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
# dagparser.py - parser and generator for concise description of DAGs
#
# Copyright 2010 Peter Arrenbrecht <peter@arrenbrecht.ch>
#
# 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 re
import string
from .i18n import _
from . import (
error,
pycompat,
)
from .utils import (
stringutil,
)
def parsedag(desc):
'''parses a DAG from a concise textual description; generates events
"+n" is a linear run of n nodes based on the current default parent
"." is a single node based on the current default parent
"$" resets the default parent to -1 (implied at the start);
otherwise the default parent is always the last node created
"<p" sets the default parent to the backref p
"*p" is a fork at parent p, where p is a backref
"*p1/p2/.../pn" is a merge of parents p1..pn, where the pi are backrefs
"/p2/.../pn" is a merge of the preceding node and p2..pn
":name" defines a label for the preceding node; labels can be redefined
"@text" emits an annotation event for text
"!command" emits an action event for the current node
"!!my command\n" is like "!", but to the end of the line
"#...\n" is a comment up to the end of the line
Whitespace between the above elements is ignored.
A backref is either
* a number n, which references the node curr-n, where curr is the current
node, or
* the name of a label you placed earlier using ":name", or
* empty to denote the default parent.
All string valued-elements are either strictly alphanumeric, or must
be enclosed in double quotes ("..."), with "\" as escape character.
Generates sequence of
('n', (id, [parentids])) for node creation
('l', (id, labelname)) for labels on nodes
('a', text) for annotations
('c', command) for actions (!)
('C', command) for line actions (!!)
Examples
--------
Example of a complex graph (output not shown for brevity):
>>> len(list(parsedag(b"""
...
... +3 # 3 nodes in linear run
... :forkhere # a label for the last of the 3 nodes from above
... +5 # 5 more nodes on one branch
... :mergethis # label again
... <forkhere # set default parent to labeled fork node
... +10 # 10 more nodes on a parallel branch
... @stable # following nodes will be annotated as "stable"
... +5 # 5 nodes in stable
... !addfile # custom command; could trigger new file in next node
... +2 # two more nodes
... /mergethis # merge last node with labeled node
... +4 # 4 more nodes descending from merge node
...
... """)))
34
Empty list:
>>> list(parsedag(b""))
[]
A simple linear run:
>>> list(parsedag(b"+3"))
[('n', (0, [-1])), ('n', (1, [0])), ('n', (2, [1]))]
Some non-standard ways to define such runs:
>>> list(parsedag(b"+1+2"))
[('n', (0, [-1])), ('n', (1, [0])), ('n', (2, [1]))]
>>> list(parsedag(b"+1*1*"))
[('n', (0, [-1])), ('n', (1, [0])), ('n', (2, [1]))]
>>> list(parsedag(b"*"))
[('n', (0, [-1]))]
>>> list(parsedag(b"..."))
[('n', (0, [-1])), ('n', (1, [0])), ('n', (2, [1]))]
A fork and a join, using numeric back references:
>>> list(parsedag(b"+2*2*/2"))
[('n', (0, [-1])), ('n', (1, [0])), ('n', (2, [0])), ('n', (3, [2, 1]))]
>>> list(parsedag(b"+2<2+1/2"))
[('n', (0, [-1])), ('n', (1, [0])), ('n', (2, [0])), ('n', (3, [2, 1]))]
Placing a label:
>>> list(parsedag(b"+1 :mylabel +1"))
[('n', (0, [-1])), ('l', (0, 'mylabel')), ('n', (1, [0]))]
An empty label (silly, really):
>>> list(parsedag(b"+1:+1"))
[('n', (0, [-1])), ('l', (0, '')), ('n', (1, [0]))]
Fork and join, but with labels instead of numeric back references:
>>> list(parsedag(b"+1:f +1:p2 *f */p2"))
[('n', (0, [-1])), ('l', (0, 'f')), ('n', (1, [0])), ('l', (1, 'p2')),
('n', (2, [0])), ('n', (3, [2, 1]))]
>>> list(parsedag(b"+1:f +1:p2 <f +1 /p2"))
[('n', (0, [-1])), ('l', (0, 'f')), ('n', (1, [0])), ('l', (1, 'p2')),
('n', (2, [0])), ('n', (3, [2, 1]))]
Restarting from the root:
>>> list(parsedag(b"+1 $ +1"))
[('n', (0, [-1])), ('n', (1, [-1]))]
Annotations, which are meant to introduce sticky state for subsequent nodes:
>>> list(parsedag(b"+1 @ann +1"))
[('n', (0, [-1])), ('a', 'ann'), ('n', (1, [0]))]
>>> list(parsedag(b'+1 @"my annotation" +1'))
[('n', (0, [-1])), ('a', 'my annotation'), ('n', (1, [0]))]
Commands, which are meant to operate on the most recently created node:
>>> list(parsedag(b"+1 !cmd +1"))
[('n', (0, [-1])), ('c', 'cmd'), ('n', (1, [0]))]
>>> list(parsedag(b'+1 !"my command" +1'))
[('n', (0, [-1])), ('c', 'my command'), ('n', (1, [0]))]
>>> list(parsedag(b'+1 !!my command line\\n +1'))
[('n', (0, [-1])), ('C', 'my command line'), ('n', (1, [0]))]
Comments, which extend to the end of the line:
>>> list(parsedag(b'+1 # comment\\n+1'))
[('n', (0, [-1])), ('n', (1, [0]))]
Error:
>>> try: list(parsedag(b'+1 bad'))
... except Exception as e: print(pycompat.sysstr(bytes(e)))
invalid character in dag description: bad...
