view tests/test-hardlinks.t @ 51681:522b4d729e89

mmap: populate the mapping by default

Without pre-population, accessing all data through a mmap can result in many
pagefault, reducing performance significantly. If the mmap is prepopulated, the
performance can no longer get slower than a full read.
(See benchmark number below)

In some cases were very few data is read, prepopulating can be overkill and
slower than populating on access (through page fault). So that behavior can be
controlled when the caller can pre-determine the best behavior.
(See benchmark number below)

In addition, testing with populating in a secondary thread yield great result
combining the best of each approach. This might be implemented in later
changesets.

In all cases, using mmap has a great effect on memory usage when many processes
run in parallel on the same machine.

### Benchmarks

# What did I run

A couple of month back I ran a large benchmark campaign to assess the impact of
various approach for using mmap with the revlog (and other files), it
highlighted a few benchmarks that capture the impact of the changes well. So to
validate this change I checked the following:

- log command displaying various revisions
  (read the changelog index)
- log command displaying the patch of listed revisions
  (read the changelog index, the manifest index and a few files indexes)
- unbundling a few revisions
  (read and write changelog, manifest and few files indexes, and walk the graph
  to update some cache)
- pushing a few revisions
  (read and write changelog, manifest and few files indexes, walk the graph to
  update some cache, performs various accesses locally and remotely during
  discovery)

Benchmarks were run using the default module policy (c+py) and the rust one. No
significant difference were found between the two implementation, so we will
present result using the default policy (unless otherwise specified).

I ran them on a few repositories :

- mercurial: a "public changeset only" copy of mercurial from 2018-08-01 using
  zstd compression and sparse-revlog

- pypy: a copy of pypy from 2018-08-01 using zstd compression and sparse-revlog

- netbeans: a copy of netbeans from 2018-08-01 using zstd compression and
  sparse-revlog

- mozilla-try: a copy of mozilla-try from 2019-02-18 using zstd compression and
  sparse-revlog

- mozilla-try persistent-nodemap: Same as the above but with a persistent
  nodemap. Used for the log --patch benchmark only

# Results

For the smaller repositories (mercurial, pypy), the impact of mmap is almost
imperceptible, other cost dominating the operation. The impact of prepopulating
is undiscernible in the benchmark we ran.

For larger repositories the benchmark support explanation given above:

On netbeans, the log can be about 1% faster without repopulation (for a
difference < 100ms) but unbundle becomes a bit slower, even when small.

### data-env-vars.name                             = netbeans-2018-08-01-zstd-sparse-revlog
  # benchmark.name                                 = hg.command.unbundle
  # benchmark.variants.issue6528                   = disabled
  # benchmark.variants.reuse-external-delta-parent = yes
  # benchmark.variants.revs                        = any-1-extra-rev
  # benchmark.variants.source                      = unbundle
  # benchmark.variants.verbosity                   = quiet
with-populate: 0.240157
no-populate:   0.265087  (+10.38%, +0.02)
  # benchmark.variants.revs                        = any-100-extra-rev
with-populate: 1.459518
no-populate:   1.481290   (+1.49%, +0.02)
 ## benchmark.name                                 = hg.command.push
  # benchmark.variants.explicit-rev                = none
  # benchmark.variants.issue6528                   = disabled
  # benchmark.variants.protocol                    = ssh
  # benchmark.variants.reuse-external-delta-parent = yes
  # benchmark.variants.revs                        = any-1-extra-rev
with-populate: 0.771919
no-populate:   0.792025   (+2.60%, +0.02)
  # benchmark.variants.revs                        = any-100-extra-rev
with-populate: 1.459518
no-populate:   1.481290   (+1.49%, +0.02)


For mozilla-try, the "slow down" from pre-populate for small `hg log` is more
visible, but still small in absolute time. (using rust value for the persistent
nodemap value to be relevant).

