Mercurial > hg
view tests/test-pathencode.py @ 30442:41a8106789ca
util: implement zstd compression engine
Now that zstd is vendored and being built (in some configurations), we
can implement a compression engine for zstd!
The zstd engine is a little different from existing engines. Because
it may not always be present, we have to defer load the module in case
importing it fails. We facilitate this via a cached property that holds
a reference to the module or None. The "available" method is
implemented to reflect reality.
The zstd engine declares its ability to handle bundles using the
"zstd" human name and the "ZS" internal name. The latter was chosen
because internal names are 2 characters (by only convention I think)
and "ZS" seems reasonable.
The engine, like others, supports specifying the compression level.
However, there are no consumers of this API that yet pass in that
argument. I have plans to change that, so stay tuned.
Since all we need to do to support bundle generation with a new
compression engine is implement and register the compression engine,
bundle generation with zstd "just works!" Tests demonstrating this
have been added.
How does performance of zstd for bundle generation compare? On the
mozilla-unified repo, `hg bundle --all -t <engine>-v2` yields the
following on my i7-6700K on Linux:
engine CPU time bundle size vs orig size throughput
none 97.0s 4,054,405,584 100.0% 41.8 MB/s
bzip2 (l=9) 393.6s 975,343,098 24.0% 10.3 MB/s
gzip (l=6) 184.0s 1,140,533,074 28.1% 22.0 MB/s
zstd (l=1) 108.2s 1,119,434,718 27.6% 37.5 MB/s
zstd (l=2) 111.3s 1,078,328,002 26.6% 36.4 MB/s
zstd (l=3) 113.7s 1,011,823,727 25.0% 35.7 MB/s
zstd (l=4) 116.0s 1,008,965,888 24.9% 35.0 MB/s
zstd (l=5) 121.0s 977,203,148 24.1% 33.5 MB/s
zstd (l=6) 131.7s 927,360,198 22.9% 30.8 MB/s
zstd (l=7) 139.0s 912,808,505 22.5% 29.2 MB/s
zstd (l=12) 198.1s 854,527,714 21.1% 20.5 MB/s
zstd (l=18) 681.6s 789,750,690 19.5% 5.9 MB/s
On compression, zstd for bundle generation delivers:
* better compression than gzip with significantly less CPU utilization
* better than bzip2 compression ratios while still being significantly
faster than gzip
* ability to aggressively tune compression level to achieve
significantly smaller bundles
That last point is important. With clone bundles, a server can
pre-generate a bundle file, upload it to a static file server, and
redirect clients to transparently download it during clone. The server
could choose to produce a zstd bundle with the highest compression
settings possible. This would take a very long time - a magnitude
longer than a typical zstd bundle generation - but the result would
be hundreds of megabytes smaller! For the clone volume we do at
Mozilla, this could translate to petabytes of bandwidth savings
per year and faster clones (due to smaller transfer size).
I don't have detailed numbers to report on decompression. However,
zstd decompression is fast: >1 GB/s output throughput on this machine,
even through the Python bindings. And it can do that regardless of the
compression level of the input. By the time you have enough data to
worry about overhead of decompression, you have plenty of other things
to worry about performance wise.
zstd is wins all around. I can't wait to implement support for it
on the wire protocol and in revlogs.
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Fri, 11 Nov 2016 01:10:07 -0800 |
parents | 59481bfdb7f3 |
children | 0f200e2310ca |
line wrap: on
line source
# This is a randomized test that generates different pathnames every # time it is invoked, and tests the encoding of those pathnames. # # It uses a simple probabilistic model to generate valid pathnames # that have proven likely to expose bugs and divergent behavior in # different encoding implementations. from __future__ import absolute_import, print_function import binascii import collections import itertools import math import os import random import sys import time from mercurial import ( store, ) validchars = set(map(chr, range(0, 256))) alphanum = range(ord('A'), ord('Z')) for c in '\0/': validchars.remove(c) winreserved = ('aux con prn nul'.split() + ['com%d' % i for i in xrange(1, 10)] + ['lpt%d' % i for i in xrange(1, 10)]) def casecombinations(names): '''Build all case-diddled combinations of names.''' combos = set() for r in names: for i in xrange(len(r) + 1): for c in itertools.combinations(xrange(len(r)), i): d = r for j in c: d = ''.join((d[:j], d[j].upper(), d[j + 1:])) combos.add(d) return sorted(combos) def buildprobtable(fp, cmd='hg manifest tip'): '''Construct and print a table of probabilities for path name components. The numbers are percentages.''' counts = collections.defaultdict(lambda: 0) for line in os.popen(cmd).read().splitlines(): if line[-2:] in ('.i', '.d'): line = line[:-2] if line.startswith('data/'): line = line[5:] for c in line: counts[c] += 1 for c in '\r/\n': counts.pop(c, None) t = sum(counts.itervalues()) / 100.0 fp.write('probtable = (') for i, (k, v) in enumerate(sorted(counts.iteritems(), key=lambda x: x[1], reverse=True)): if (i % 5) == 0: fp.write('\n ') vt = v / t if vt < 0.0005: break fp.write('(%r, %.03f), ' % (k, vt)) fp.write('\n )\n') # A table of character frequencies (as percentages), gleaned by # looking at filelog names from a real-world, very large repo. probtable = ( ('t', 9.828), ('e', 9.042), ('s', 8.011), ('a', 6.801), ('i', 6.618), ('g', 5.053), ('r', 5.030), ('o', 4.887), ('p', 4.363), ('n', 4.258), ('l', 3.830), ('h', 3.693), ('_', 3.659), ('.', 3.377), ('m', 3.194), ('u', 2.364), ('d', 2.296), ('c', 2.163), ('b', 1.739), ('f', 1.625), ('6', 0.666), ('j', 0.610), ('y', 0.554), ('x', 0.487), ('w', 0.477), ('k', 0.476), ('v', 0.473), ('3', 0.336), ('1', 0.335), ('2', 0.326), ('4', 0.310), ('5', 0.305), ('9', 0.302), ('8', 0.300), ('7', 0.299), ('q', 0.298), ('0', 0.250), ('z', 0.223), ('-', 0.118), ('C', 0.095), ('T', 0.087), ('F', 0.085), ('B', 0.077), ('S', 0.076), ('P', 0.076), ('L', 0.059), ('A', 0.058), ('N', 0.051), ('D', 0.049), ('M', 0.046), ('E', 0.039), ('I', 0.035), ('R', 0.035), ('G', 0.028), ('U', 0.026), ('W', 0.025), ('O', 0.017), ('V', 0.015), ('H', 0.013), ('Q', 0.011), ('J', 0.007), ('K', 0.005), ('+', 0.004), ('X', 0.003), ('Y', 0.001), ) for c, _ in probtable: validchars.remove(c) validchars = list(validchars) def pickfrom(rng, table): c = 0 r = rng.random() * sum(i[1] for i in table) for i, p in table: c += p if c >= r: return i reservedcombos = casecombinations(winreserved) # The first component of a name following a slash. firsttable = ( (lambda rng: pickfrom(rng, probtable), 90), (lambda rng: rng.choice(validchars), 5), (lambda rng: rng.choice(reservedcombos), 5), ) # Components of a name following the first. resttable = firsttable[:-1] # Special suffixes. internalsuffixcombos = casecombinations('.hg .i .d'.split()) # The last component of a path, before a slash or at the end of a name. lasttable = resttable + ( (lambda rng: '', 95), (lambda rng: rng.choice(internalsuffixcombos), 5), ) def makepart(rng, k): '''Construct a part of a pathname, without slashes.''' p = pickfrom(rng, firsttable)(rng) l = len(p) ps = [p] maxl = rng.randint(1, k) while l < maxl: p = pickfrom(rng, resttable)(rng) l += len(p) ps.append(p) ps.append(pickfrom(rng, lasttable)(rng)) return ''.join(ps) def makepath(rng, j, k): '''Construct a complete pathname.''' return ('data/' + '/'.join(makepart(rng, k) for _ in xrange(j)) + rng.choice(['.d', '.i'])) def genpath(rng, count): '''Generate random pathnames with gradually increasing lengths.''' mink, maxk = 1, 4096 def steps(): for i in xrange(count): yield mink + int(round(math.sqrt((maxk - mink) * float(i) / count))) for k in steps(): x = rng.randint(1, k) y = rng.randint(1, k) yield makepath(rng, x, y) def runtests(rng, seed, count): nerrs = 0 for p in genpath(rng, count): h = store._pathencode(p) # uses C implementation, if available r = store._hybridencode(p, True) # reference implementation in Python if h != r: if nerrs == 0: print('seed:', hex(seed)[:-1], file=sys.stderr) print("\np: '%s'" % p.encode("string_escape"), file=sys.stderr) print("h: '%s'" % h.encode("string_escape"), file=sys.stderr) print("r: '%s'" % r.encode("string_escape"), file=sys.stderr) nerrs += 1 return nerrs def main(): import getopt # Empirically observed to take about a second to run count = 100 seed = None opts, args = getopt.getopt(sys.argv[1:], 'c:s:', ['build', 'count=', 'seed=']) for o, a in opts: if o in ('-c', '--count'): count = int(a) elif o in ('-s', '--seed'): seed = long(a, base=0) # accepts base 10 or 16 strings elif o == '--build': buildprobtable(sys.stdout, 'find .hg/store/data -type f && ' 'cat .hg/store/fncache 2>/dev/null') sys.exit(0) if seed is None: try: seed = long(binascii.hexlify(os.urandom(16)), 16) except AttributeError: seed = long(time.time() * 1000) rng = random.Random(seed) if runtests(rng, seed, count): sys.exit(1) if __name__ == '__main__': main()