Mercurial > public > mercurial-scm > hg
view mercurial/wireprototypes.py @ 40021:c537144fdbef
wireprotov2: support response caching
One of the things I've learned from managing VCS servers over the
years is that they are hard to scale. It is well known that some
companies have very beefy (read: very expensive) servers to power
their VCS needs. It is also known that specialized servers for
various VCS exist in order to facilitate scaling servers. (Mercurial
is in this boat.)
One of the aspects that make a VCS server hard to scale is the
high CPU load incurred by constant client clone/pull operations.
To alleviate the scaling pain associated with data retrieval
operations, I want to integrate caching into the Mercurial wire
protocol server as robustly as possible such that servers can
aggressively cache responses and defer as much server load as
possible.
This commit represents the initial implementation of a general
caching layer in wire protocol version 2.
We define a new interface and behavior for a wire protocol cacher
in repository.py. (This is probably where a reviewer should look
first to understand what is going on.)
The bulk of the added code is in wireprotov2server.py, where we
define how a command can opt in to being cached and integrate
caching into command dispatching.
From a very high-level:
* A command can declare itself as cacheable by providing a callable
that can be used to derive a cache key.
* At dispatch time, if a command is cacheable, we attempt to
construct a cacher and use it for serving the request and/or
caching the request.
* The dispatch layer handles the bulk of the business logic for
caching, making cachers mostly "dumb content stores."
* The mechanism for invalidating cached entries (one of the harder
parts about caching in general) is by varying the cache key when
state changes. As such, cachers don't need to be concerned with
cache invalidation.
Initially, we've hooked up support for caching "manifestdata" and
"filedata" commands. These are the simplest to cache, as they should
be immutable over time. Caching of commands related to changeset
data is a bit harder (because cache validation is impacted by
changes to bookmarks, phases, etc). This will be implemented later.
(Strictly speaking, censoring a file should invalidate caches. I've
added an inline TODO to track this edge case.)
To prove it works, this commit implements a test-only extension
providing in-memory caching backed by an lrucachedict. A new test
showing this extension behaving properly is added. FWIW, the
cacher is ~50 lines of code, demonstrating the relative ease with
which a cache can be added to a server.
While the test cacher is not suitable for production workloads, just
for kicks I performed a clone of just the changeset and manifest data
for the mozilla-unified repository. With a fully warmed cache (of just
the manifest data since changeset data is not cached), server-side
CPU usage dropped from ~73s to ~28s. That's pretty significant and
demonstrates the potential that response caching has on server
scalability!
Differential Revision: https://phab.mercurial-scm.org/D4773
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Wed, 26 Sep 2018 17:16:56 -0700 |
parents | ed919b90acda |
children | b099e6032f38 |
line wrap: on
line source
# Copyright 2018 Gregory Szorc <gregory.szorc@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 from .node import ( bin, hex, ) from .i18n import _ from .thirdparty import ( attr, ) from . import ( error, util, ) from .utils import ( interfaceutil, ) # Names of the SSH protocol implementations. SSHV1 = 'ssh-v1' # These are advertised over the wire. Increment the counters at the end # to reflect BC breakages. SSHV2 = 'exp-ssh-v2-0002' HTTP_WIREPROTO_V2 = 'exp-http-v2-0002' # All available wire protocol transports. TRANSPORTS = { SSHV1: { 'transport': 'ssh', 'version': 1, }, SSHV2: { 'transport': 'ssh', # TODO mark as version 2 once all commands are implemented. 