localrepo: experimental support for non-zlib revlog compression
The final part of integrating the compression manager APIs into
revlog storage is the plumbing for repositories to advertise they
are using non-zlib storage and for revlogs to instantiate a non-zlib
compression engine.
The main intent of the compression manager work was to zstd all
of the things. Adding zstd to revlogs has proved to be more involved
than other places because revlogs are... special. Very small inputs
and the use of delta chains (which are themselves a form of
compression) are a completely different use case from streaming
compression, which bundles and the wire protocol employ. I've
conducted numerous experiments with zstd in revlogs and have yet
to formalize compression settings and a storage architecture that
I'm confident I won't regret later. In other words, I'm not yet
ready to commit to a new mechanism for using zstd - or any other
compression format - in revlogs.
That being said, having some support for zstd (and other compression
formats) in revlogs in core is beneficial. It can allow others to
conduct experiments.
This patch introduces *highly experimental* support for non-zlib
compression formats in revlogs. Introduced is a config option to
control which compression engine to use. Also introduced is a namespace
of "exp-compression-*" requirements to denote support for non-zlib
compression in revlogs. I've prefixed the namespace with "exp-"
(short for "experimental") because I'm not confident of the
requirements "schema" and in no way want to give the illusion of
supporting these requirements in the future. I fully intend to drop
support for these requirements once we figure out what we're doing
with zstd in revlogs.
A good portion of the patch is teaching the requirements system
about registered compression engines and passing the requested
compression engine as an opener option so revlogs can instantiate
the proper compression engine for new operations.
That's a verbose way of saying "we can now use zstd in revlogs!"
On an `hg pull` conversion of the mozilla-unified repo with no extra
redelta settings (like aggressivemergedeltas), we can see the impact
of zstd vs zlib in revlogs:
$ hg perfrevlogchunks -c
! chunk
! wall 2.032052 comb 2.040000 user 1.990000 sys 0.050000 (best of 5)
! wall 1.866360 comb 1.860000 user 1.820000 sys 0.040000 (best of 6)
! chunk batch
! wall 1.877261 comb 1.870000 user 1.860000 sys 0.010000 (best of 6)
! wall 1.705410 comb 1.710000 user 1.690000 sys 0.020000 (best of 6)
$ hg perfrevlogchunks -m
! chunk
! wall 2.721427 comb 2.720000 user 2.640000 sys 0.080000 (best of 4)
! wall 2.035076 comb 2.030000 user 1.950000 sys 0.080000 (best of 5)
! chunk batch
! wall 2.614561 comb 2.620000 user 2.580000 sys 0.040000 (best of 4)
! wall 1.910252 comb 1.910000 user 1.880000 sys 0.030000 (best of 6)
$ hg perfrevlog -c -d 1
! wall 4.812885 comb 4.820000 user 4.800000 sys 0.020000 (best of 3)
! wall 4.699621 comb 4.710000 user 4.700000 sys 0.010000 (best of 3)
$ hg perfrevlog -m -d 1000
! wall 34.252800 comb 34.250000 user 33.730000 sys 0.520000 (best of 3)
! wall 24.094999 comb 24.090000 user 23.320000 sys 0.770000 (best of 3)
Only modest wins for the changelog. But manifest reading is
significantly faster. What's going on?
One reason might be data volume. zstd decompresses faster. So given
more bytes, it will put more distance between it and zlib.
Another reason is size. In the current design, zstd revlogs are
*larger*:
debugcreatestreamclonebundle (size in bytes)
zlib: 1,638,852,492
zstd: 1,680,601,332
I haven't investigated this fully, but I reckon a significant cause of
larger revlogs is that the zstd frame/header has more bytes than
zlib's. For very small inputs or data that doesn't compress well, we'll
tend to store more uncompressed chunks than with zlib (because the
compressed size isn't smaller than original). This will make revlog
reading faster because it is doing less decompression.
Moving on to bundle performance:
$ hg bundle -a -t none-v2 (total CPU time)
zlib: 102.79s
zstd: 97.75s
So, marginal CPU decrease for reading all chunks in all revlogs
(this is somewhat disappointing).