### data-env-vars.name            = mozilla-try-2019-02-18-ds2-pnm
  # benchmark.name                = hg.command.log
  # bin-env-vars.hg.flavor        = rust
  # benchmark.variants.patch      = yes
  # benchmark.variants.limit-rev  = 1
with-populate: 0.237813
no-populate:   0.229452   (-3.52%, -0.01)
  # benchmark.variants.limit-rev  = 10
  # benchmark.variants.patch      = yes
with-populate: 1.213578
no-populate:   1.205189
### data-env-vars.name            = mozilla-try-2019-02-18-zstd-sparse-revlog
  # benchmark.variants.limit-rev  = 1000
  # benchmark.variants.patch      = no
  # benchmark.variants.rev        = tip
with-populate: 0.198607
no-populate:   0.195038   (-1.80%, -0.00)

However pre-populating provide a significant boost on more complex operations
like unbundle or push:

### data-env-vars.name                             = mozilla-try-2019-02-18-zstd-sparse-revlog
  # benchmark.name                                 = hg.command.push
  # benchmark.variants.explicit-rev                = none
  # benchmark.variants.issue6528                   = disabled
  # benchmark.variants.protocol                    = ssh
  # benchmark.variants.reuse-external-delta-parent = yes
  # benchmark.variants.revs                        = any-1-extra-rev
with-populate: 4.798632
no-populate:   4.953295   (+3.22%, +0.15)
  # benchmark.variants.revs                        = any-100-extra-rev
with-populate: 4.903618
no-populate:   5.014963   (+2.27%, +0.11)
 ## benchmark.name                                 = hg.command.unbundle
  # benchmark.variants.revs                        = any-1-extra-rev
with-populate: 1.423411
no-populate:   1.585365  (+11.38%, +0.16)
  # benchmark.variants.revs                        = any-100-extra-rev
with-populate: 1.537909
no-populate:   1.688489   (+9.79%, +0.15)

#require hardlink reporevlogstore

  $ cat > nlinks.py <<EOF
  > import sys
  > from mercurial import pycompat, util
  > for f in sorted(sys.stdin.readlines()):
  >     f = f[:-1]
  >     print(util.nlinks(pycompat.fsencode(f)), f)
  > EOF

  $ nlinksdir()
  > {
  >     find "$@" -type f | "$PYTHON" $TESTTMP/nlinks.py
  > }

Some implementations of cp can't create hardlinks (replaces 'cp -al' on Linux):

  $ cat > linkcp.py <<EOF
  > import sys
  > from mercurial import pycompat, util
  > util.copyfiles(pycompat.fsencode(sys.argv[1]),
  >                pycompat.fsencode(sys.argv[2]), hardlink=True)
  > EOF

  $ linkcp()
  > {
  >     "$PYTHON" $TESTTMP/linkcp.py $1 $2
  > }

Prepare repo r1:

  $ hg init r1
  $ cd r1

  $ echo c1 > f1
  $ hg add f1
  $ hg ci -m0

  $ mkdir d1
  $ cd d1
  $ echo c2 > f2
  $ hg add f2
  $ hg ci -m1
  $ cd ../..

  $ nlinksdir r1/.hg/store
  1 r1/.hg/store/00changelog-b870a51b.nd (rust !)
  1 r1/.hg/store/00changelog.d
  1 r1/.hg/store/00changelog.i
  1 r1/.hg/store/00changelog.n (rust !)
  1 r1/.hg/store/00manifest.i
  1 r1/.hg/store/data/d1/f2.i
  1 r1/.hg/store/data/f1.i
  1 r1/.hg/store/fncache (repofncache !)
  1 r1/.hg/store/phaseroots
  1 r1/.hg/store/requires
  1 r1/.hg/store/undo
  1 r1/.hg/store/undo.backup.00changelog.n.bck (rust !)
  1 r1/.hg/store/undo.backup.fncache.bck (repofncache !)
  1 r1/.hg/store/undo.backupfiles


Create hardlinked clone r2:

