Mercurial > hg
view mercurial/namespaces.py @ 35793:4fb2bb61597c
bundle2: increase payload part chunk size to 32kb
Bundle2 payload parts are framed chunks. Esentially, we obtain
data in equal size chunks of size `preferedchunksize` and emit those
to a generator. That generator is fed into a compressor (which can
be the no-op compressor, which just re-emits the generator). And
the output from the compressor likely goes to a file descriptor
or socket.
What this means is that small chunk sizes create more Python objects
and Python function calls than larger chunk sizes. And as we know,
Python object and function call overhead in performance sensitive
code matters (at least with CPython).
This commit increases the bundle2 part payload chunk size from 4k
to 32k. Practically speaking, this means that the chunks we feed
into a compressor (implemented in C code) or feed directly into a
file handle or socket write() are larger. It's possible the chunks
might be larger than what the receiver can handle in one logical
operation. But at that point, we're in C code, which is much more
efficient at dealing with splitting up the chunk and making multiple
function calls than Python is.
A downside to larger chunks is that the receiver has to wait for that
much data to arrive (either raw or from a decompressor) before it
can process the chunk. But 32kb still feels like a small buffer to
have to wait for. And in many cases, the client will convert from
8 read(4096) to 1 read(32768). That's happening in Python land. So
we cut down on the number of Python objects and function calls,
making the client faster as well. I don't think there are any
significant concerns to increasing the payload chunk size to 32kb.
The impact of this change on performance significant. Using `curl`
to obtain a stream clone bundle2 payload from a server on localhost
serving the mozilla-unified repository:
before: 20.78 user; 7.71 system; 80.5 MB/s
after: 13.90 user; 3.51 system; 132 MB/s
legacy: 9.72 user; 8.16 system; 132 MB/s
bundle2 stream clone generation is still more resource intensive than
legacy stream clone (that's likely because of the use of a
util.chunkbuffer). But the throughput is the same. We might
be in territory we're this is effectively a benchmark of the
networking stack or Python's syscall throughput.
From the client perspective, `hg clone -U --stream`:
before: 33.50 user; 7.95 system; 53.3 MB/s
after: 22.82 user; 7.33 system; 72.7 MB/s
legacy: 29.96 user; 7.94 system; 58.0 MB/s
And for `hg clone --stream` with a working directory update of
~230k files:
after: 119.55 user; 26.47 system; 0:57.08 wall
legacy: 126.98 user; 26.94 system; 1:05.56 wall
So, it appears that bundle2's stream clone is now definitively faster
than legacy stream clone!
Differential Revision: https://phab.mercurial-scm.org/D1932
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Sat, 20 Jan 2018 22:55:42 -0800 |
parents | c7b45db8f317 |
children | b0054f3c055a |
line wrap: on
line source
from __future__ import absolute_import from .i18n import _ from . import ( templatekw, util, ) def tolist(val): """ a convenience method to return an empty list instead of None """ if val is None: return [] else: return [val] class namespaces(object): """provides an interface to register and operate on multiple namespaces. See the namespace class below for details on the namespace object. """ _names_version = 0 def __init__(self): self._names = util.sortdict() columns = templatekw.getlogcolumns() # we need current mercurial named objects (bookmarks, tags, and # branches) to be initialized somewhere, so that place is here bmknames = lambda repo: repo._bookmarks.keys() bmknamemap = lambda repo, name: tolist(repo._bookmarks.get(name)) bmknodemap = lambda repo, node: repo.nodebookmarks(node) n = namespace("bookmarks", templatename="bookmark", logfmt=columns['bookmark'], listnames=bmknames, namemap=bmknamemap, nodemap=bmknodemap, builtin=True) self.addnamespace(n) tagnames = lambda repo: [t for t, n in repo.tagslist()] tagnamemap = lambda repo, name: tolist(repo._tagscache.tags.get(name)) tagnodemap = lambda repo, node: repo.nodetags(node) n = namespace("tags", templatename="tag", logfmt=columns['tag'], listnames=tagnames, namemap=tagnamemap, nodemap=tagnodemap, deprecated={'tip'}, builtin=True) self.addnamespace(n) bnames = lambda repo: repo.branchmap().keys() bnamemap = lambda repo, name: tolist(repo.branchtip(name, True)) bnodemap = lambda repo, node: [repo[node].branch()] n = namespace("branches", templatename="branch", logfmt=columns['branch'], listnames=bnames, namemap=bnamemap, nodemap=bnodemap, builtin=True) self.addnamespace(n) def __getitem__(self, namespace): """returns the namespace object""" return self._names[namespace] def __iter__(self): return self._names.