated to precisely tracking the changes introduced in the
# transaction.
self.changes = {}
# a dict of arguments to be passed to hooks
self.hookargs = {}
self.file = opener.open(self.journal, "w")
# a list of ('location', 'path', 'backuppath', cache) entries.
# - if 'backuppath' is empty, no file existed at backup time
# - if 'path' is empty, this is a temporary transaction file
# - if 'location' is not empty, the path is outside main opener reach.
# use 'location' value as a key in a vfsmap to find the right 'vfs'
# (cache is currently unused)
self._backupentries = []
self._backupmap = {}
self._backupjournal = "%s.backupfiles" % self.journal
self._backupsfile = opener.open(self._backupjournal, 'w')
self._backupsfile.write('%d\n' % version)
if createmode is not None:
opener.chmod(self.journal, createmode & 0o666)
opener.chmod(self._backupjournal, createmode & 0o666)
# hold file generations to be performed on commit
self._filegenerators = {}
# hold callback to write pending data for hooks
self._pendingcallback = {}
# True is any pending data have been written ever
self._anypending = False
# holds callback to call when writing the transaction
self._finalizecallback = {}
# hold callback for post transaction close
self._postclosecallback = {}
# holds callbacks to call during abort
self._abortcallback = {}
def __repr__(self):
name = r'/'.join(self.names)
return (r'<transaction name=%s, count=%d, usages=%d>' %
(name, self.count, self.usages))
def __del__(self):
if self.journal:
self._abort()
@active
def startgroup(self):
"""delay registration of file entry
This is used by strip to delay vision of strip offset. The transaction
sees either none or all of the strip actions to be done."""
self._queue.append([])
@active
def endgroup(self):
"""apply delayed registration of file entry.
This is used by strip to delay vision of strip offset. The transaction
sees either none or all of the strip actions to be done."""
q = self._queue.pop()
for f, o, data in q:
self._addentry(f, o, data)
@active
def add(self, file, offset, data=None):
"""record the state of an append-only file before update"""
if file in self.map or file in self._backupmap:
return
if self._queue:
self._queue[-1].append((file, offset, data))
return
self._addentry(file, offset, data)
def _addentry(self, file, offset, data):
"""add a append-only entry to memory and on-disk state"""
if file in self.map or file in self._backupmap:
return
self.entries.append((file, offset, data))
self.map[file] = len(self.entries) - 1
# add enough data to the journal to do the truncate
self.file.write("%s\0%d\n" % (file, offset))
self.file.flush()
@active
def addbackup(self, file, hardlink=True, location=''):
"""Adds a backup of the file to the transaction
Calling addbackup() creates a hardlink backup of the specified file
that is used to recover the file in the event of the transaction
aborting.
* `file`: the file path, relative to .hg/store
* `hardlink`: use a hardlink to quickly create the backup
"""
if self._queue:
msg = 'cannot use transaction.addbackup inside "group"'
raise error.ProgrammingError(msg)
if file in self.map or file in self._backupmap:
return
vfs = self._vfsmap[location]
dirname, filename = vfs.split(file)
backupfilename = "%s.backup.%s" % (self.journal, filename)
backupfile = vfs.reljoin(dirname, backupfilename)
if vfs.exists(file):
filepath = vfs.join(file)
backuppath = vfs.join(backupfile)
util.copyfile(filepath, backuppath, hardlink=hardlink)
else:
backupfile = ''
self._addbackupentry((location, file, backupfile, False))
def _addbackupentry(self, entry):
"""register a new backup entry and write it to disk"""
self._backupentries.append(entry)
self._backupmap[entry[1]] = len(self._backupentries) - 1
self._backupsfile.write("%s\0%s\0%s\0%d\n" % entry)
self._backupsfile.flush()
@active
def registertmp(self, tmpfile, location=''):
"""register a temporary transaction file
Such files will be deleted when the transaction exits (on both
failure and success).
"""
self._addbackupentry((location, '', tmpfile, False))
@active
def addfilegenerator(self, genid, filenames, genfunc, order=0,
location=''):
"""add a function to generates some files at transaction commit
The `genfunc` argument is a function capable of generating proper
content of each entry in the `filename` tuple.
At transaction close time, `genfunc` will be called with one file
object argument per entries in `filenames`.
The transaction itself is responsible for the backup, creation and
final write of such file.
The `genid` argument is used to ensure the same set of file is only
generated once. Call to `addfilegenerator` for a `genid` already
present will overwrite the old entry.
The `order` argument may be used to control the order in which multiple
generator will be executed.