'version': 1, }, 'http-v1': { 'transport': 'http', 'version': 1, }, HTTP_WIREPROTO_V2: { 'transport': 'http', 'version': 2, } } class bytesresponse(object): """A wire protocol response consisting of raw bytes.""" def __init__(self, data): self.data = data class ooberror(object): """wireproto reply: failure of a batch of operation Something failed during a batch call. The error message is stored in `self.message`. """ def __init__(self, message): self.message = message class pushres(object): """wireproto reply: success with simple integer return The call was successful and returned an integer contained in `self.res`. """ def __init__(self, res, output): self.res = res self.output = output class pusherr(object): """wireproto reply: failure The call failed. The `self.res` attribute contains the error message. """ def __init__(self, res, output): self.res = res self.output = output class streamres(object): """wireproto reply: binary stream The call was successful and the result is a stream. Accepts a generator containing chunks of data to be sent to the client. ``prefer_uncompressed`` indicates that the data is expected to be uncompressable and that the stream should therefore use the ``none`` engine. """ def __init__(self, gen=None, prefer_uncompressed=False): self.gen = gen self.prefer_uncompressed = prefer_uncompressed class streamreslegacy(object): """wireproto reply: uncompressed binary stream The call was successful and the result is a stream. Accepts a generator containing chunks of data to be sent to the client. Like ``streamres``, but sends an uncompressed data for "version 1" clients using the application/mercurial-0.1 media type. """ def __init__(self, gen=None): self.gen = gen # list of nodes encoding / decoding def decodelist(l, sep=' '): if l: return [bin(v) for v in l.split(sep)] return [] def encodelist(l, sep=' '): try: return sep.join(map(hex, l)) except TypeError: raise # batched call argument encoding def escapebatcharg(plain): return (plain .replace(':', ':c') .replace(',', ':o') .replace(';', ':s') .replace('=', ':e')) def unescapebatcharg(escaped): return (escaped .replace(':e', '=') .replace(':s', ';') .replace(':o', ',') .replace(':c', ':')) # mapping of options accepted by getbundle and their types # # Meant to be extended by extensions. It is extensions responsibility to ensure # such options are properly processed in exchange.getbundle. # # supported types are: # # :nodes: list of binary nodes # :csv: list of comma-separated values # :scsv: list of comma-separated values return as set # :plain: string with no transformation needed. GETBUNDLE_ARGUMENTS = { 'heads': 'nodes', 'bookmarks': 'boolean', 'common': 'nodes', 'obsmarkers': 'boolean', 'phases': 'boolean', 'bundlecaps': 'scsv', 'listkeys': 'csv', 'cg': 'boolean', 'cbattempted': 'boolean', 'stream': 'boolean', } class baseprotocolhandler(interfaceutil.Interface): """Abstract base class for wire protocol handlers. A wire protocol handler serves as an interface between protocol command handlers and the wire protocol transport layer. Protocol handlers provide methods to read command arguments, redirect stdio for the duration of the request, handle response types, etc. """ name = interfaceutil.Attribute( """The name of the protocol implementation. Used for uniquely identifying the transport type. """) def getargs(args): """return the value for arguments in <args> For version 1 transports, returns a list of values in the same order they appear in ``args``. For version 2 transports, returns a dict mapping argument name to value. """ def getprotocaps(): """Returns the list of protocol-level capabilities of client Returns a list of capabilities as declared by the client for the current request (or connection for stateful protocol handlers).""" def getpayload(): """Provide a generator for the raw payload. The caller is responsible for ensuring that the full payload is processed. """ def mayberedirectstdio(): """Context manager to possibly redirect stdio. The context manager yields a file-object like object that receives stdout and stderr output when the context manager is active. Or it yields ``None`` if no I/O redirection occurs. The intent of this context manager is to capture stdio output so it may be sent in the response. Some transports support streaming stdio to the client in real time. For these transports, stdio output won't be captured. """ def client(): """Returns a string representation of this client (as bytes).""" def addcapabilities(repo, caps): """Adds advertised capabilities specific to this protocol. Receives the list of capabilities collected so far. Returns a list of capabilities. The passed in argument can be returned. """ def checkperm(perm): """Validate that the client has permissions to perform a request. The argument is the permission required to proceed. If the client doesn't have that permission, the exception should raise or abort in a protocol specific manner. """ class commandentry(object): """Represents a declared