__iter__() def items(self): return self._names.iteritems() iteritems = items def addnamespace(self, namespace, order=None): """register a namespace namespace: the name to be registered (in plural form) order: optional argument to specify the order of namespaces (e.g. 'branches' should be listed before 'bookmarks') """ if order is not None: self._names.insert(order, namespace.name, namespace) else: self._names[namespace.name] = namespace # we only generate a template keyword if one does not already exist if namespace.name not in templatekw.keywords: def generatekw(**args): return templatekw.shownames(namespace.name, **args) templatekw.keywords[namespace.name] = generatekw def singlenode(self, repo, name): """ Return the 'best' node for the given name. Best means the first node in the first nonempty list returned by a name-to-nodes mapping function in the defined precedence order. Raises a KeyError if there is no such node. """ for ns, v in self._names.iteritems(): n = v.namemap(repo, name) if n: # return max revision number if len(n) > 1: cl = repo.changelog maxrev = max(cl.rev(node) for node in n) return cl.node(maxrev) return n[0] raise KeyError(_('no such name: %s') % name) class namespace(object): """provides an interface to a namespace Namespaces are basically generic many-to-many mapping between some (namespaced) names and nodes. The goal here is to control the pollution of jamming things into tags or bookmarks (in extension-land) and to simplify internal bits of mercurial: log output, tab completion, etc. More precisely, we define a mapping of names to nodes, and a mapping from nodes to names. Each mapping returns a list. Furthermore, each name mapping will be passed a name to lookup which might not be in its domain. In this case, each method should return an empty list and not raise an error. This namespace object will define the properties we need: 'name': the namespace (plural form) 'templatename': name to use for templating (usually the singular form of the plural namespace name) 'listnames': list of all names in the namespace (usually the keys of a dictionary) 'namemap': function that takes a name and returns a list of nodes 'nodemap': function that takes a node and returns a list of names 'deprecated': set of names to be masked for ordinary use 'builtin': bool indicating if this namespace is supported by core Mercurial. """ def __init__(self, name, templatename=None, logname=None, colorname=None, logfmt=None, listnames=None, namemap=None, nodemap=None, deprecated=None, builtin=False): """create a namespace name: the namespace to be registered (in plural form) templatename: the name to use for templating logname: the name to use for log output; if not specified templatename is used colorname: the name to use for colored log output; if not specified logname is used logfmt: the format to use for (i18n-ed) log output; if not specified it is composed from logname listnames: function to list all names namemap: function that inputs a name, output node(s) nodemap: function that inputs a node, output name(s) deprecated: set of names to be masked for ordinary use builtin: whether namespace is implemented by core Mercurial """ self.name = name self.templatename = templatename self.logname = logname self.colorname = colorname self.logfmt = logfmt self.listnames = listnames self.namemap = namemap self.nodemap = nodemap # if logname is not specified, use the template name as backup if self.logname is None: self.logname = self.templatename # if colorname is not specified, just use the logname as a backup if self.colorname is None: self.colorname = self.logname # if logfmt is not specified, compose it from logname as backup if self.logfmt is None: # i18n: column positioning for "hg log" self.logfmt = ("%s:" % self.logname).ljust(13) + "%s\n" if deprecated is None: self.deprecated = set() else: self.deprecated = deprecated self.builtin = builtin def names(self, repo, node): """method that returns a (sorted) list of names in a namespace that match a given node""" return sorted(self.nodemap(repo, node)) def nodes(self, repo, name): """method that returns a list of nodes in a namespace that match a given name. """ return sorted(self.namemap(repo, name))