  $ hg clone -U --debug r1 r2 --config progress.debug=true
  linking: 1/8 files (12.50%) (no-rust !)
  linking: 2/8 files (25.00%) (no-rust !)
  linking: 3/8 files (37.50%) (no-rust !)
  linking: 4/8 files (50.00%) (no-rust !)
  linking: 5/8 files (62.50%) (no-rust !)
  linking: 6/8 files (75.00%) (no-rust !)
  linking: 7/8 files (87.50%) (no-rust !)
  linking: 8/8 files (100.00%) (no-rust !)
  linked 8 files (no-rust !)
  linking: 1/10 files (10.00%) (rust !)
  linking: 2/10 files (20.00%) (rust !)
  linking: 3/10 files (30.00%) (rust !)
  linking: 4/10 files (40.00%) (rust !)
  linking: 5/10 files (50.00%) (rust !)
  linking: 6/10 files (60.00%) (rust !)
  linking: 7/10 files (70.00%) (rust !)
  linking: 8/10 files (80.00%) (rust !)
  linking: 9/10 files (90.00%) (rust !)
  linking: 10/10 files (100.00%) (rust !)
  linked 10 files (rust !)
  updating the branch cache

Create non-hardlinked clone r3:

  $ hg clone --pull r1 r3
  requesting all changes
  adding changesets
  adding manifests
  adding file changes
  added 2 changesets with 2 changes to 2 files
  new changesets 40d85e9847f2:7069c422939c
  updating to branch default
  2 files updated, 0 files merged, 0 files removed, 0 files unresolved


Repos r1 and r2 should now contain hardlinked files:

  $ nlinksdir r1/.hg/store
  1 r1/.hg/store/00changelog-b870a51b.nd (rust !)
  2 r1/.hg/store/00changelog.d
  2 r1/.hg/store/00changelog.i
  1 r1/.hg/store/00changelog.n (rust !)
  2 r1/.hg/store/00manifest.i
  2 r1/.hg/store/data/d1/f2.i
  2 r1/.hg/store/data/f1.i
  1 r1/.hg/store/fncache (repofncache !)
  1 r1/.hg/store/phaseroots
  1 r1/.hg/store/requires
  1 r1/.hg/store/undo
  1 r1/.hg/store/undo.backup.00changelog.n.bck (rust !)
  1 r1/.hg/store/undo.backup.fncache.bck (repofncache !)
  1 r1/.hg/store/undo.backupfiles

  $ nlinksdir r2/.hg/store
  1 r2/.hg/store/00changelog-b870a51b.nd (rust !)
  2 r2/.hg/store/00changelog.d
  2 r2/.hg/store/00changelog.i
  1 r2/.hg/store/00changelog.n (rust !)
  2 r2/.hg/store/00manifest.i
  2 r2/.hg/store/data/d1/f2.i
  2 r2/.hg/store/data/f1.i
  1 r2/.hg/store/fncache (repofncache !)
  1 r2/.hg/store/requires

Repo r3 should not be hardlinked:

  $ nlinksdir r3/.hg/store
  1 r3/.hg/store/00changelog-88698448.nd (rust !)
  1 r3/.hg/store/00changelog.d
  1 r3/.hg/store/00changelog.i
  1 r3/.hg/store/00changelog.n (rust !)
  1 r3/.hg/store/00manifest.i
  1 r3/.hg/store/data/d1/f2.i
  1 r3/.hg/store/data/f1.i
  1 r3/.hg/store/fncache (repofncache !)
  1 r3/.hg/store/phaseroots
  1 r3/.hg/store/requires
  1 r3/.hg/store/undo
  1 r3/.hg/store/undo.backupfiles


Create a non-inlined filelog in r3:

  $ cd r3/d1
  >>> f = open('data1', 'wb')
  >>> for x in range(10000):
  ...     f.write(b"%d\n" % x) and None
  >>> f.close()
  $ for j in 0 1 2 3 4 5 6 7 8 9; do
  >   cat data1 >> f2
  >   hg commit -m$j
  > done
  $ cd ../..