wire protocol command.""" def __init__(self, func, args='', transports=None, permission='push', cachekeyfn=None): self.func = func self.args = args self.transports = transports or set() self.permission = permission self.cachekeyfn = cachekeyfn def _merge(self, func, args): """Merge this instance with an incoming 2-tuple. This is called when a caller using the old 2-tuple API attempts to replace an instance. The incoming values are merged with data not captured by the 2-tuple and a new instance containing the union of the two objects is returned. """ return commandentry(func, args=args, transports=set(self.transports), permission=self.permission) # Old code treats instances as 2-tuples. So expose that interface. def __iter__(self): yield self.func yield self.args def __getitem__(self, i): if i == 0: return self.func elif i == 1: return self.args else: raise IndexError('can only access elements 0 and 1') class commanddict(dict): """Container for registered wire protocol commands. It behaves like a dict. But __setitem__ is overwritten to allow silent coercion of values from 2-tuples for API compatibility. """ def __setitem__(self, k, v): if isinstance(v, commandentry): pass # Cast 2-tuples to commandentry instances. elif isinstance(v, tuple): if len(v) != 2: raise ValueError('command tuples must have exactly 2 elements') # It is common for extensions to wrap wire protocol commands via # e.g. ``wireproto.commands[x] = (newfn, args)``. Because callers # doing this aren't aware of the new API that uses objects to store # command entries, we automatically merge old state with new. if k in self: v = self[k]._merge(v[0], v[1]) else: # Use default values from @wireprotocommand. v = commandentry(v[0], args=v[1], transports=set(TRANSPORTS), permission='push') else: raise ValueError('command entries must be commandentry instances ' 'or 2-tuples') return super(commanddict, self).__setitem__(k, v) def commandavailable(self, command, proto): """Determine if a command is available for the requested protocol.""" assert proto.name in TRANSPORTS entry = self.get(command) if not entry: return False if proto.name not in entry.transports: return False return True def supportedcompengines(ui, role): """Obtain the list of supported compression engines for a request.""" assert role in (util.CLIENTROLE, util.SERVERROLE) compengines = util.compengines.supportedwireengines(role) # Allow config to override default list and ordering. if role == util.SERVERROLE: configengines = ui.configlist('server', 'compressionengines') config = 'server.compressionengines' else: # This is currently implemented mainly to facilitate testing. In most # cases, the server should be in charge of choosing a compression engine # because a server has the most to lose from a sub-optimal choice. (e.g. # CPU DoS due to an expensive engine or a network DoS due to poor # compression ratio). configengines = ui.configlist('experimental', 'clientcompressionengines') config = 'experimental.clientcompressionengines' # No explicit config. Filter out the ones that aren't supposed to be # advertised and return default ordering. if not configengines: attr = 'serverpriority' if role == util.SERVERROLE else 'clientpriority' return [e for e in compengines if getattr(e.wireprotosupport(), attr) > 0] # If compression engines are listed in the config, assume there is a good # reason for it (like server operators wanting to achieve specific # performance characteristics). So fail fast if the config references # unusable compression engines. validnames = set(e.name() for e in compengines) invalidnames = set(e for e in configengines if e not in validnames) if invalidnames: raise error.Abort(_('invalid compression engine defined in %s: %s') % (config, ', '.join(sorted(invalidnames)))) compengines = [e for e in compengines if e.name() in configengines] compengines = sorted(compengines, key=lambda e: configengines.index(e.name())) if not compengines: raise error.Abort(_('%s config option does not specify any known ' 'compression engines') % config, hint=_('usable compression engines: %s') % ', '.sorted(validnames)) return compengines @attr.s class encodedresponse(object): """Represents response data that is already content encoded. Wire protocol version 2 only. Commands typically emit Python objects that are encoded and sent over the wire. If commands emit an object of this type, the encoding step is bypassed and the content from this object is used instead. """ data = attr.ib()