'''
if not desc:
return
wordchars = pycompat.bytestr(string.ascii_letters + string.digits)
labels = {}
p1 = -1
r = 0
def resolve(ref):
if not ref:
return p1
elif ref[0] in pycompat.bytestr(string.digits):
return r - int(ref)
else:
return labels[ref]
chiter = pycompat.iterbytestr(desc)
def nextch():
return next(chiter, '\0')
def nextrun(c, allow):
s = ''
while c in allow:
s += c
c = nextch()
return c, s
def nextdelimited(c, limit, escape):
s = ''
while c != limit:
if c == escape:
c = nextch()
s += c
c = nextch()
return nextch(), s
def nextstring(c):
if c == '"':
return nextdelimited(nextch(), '"', '\\')
else:
return nextrun(c, wordchars)
c = nextch()
while c != '\0':
while c in pycompat.bytestr(string.whitespace):
c = nextch()
if c == '.':
yield 'n', (r, [p1])
p1 = r
r += 1
c = nextch()
elif c == '+':
c, digs = nextrun(nextch(), pycompat.bytestr(string.digits))
n = int(digs)
for i in pycompat.xrange(0, n):
yield 'n', (r, [p1])
p1 = r
r += 1
elif c in '*/':
if c == '*':
c = nextch()
c, pref = nextstring(c)
prefs = [pref]
while c == '/':
c, pref = nextstring(nextch())
prefs.append(pref)
ps = [resolve(ref) for ref in prefs]
yield 'n', (r, ps)
p1 = r
r += 1
elif c == '<':
c, ref = nextstring(nextch())
p1 = resolve(ref)
elif c == ':':
c, name = nextstring(nextch())
labels[name] = p1
yield 'l', (p1, name)
elif c == '@':
c, text = nextstring(nextch())
yield 'a', text
elif c == '!':
c = nextch()
if c == '!':
cmd = ''
c = nextch()
while c not in '\n\r\0':
cmd += c
c = nextch()
yield 'C', cmd
else:
c, cmd = nextstring(c)
yield 'c', cmd
elif c == '#':
while c not in '\n\r\0':
c = nextch()
elif c == '$':
p1 = -1
c = nextch()
elif c == '\0':
return # in case it was preceded by whitespace
else:
s = ''
i = 0
while c != '\0' and i < 10:
s += c
i += 1
c = nextch()
raise error.Abort(_('invalid character in dag description: '
'%s...') % s)
def dagtextlines(events,
addspaces=True,
wraplabels=False,
wrapannotations=False,
wrapcommands=False,
wrapnonlinear=False,
usedots=False,
maxlinewidth=70):
'''generates single lines for dagtext()'''
def wrapstring(text):
if re.match("^[0-9a-z]*$", text):
return text
return '"' + text.replace('\\', '\\\\').replace('"', '\"') + '"'
def gen():
labels = {}
run = 0
wantr = 0
needroot = False
for kind, data in events:
if kind == 'n':
r, ps = data
# sanity check
if r != wantr:
raise error.Abort(_("expected id %i, got %i") % (wantr, r))
if not ps:
ps = [-1]
else:
for p in ps:
if p >= r:
raise error.Abort(_("parent id %i is larger than "
"current id %i") % (p, r))
wantr += 1
# new root?
p1 = r - 1
if len(ps) == 1 and ps[0] == -1:
if needroot:
if run:
yield '+%d' % run
run = 0
if wrapnonlinear:
yield '\n'
yield '$'
p1 = -1
else:
needroot = True
if len(ps) == 1 and ps[0] == p1:
if usedots:
yield "."