$ hg bundle -a -t <engine>-v2 (total CPU time)
zlib: 191.59s
zstd: 115.36s
This last test effectively measures the difference between zlib->zlib
and zstd->zstd for revlogs to bundle. This is a rough approximation of
what a server does during `hg clone`.
There are some promising results for zstd. But not enough for me to
feel comfortable advertising it to users. We'll get there...
from __future__ import absolute_import, print_function
import binascii
import getopt
import math
import os
import random
import sys
import time
from mercurial.node import nullrev
from mercurial import (
ancestor,
debugcommands,
hg,
ui as uimod,
util,
)
def buildgraph(rng, nodes=100, rootprob=0.05, mergeprob=0.2, prevprob=0.7):
'''nodes: total number of nodes in the graph
rootprob: probability that a new node (not 0) will be a root
mergeprob: probability that, excluding a root a node will be a merge
prevprob: probability that p1 will be the previous node
return value is a graph represented as an adjacency list.
'''
graph = [None] * nodes
for i in xrange(nodes):
if i == 0 or rng.random() < rootprob:
graph[i] = [nullrev]
elif i == 1:
graph[i] = [0]
elif rng.random() < mergeprob:
if i == 2 or rng.random() < prevprob:
# p1 is prev
p1 = i - 1
else:
p1 = rng.randrange(i - 1)
p2 = rng.choice(range(0, p1) + range(p1 + 1, i))
graph[i] = [p1, p2]
elif rng.random() < prevprob:
graph[i] = [i - 1]
else:
graph[i] = [rng.randrange(i - 1)]
return graph
def buildancestorsets(graph):
ancs = [None] * len(graph)
for i in xrange(len(graph)):
ancs[i] = set([i])
if graph[i] == [nullrev]:
continue
for p in graph[i]:
ancs[i].update(ancs[p])
return ancs
class naiveincrementalmissingancestors(object):
def __init__(self, ancs, bases):
self.ancs = ancs
self.bases = set(bases)
def addbases(self, newbases):
self.bases.update(newbases)
def removeancestorsfrom(self, revs):
for base in self.bases:
if base != nullrev:
revs.difference_update(self.ancs[base])
revs.discard(nullrev)
def missingancestors(self, revs):
res = set()
for rev in revs:
if rev != nullrev:
res.update(self.ancs[rev])
for base in self.bases:
if base != nullrev:
res.difference_update(self.ancs[base])
return sorted(res)
def test_missingancestors(seed, rng):
# empirically observed to take around 1 second
graphcount = 100
testcount = 10
inccount = 10
nerrs = [0]
# the default mu and sigma give us a nice distribution of mostly
# single-digit counts (including 0) with some higher ones
def lognormrandom(mu, sigma):
return int(math.floor(rng.lognormvariate(mu, sigma)))
def samplerevs(nodes, mu=1.1, sigma=0.8):
count = min(lognormrandom(mu, sigma), len(nodes))
return rng.sample(nodes, count)
def err(seed, graph, bases, seq, output, expected):
if nerrs[0] == 0:
print('seed:', hex(seed)[:-1], file=sys.stderr)
if gerrs[0] == 0:
print('graph:', graph, file=sys.stderr)
print('* bases:', bases, file=sys.stderr)
print('* seq: ', seq, file=sys.stderr)
print('* output: ', output, file=sys.stderr)
print('* expected:', expected, file=sys.stderr)
nerrs[0] += 1
gerrs[0] += 1
for g in xrange(graphcount):
graph = buildgraph(rng)
ancs = buildancestorsets(graph)
gerrs = [0]
for _ in xrange(testcount):
# start from nullrev to include it as a possibility
graphnodes = range(nullrev, len(graph))
bases = samplerevs(graphnodes)
# fast algorithm
inc = ancestor.incrementalmissingancestors(graph.__getitem__, bases)
# reference slow algorithm
naiveinc = naiveincrementalmissingancestors(ancs, bases)
seq = []
revs = []
for _ in xrange(inccount):