The `location` arguments may be used to indicate the files are located
outside of the the standard directory for transaction. It should match
one of the key of the `transaction.vfsmap` dictionary.
"""
# For now, we are unable to do proper backup and restore of custom vfs
# but for bookmarks that are handled outside this mechanism.
self._filegenerators[genid] = (order, filenames, genfunc, location)
@active
def removefilegenerator(self, genid):
"""reverse of addfilegenerator, remove a file generator function"""
if genid in self._filegenerators:
del self._filegenerators[genid]
def _generatefiles(self, suffix='', group=gengroupall):
# write files registered for generation
any = False
for id, entry in sorted(self._filegenerators.iteritems()):
any = True
order, filenames, genfunc, location = entry
# for generation at closing, check if it's before or after finalize
postfinalize = group == gengrouppostfinalize
if (group != gengroupall and
(id in postfinalizegenerators) != (postfinalize)):
continue
vfs = self._vfsmap[location]
files = []
try:
for name in filenames:
name += suffix
if suffix:
self.registertmp(name, location=location)
checkambig = False
else:
self.addbackup(name, location=location)
checkambig = (name, location) in self.checkambigfiles
files.append(vfs(name, 'w', atomictemp=True,
checkambig=checkambig))
genfunc(*files)
finally:
for f in files:
f.close()
return any
@active
def find(self, file):
if file in self.map:
return self.entries[self.map[file]]
if file in self._backupmap:
return self._backupentries[self._backupmap[file]]
return None
@active
def replace(self, file, offset, data=None):
'''
replace can only replace already committed entries
that are not pending in the queue
'''
if file not in self.map:
raise KeyError(file)
index = self.map[file]
self.entries[index] = (file, offset, data)
self.file.write("%s\0%d\n" % (file, offset))
self.file.flush()
@active
def nest(self, name=r'<unnamed>'):
self.count += 1
self.usages += 1
self.names.append(name)
return self
def release(self):
if self.count > 0:
self.usages -= 1
if self.names:
self.names.pop()
# if the transaction scopes are left without being closed, fail
if self.count > 0 and self.usages == 0:
self._abort()
def running(self):
return self.count > 0
def addpending(self, category, callback):
"""add a callback to be called when the transaction is pending
The transaction will be given as callback's first argument.
Category is a unique identifier to allow overwriting an old callback
with a newer callback.
"""
self._pendingcallback[category] = callback
@active
def writepending(self):
'''write pending file to temporary version
This is used to allow hooks to view a transaction before commit'''
categories = sorted(self._pendingcallback)
for cat in categories:
# remove callback since the data will have been flushed
any = self._pendingcallback.pop(cat)(self)
self._anypending = self._anypending or any
self._anypending |= self._generatefiles(suffix='.pending')
return self._anypending
@active
def addfinalize(self, category, callback):
"""add a callback to be called when the transaction is closed
The transaction will be given as callback's first argument.
Category is a unique identifier to allow overwriting old callbacks with
newer callbacks.
"""
self._finalizecallback[category] = callback
@active
def addpostclose(self, category, callback):
"""add or replace a callback to be called after the transaction closed
The transaction will be given as callback's first argument.
Category is a unique identifier to allow overwriting an old callback
with a newer callback.
"""
self._postclosecallback[category] = callback
@active
def getpostclose(self, category):
"""return a postclose callback added before, or None"""
return self._postclosecallback.get(category, None)
@active
def addabort(self, category, callback):
"""add a callback to be called when the transaction is aborted.
The transaction will be given as the first argument to the callback.
Category is a unique identifier to allow overwriting an old callback
with a newer callback.
"""
self._abortcallback[category] = callback
@active
def close(self):
'''commit the transaction'''
if self.count == 1:
self.validator(self) # will raise exception if needed
self.validator = None # Help prevent cycles.
self._generatefiles(group=gengroupprefinalize)
categories = sorted(self._finalizecallback)
for cat in categories:
self._finalizecallback[cat](self)
# Prevent double usage and help clear cycles.
self._finalizecallback = None
self._generatefiles(group=gengrouppostfinalize)
self.count -= 1
if self.count != 0:
return
self.file.close()
self._backupsfile.close()
# cleanup temporary files
for l, f, b, c in self._backupentries:
if l not in self._vfsmap and c:
self.report("couldn't remove %s: unknown cache location %s\n"
% (b, l))
continue
vfs = self._vfsmap[l]
if not f and b and vfs.exists(b):
try:
vfs.unlink(b)
except (IOError, OSError, error.Abort) as inst:
if not c:
raise
# Abort may be raise by read only opener
self.report("couldn't remove %s: %s\n"
% (vfs.join(b), inst))
self.entries = []
self._writeundo()
if self.after:
self.after()
self.after = None # Help prevent cycles.