else:
run += 1
else:
if run:
yield '+%d' % run
run = 0
if wrapnonlinear:
yield '\n'
prefs = []
for p in ps:
if p == p1:
prefs.append('')
elif p in labels:
prefs.append(labels[p])
else:
prefs.append('%d' % (r - p))
yield '*' + '/'.join(prefs)
else:
if run:
yield '+%d' % run
run = 0
if kind == 'l':
rid, name = data
labels[rid] = name
yield ':' + name
if wraplabels:
yield '\n'
elif kind == 'c':
yield '!' + wrapstring(data)
if wrapcommands:
yield '\n'
elif kind == 'C':
yield '!!' + data
yield '\n'
elif kind == 'a':
if wrapannotations:
yield '\n'
yield '@' + wrapstring(data)
elif kind == '#':
yield '#' + data
yield '\n'
else:
raise error.Abort(_("invalid event type in dag: "
"('%s', '%s')")
% (stringutil.escapestr(kind),
stringutil.escapestr(data)))
if run:
yield '+%d' % run
line = ''
for part in gen():
if part == '\n':
if line:
yield line
line = ''
else:
if len(line) + len(part) >= maxlinewidth:
yield line
line = ''
elif addspaces and line and part != '.':
line += ' '
line += part
if line:
yield line
def dagtext(dag,
addspaces=True,
wraplabels=False,
wrapannotations=False,
wrapcommands=False,
wrapnonlinear=False,
usedots=False,
maxlinewidth=70):
'''generates lines of a textual representation for a dag event stream
events should generate what parsedag() does, so:
('n', (id, [parentids])) for node creation
('l', (id, labelname)) for labels on nodes
('a', text) for annotations
('c', text) for commands
('C', text) for line commands ('!!')
('#', text) for comment lines
Parent nodes must come before child nodes.
Examples
--------
Linear run:
>>> dagtext([(b'n', (0, [-1])), (b'n', (1, [0]))])
'+2'
Two roots:
>>> dagtext([(b'n', (0, [-1])), (b'n', (1, [-1]))])
'+1 $ +1'
Fork and join:
>>> dagtext([(b'n', (0, [-1])), (b'n', (1, [0])), (b'n', (2, [0])),
... (b'n', (3, [2, 1]))])
'+2 *2 */2'
Fork and join with labels:
>>> dagtext([(b'n', (0, [-1])), (b'l', (0, b'f')), (b'n', (1, [0])),
... (b'l', (1, b'p2')), (b'n', (2, [0])), (b'n', (3, [2, 1]))])
'+1 :f +1 :p2 *f */p2'
Annotations:
>>> dagtext([(b'n', (0, [-1])), (b'a', b'ann'), (b'n', (1, [0]))])
'+1 @ann +1'
>>> dagtext([(b'n', (0, [-1])),
... (b'a', b'my annotation'),
... (b'n', (1, [0]))])
'+1 @"my annotation" +1'
Commands:
>>> dagtext([(b'n', (0, [-1])), (b'c', b'cmd'), (b'n', (1, [0]))])
'+1 !cmd +1'
>>> dagtext([(b'n', (0, [-1])),
... (b'c', b'my command'),
... (b'n', (1, [0]))])
'+1 !"my command" +1'
>>> dagtext([(b'n', (0, [-1])),
... (b'C', b'my command line'),
... (b'n', (1, [0]))])
'+1 !!my command line\\n+1'
Comments:
>>> dagtext([(b'n', (0, [-1])), (b'#', b' comment'), (b'n', (1, [0]))])
'+1 # comment\\n+1'
>>> dagtext([])
''
Combining parsedag and dagtext:
>>> dagtext(parsedag(b'+1 :f +1 :p2 *f */p2'))
'+1 :f +1 :p2 *f */p2'
'''
return "\n".join(dagtextlines(dag,
addspaces,
wraplabels,
wrapannotations,
wrapcommands,
wrapnonlinear,
usedots,
maxlinewidth))