changegroup: compute seen files as changesets are added (
issue4750)
Before this patch, addchangegroup() would walk the changelog and compute
the set of seen files between applying changesets and applying
manifests. When cloning large repositories such as mozilla-central,
this consumed a non-trivial amount of time. On my MBP, this walk takes
~10s. On a dainty EC2 instance, this was measured to take ~125s! On the
latter machine, this delay was enough for the Mercurial server to
disconnect the client, thinking it had timed out, thus causing a clone
to abort.
This patch enables the changelog to compute the set of changed files as
new revisions are added. By doing so, we:
* avoid a potentially heavy computation between changelog and manifest
processing by spreading the computation across all changelog additions
* avoid extra reads from the changelog by operating on the data as it is
added
The downside of this is that the add revision callback does result in
extra I/O. Before, we would perform a flush (and subsequent read to
construct the full revision) when new delta chains were created. For
changelogs, this is typically every 2-4 revisions. Using the callback
guarantees there will be a flush after every added revision *and* an
open + read of the changelog to obtain the full revision in order to
read the added files. So, this increases the frequency of these
operations by the average chain length. In the future, the revlog
should be smart enough to know how to read revisions that haven't been
flushed yet, thus eliminating this extra I/O.
On my MBP, the total CPU times for an `hg unbundle` with a local
mozilla-central gzip bundle containing 251,934 changesets and 211,065
files did not have a statistically significant change with this patch,
holding steady around 360s. So, the increased revlog flushing did not
have an effect.
With this patch, there is no longer a visible pause between applying
changeset and manifest data. Before, it sure felt like Mercurial was
lethargic making this transition. Now, the transition is nearly
instantaneous, giving the impression that Mercurial is faster. Of course,
eliminating this pause means that the potential for network disconnect due
to channel inactivity during the changelog walk is eliminated as well.
And that is the impetus behind this change.
import sys, os, subprocess
if subprocess.call(['python', '%s/hghave' % os.environ['TESTDIR'],
'cacheable']):
sys.exit(80)
from mercurial import util, scmutil, extensions, hg, ui
filecache = scmutil.filecache
class fakerepo(object):
def __init__(self):
self._filecache = {}
def join(self, p):
return p
def sjoin(self, p):
return p
@filecache('x', 'y')
def cached(self):
print 'creating'
return 'string from function'
def invalidate(self):
for k in self._filecache:
try:
delattr(self, k)
except AttributeError:
pass
def basic(repo):
print "* neither file exists"
# calls function
repo.cached
repo.invalidate()
print "* neither file still exists"
# uses cache
repo.cached
# create empty file
f = open('x', 'w')
f.close()
repo.invalidate()
print "* empty file x created"
# should recreate the object
repo.cached
f = open('x', 'w')
f.write('a')
f.close()
repo.invalidate()
print "* file x changed size"
# should recreate the object
repo.cached
repo.invalidate()
print "* nothing changed with either file"
# stats file again, reuses object
repo.cached
# atomic replace file, size doesn't change
# hopefully st_mtime doesn't change as well so this doesn't use the cache
# because of inode change
f = scmutil.opener('.')('x', 'w', atomictemp=True)
f.write('b')
f.close()
repo.invalidate()
print "* file x changed inode"
repo.cached
# create empty file y
f = open('y', 'w')
f.close()
repo.invalidate()
print "* empty file y created"
# should recreate the object
repo.cached
f = open('y', 'w')
f.write('A')
f.close()
repo.invalidate()
print "* file y changed size"
# should recreate the object
repo.cached
f = scmutil.opener('.')('y', 'w', atomictemp=True)
f.write('B')
f.close()
repo.invalidate()
print "* file y changed inode"
repo.cached
f = scmutil.opener('.')('x', 'w', atomictemp=True)
f.write('c')
f.close()
f = scmutil.opener('.')('y', 'w', atomictemp=True)
f.write('C')
f.close()
repo.invalidate()
print "* both files changed inode"
repo.cached
def fakeuncacheable():
def wrapcacheable(orig, *args, **kwargs):
return False
def wrapinit(orig, *args, **kwargs):
pass
originit = extensions.wrapfunction(util.cachestat, '__init__', wrapinit)
origcacheable = extensions.wrapfunction(util.cachestat, 'cacheable',
wrapcacheable)
for fn in ['x', 'y']:
try:
os.remove(fn)
except OSError:
pass
basic(fakerepo())
util.cachestat.cacheable = origcacheable
util.cachestat.__init__ = originit
def test_filecache_synced():
# test old behaviour that caused filecached properties to go out of sync
os.system('hg init && echo a >> a && hg ci -qAm.')
repo = hg.repository(ui.ui())
# first rollback clears the filecache, but changelog to stays in __dict__
repo.rollback()
repo.commit('.')
# second rollback comes along and touches the changelog externally
# (file is moved)
repo.rollback()
# but since changelog isn't under the filecache control anymore, we don't
# see that it changed, and return the old changelog without reconstructing
# it
repo.commit('.')
def setbeforeget(repo):
os.remove('x')
os.remove('y')
repo.cached = 'string set externally'
repo.invalidate()
print "* neither file exists"
print repo.cached
repo.invalidate()
f = open('x', 'w')
f.write('a')
f.close()
print "* file x created"
print repo.cached
repo.cached = 'string 2 set externally'
repo.invalidate()
print "* string set externally again"
print repo.cached
repo.invalidate()
f = open('y', 'w')
f.write('b')
f.close()
print "* file y created"
print repo.cached
print 'basic:'
print
basic(fakerepo())
print
print 'fakeuncacheable:'
print
fakeuncacheable()
test_filecache_synced()
print
print 'setbeforeget:'
print
setbeforeget(fakerepo())