if rng.random() < 0.2:
newbases = samplerevs(graphnodes)
seq.append(('addbases', newbases))
inc.addbases(newbases)
naiveinc.addbases(newbases)
if rng.random() < 0.4:
# larger set so that there are more revs to remove from
revs = samplerevs(graphnodes, mu=1.5)
seq.append(('removeancestorsfrom', revs))
hrevs = set(revs)
rrevs = set(revs)
inc.removeancestorsfrom(hrevs)
naiveinc.removeancestorsfrom(rrevs)
if hrevs != rrevs:
err(seed, graph, bases, seq, sorted(hrevs),
sorted(rrevs))
else:
revs = samplerevs(graphnodes)
seq.append(('missingancestors', revs))
h = inc.missingancestors(revs)
r = naiveinc.missingancestors(revs)
if h != r:
err(seed, graph, bases, seq, h, r)
# graph is a dict of child->parent adjacency lists for this graph:
# o 13
# |
# | o 12
# | |
# | | o 11
# | | |\
# | | | | o 10
# | | | | |
# | o---+ | 9
# | | | | |
# o | | | | 8
# / / / /
# | | o | 7
# | | | |
# o---+ | 6
# / / /
# | | o 5
# | |/
# | o 4
# | |
# o | 3
# | |
# | o 2
# |/
# o 1
# |
# o 0
graph = {0: [-1], 1: [0], 2: [1], 3: [1], 4: [2], 5: [4], 6: [4],
7: [4], 8: [-1], 9: [6, 7], 10: [5], 11: [3, 7], 12: [9],
13: [8]}
def genlazyancestors(revs, stoprev=0, inclusive=False):
print(("%% lazy ancestor set for %s, stoprev = %s, inclusive = %s" %
(revs, stoprev, inclusive)))
return ancestor.lazyancestors(graph.get, revs, stoprev=stoprev,
inclusive=inclusive)
def printlazyancestors(s, l):
print('membership: %r' % [n for n in l if n in s])
print('iteration: %r' % list(s))
def test_lazyancestors():
# Empty revs
s = genlazyancestors([])
printlazyancestors(s, [3, 0, -1])
# Standard example
s = genlazyancestors([11, 13])
printlazyancestors(s, [11, 13, 7, 9, 8, 3, 6, 4, 1, -1, 0])
# Standard with ancestry in the initial set (1 is ancestor of 3)
s = genlazyancestors([1, 3])
printlazyancestors(s, [1, -1, 0])
# Including revs
s = genlazyancestors([11, 13], inclusive=True)
printlazyancestors(s, [11, 13, 7, 9, 8, 3, 6, 4, 1, -1, 0])
# Test with stoprev
s = genlazyancestors([11, 13], stoprev=6)
printlazyancestors(s, [11, 13, 7, 9, 8, 3, 6, 4, 1, -1, 0])
s = genlazyancestors([11, 13], stoprev=6, inclusive=True)
printlazyancestors(s, [11, 13, 7, 9, 8, 3, 6, 4, 1, -1, 0])
# The C gca algorithm requires a real repo. These are textual descriptions of
# DAGs that have been known to be problematic.
dagtests = [
'+2*2*2/*3/2',
'+3*3/*2*2/*4*4/*4/2*4/2*2',
]
def test_gca():
u = uimod.ui.load()
for i, dag in enumerate(dagtests):
repo = hg.repository(u, 'gca%d' % i, create=1)
cl = repo.changelog
if not util.safehasattr(cl.index, 'ancestors'):
# C version not available
return
debugcommands.debugbuilddag(u, repo, dag)
# Compare the results of the Python and C versions. This does not
# include choosing a winner when more than one gca exists -- we make
# sure both return exactly the same set of gcas.
for a in cl:
for b in cl:
cgcas = sorted(cl.index.ancestors(a, b))
pygcas = sorted(ancestor.ancestors(cl.parentrevs, a, b))
if cgcas != pygcas:
print("test_gca: for dag %s, gcas for %d, %d:"
% (dag, a, b))
print(" C returned: %s" % cgcas)
print(" Python returned: %s" % pygcas)
def main():
seed = None
opts, args = getopt.getopt(sys.argv[1:], 's:', ['seed='])
for o, a in opts:
if o in ('-s', '--seed'):
seed = long(a, base=0) # accepts base 10 or 16 strings
if seed is None:
try:
seed = long(binascii.hexlify(os.urandom(16)), 16)
except AttributeError:
seed = long(time.time() * 1000)
rng = random.Random(seed)
test_missingancestors(seed, rng)
test_lazyancestors()
test_gca()
if __name__ == '__main__':
main()