if self.opener.isfile(self._backupjournal):
self.opener.unlink(self._backupjournal)
if self.opener.isfile(self.journal):
self.opener.unlink(self.journal)
for l, _f, b, c in self._backupentries:
if l not in self._vfsmap and c:
self.report("couldn't remove %s: unknown cache location"
"%s\n" % (b, l))
continue
vfs = self._vfsmap[l]
if b and vfs.exists(b):
try:
vfs.unlink(b)
except (IOError, OSError, error.Abort) as inst:
if not c:
raise
# Abort may be raise by read only opener
self.report("couldn't remove %s: %s\n"
% (vfs.join(b), inst))
self._backupentries = []
self.journal = None
self.releasefn(self, True) # notify success of closing transaction
self.releasefn = None # Help prevent cycles.
# run post close action
categories = sorted(self._postclosecallback)
for cat in categories:
self._postclosecallback[cat](self)
# Prevent double usage and help clear cycles.
self._postclosecallback = None
@active
def abort(self):
'''abort the transaction (generally called on error, or when the
transaction is not explicitly committed before going out of
scope)'''
self._abort()
def _writeundo(self):
"""write transaction data for possible future undo call"""
if self.undoname is None:
return
undobackupfile = self.opener.open("%s.backupfiles" % self.undoname, 'w')
undobackupfile.write('%d\n' % version)
for l, f, b, c in self._backupentries:
if not f: # temporary file
continue
if not b:
u = ''
else:
if l not in self._vfsmap and c:
self.report("couldn't remove %s: unknown cache location"
"%s\n" % (b, l))
continue
vfs = self._vfsmap[l]
base, name = vfs.split(b)
assert name.startswith(self.journal), name
uname = name.replace(self.journal, self.undoname, 1)
u = vfs.reljoin(base, uname)
util.copyfile(vfs.join(b), vfs.join(u), hardlink=True)
undobackupfile.write("%s\0%s\0%s\0%d\n" % (l, f, u, c))
undobackupfile.close()
def _abort(self):
self.count = 0
self.usages = 0
self.file.close()
self._backupsfile.close()
try:
if not self.entries and not self._backupentries:
if self._backupjournal:
self.opener.unlink(self._backupjournal)
if self.journal:
self.opener.unlink(self.journal)
return
self.report(_("transaction abort!\n"))
try:
for cat in sorted(self._abortcallback):
self._abortcallback[cat](self)
# Prevent double usage and help clear cycles.
self._abortcallback = None
_playback(self.journal, self.report, self.opener, self._vfsmap,
self.entries, self._backupentries, False,
checkambigfiles=self.checkambigfiles)
self.report(_("rollback completed\n"))
except BaseException:
self.report(_("rollback failed - please run hg recover\n"))
finally:
self.journal = None
self.releasefn(self, False) # notify failure of transaction
self.releasefn = None # Help prevent cycles.
def rollback(opener, vfsmap, file, report, checkambigfiles=None):
"""Rolls back the transaction contained in the given file
Reads the entries in the specified file, and the corresponding
'*.backupfiles' file, to recover from an incomplete transaction.
* `file`: a file containing a list of entries, specifying where
to truncate each file. The file should contain a list of
file\0offset pairs, delimited by newlines. The corresponding
'*.backupfiles' file should contain a list of file\0backupfile
pairs, delimited by \0.
`checkambigfiles` is a set of (path, vfs-location) tuples,
which determine whether file stat ambiguity should be avoided at
restoring corresponded files.
"""
entries = []
backupentries = []
fp = opener.open(file)
lines = fp.readlines()
fp.close()
for l in lines:
try:
f, o = l.split('\0')
entries.append((f, int(o), None))
except ValueError:
report(
_("couldn't read journal entry %r!\n") % pycompat.bytestr(l))
backupjournal = "%s.backupfiles" % file
if opener.exists(backupjournal):
fp = opener.open(backupjournal)
lines = fp.readlines()
if lines:
ver = lines[0][:-1]
if ver == (b'%d' % version):
for line in lines[1:]:
if line:
# Shave off the trailing newline
line = line[:-1]
l, f, b, c = line.split('\0')
backupentries.append((l, f, b, bool(c)))
else:
report(_("journal was created by a different version of "
"Mercurial\n"))
_playback(file, report, opener, vfsmap, entries, backupentries,
checkambigfiles=checkambigfiles)