  $ nlinksdir r3/.hg/store
  1 r3/.hg/store/00changelog-ea337809.nd (rust !)
  1 r3/.hg/store/00changelog.d
  1 r3/.hg/store/00changelog.i
  1 r3/.hg/store/00changelog.n (rust !)
  1 r3/.hg/store/00manifest.i
  1 r3/.hg/store/data/d1/f2.d
  1 r3/.hg/store/data/d1/f2.i
  1 r3/.hg/store/data/f1.i
  1 r3/.hg/store/fncache (repofncache !)
  1 r3/.hg/store/phaseroots
  1 r3/.hg/store/requires
  1 r3/.hg/store/undo
  1 r3/.hg/store/undo.backup.00changelog.n.bck (rust !)
  1 r3/.hg/store/undo.backupfiles

Push to repo r1 should break up most hardlinks in r2:

  $ hg -R r2 verify -q

  $ cd r3
  $ hg push
  pushing to $TESTTMP/r1
  searching for changes
  adding changesets
  adding manifests
  adding file changes
  added 10 changesets with 10 changes to 1 files

  $ cd ..

  $ nlinksdir r2/.hg/store
  1 r2/.hg/store/00changelog-b870a51b.nd (rust !)
  1 r2/.hg/store/00changelog.d
  1 r2/.hg/store/00changelog.i
  1 r2/.hg/store/00changelog.n (rust !)
  1 r2/.hg/store/00manifest.i
  1 r2/.hg/store/data/d1/f2.i
  2 r2/.hg/store/data/f1.i
  [12] r2/\.hg/store/fncache (re) (repofncache !)
  1 r2/.hg/store/requires

#if hardlink-whitelisted repofncache
  $ nlinksdir r2/.hg/store/fncache
  1 r2/.hg/store/fncache
#endif

  $ hg -R r2 verify -q

  $ cd r1
  $ hg up
  1 files updated, 0 files merged, 0 files removed, 0 files unresolved

Committing a change to f1 in r1 must break up hardlink f1.i in r2:

  $ echo c1c1 >> f1
  $ hg ci -m00
  $ cd ..

  $ nlinksdir r2/.hg/store
  1 r2/.hg/store/00changelog-b870a51b.nd (rust !)
  1 r2/.hg/store/00changelog.d
  1 r2/.hg/store/00changelog.i
  1 r2/.hg/store/00changelog.n (rust !)
  1 r2/.hg/store/00manifest.i
  1 r2/.hg/store/data/d1/f2.i
  1 r2/.hg/store/data/f1.i
  1 r2/.hg/store/fncache (repofncache !)
  1 r2/.hg/store/requires

#if hardlink-whitelisted repofncache
  $ nlinksdir r2/.hg/store/fncache
  1 r2/.hg/store/fncache
#endif

Create a file which exec permissions we will change
  $ cd r3
  $ echo "echo hello world" > f3
  $ hg add f3
  $ hg ci -mf3
  $ cd ..

  $ cd r3
  $ hg tip --template '{rev}:{node|short}\n'
  12:d3b77733a28a
  $ echo bla > f1
  $ chmod +x f3
  $ hg ci -m1
  $ cd ..

Create hardlinked copy r4 of r3 (on Linux, we would call 'cp -al'):

  $ linkcp r3 r4

'checklink' is produced by hardlinking a symlink, which is undefined whether
the symlink should be followed or not. It does behave differently on Linux and
BSD. Just remove it so the test pass on both platforms.

  $ rm -f r4/.hg/wcache/checklink

r4 has hardlinks in the working dir (not just inside .hg):

  $ nlinksdir r4
  2 r4/.hg/00changelog.i
  [24] r4/.hg/branch (re)
  2 r4/.hg/cache/branch2-base
  2 r4/.hg/cache/branch2-served
  2 r4/.hg/cache/rbc-names-v1
  2 r4/.hg/cache/rbc-revs-v1
  2 r4/.hg/cache/tags2
  2 r4/.hg/cache/tags2-served
  2 r4/.hg/dirstate
  2 r4/.hg/fsmonitor.state (fsmonitor !)
  2 r4/.hg/hgrc
  2 r4/.hg/last-message.txt
  2 r4/.hg/requires
  2 r4/.hg/store/00changelog-7f2eb713.nd (rust !)
  2 r4/.hg/store/00changelog.d
  2 r4/.hg/store/00changelog.i
  2 r4/.hg/store/00changelog.n (rust !)
  2 r4/.hg/store/00manifest.i
  2 r4/.hg/store/data/d1/f2.d
  2 r4/.hg/store/data/d1/f2.i
  2 r4/.hg/store/data/f1.i
  2 r4/.hg/store/data/f3.i
  2 r4/.hg/store/fncache (repofncache !)
  2 r4/.hg/store/phaseroots
  2 r4/.hg/store/requires
  2 r4/.hg/store/undo
  2 r4/.hg/store/undo.backup.00changelog.n.bck (rust !)
  2 r4/.hg/store/undo.backupfiles
  [24] r4/.hg/undo.backup.branch.bck (re)
  2 r4/\.hg/undo\.backup\.dirstate.bck (re)
  2 r4/.hg/undo.desc
  2 r4/.hg/wcache/checkisexec (execbit !)
  2 r4/.hg/wcache/checklink-target (symlink !)
  2 r4/.hg/wcache/checknoexec (execbit !)
  2 r4/.hg/wcache/manifestfulltextcache (reporevlogstore !)
  2 r4/d1/data1
  2 r4/d1/f2
  2 r4/f1
  2 r4/f3

Update back to revision 12 in r4 should break hardlink of file f1 and f3:
#if hardlink-whitelisted
  $ nlinksdir r4/.hg/undo.backup.dirstate.bck r4/.hg/dirstate
  2 r4/.hg/dirstate
  2 r4/.hg/undo.backup.dirstate.bck
#endif


  $ hg -R r4 up 12
  2 files updated, 0 files merged, 0 files removed, 0 files unresolved (execbit !)
  1 files updated, 0 files merged, 0 files removed, 0 files unresolved (no-execbit !)

  $ nlinksdir r4
  2 r4/.hg/00changelog.i
  1 r4/.hg/branch
  2 r4/.hg/cache/branch2-base
  2 r4/.hg/cache/branch2-served
  2 r4/.hg/cache/rbc-names-v1
  2 r4/.hg/cache/rbc-revs-v1
  2 r4/.hg/cache/tags2
  2 r4/.hg/cache/tags2-served
  1 r4/.hg/dirstate
  1 r4/.hg/fsmonitor.state (fsmonitor !)
  2 r4/.hg/hgrc
  2 r4/.hg/last-message.txt
  2 r4/.hg/requires
  2 r4/.hg/store/00changelog-7f2eb713.nd (rust !)
  2 r4/.hg/store/00changelog.d
  2 r4/.hg/store/00changelog.i
  2 r4/.hg/store/00changelog.n (rust !)
  2 r4/.hg/store/00manifest.i
  2 r4/.hg/store/data/d1/f2.d
  2 r4/.hg/store/data/d1/f2.i
  2 r4/.hg/store/data/f1.i
  2 r4/.hg/store/data/f3.i
  2 r4/.hg/store/fncache
  2 r4/.hg/store/phaseroots
  2 r4/.hg/store/requires
  2 r4/.hg/store/undo
  2 r4/.hg/store/undo.backup.00changelog.n.bck (rust !)
  2 r4/.hg/store/undo.backupfiles
  [23] r4/.hg/undo.backup.branch.bck (re)
  2 r4/\.hg/undo\.backup\.dirstate.bck (re)
  2 r4/.hg/undo.desc
  2 r4/.hg/wcache/checkisexec (execbit !)
  2 r4/.hg/wcache/checklink-target (symlink !)
  2 r4/.hg/wcache/checknoexec (execbit !)
  1 r4/.hg/wcache/manifestfulltextcache (reporevlogstore !)
  2 r4/d1/data1
  2 r4/d1/f2
  1 r4/f1
  1 r4/f3 (execbit !)
  2 r4/f3 (no-execbit !)

#if hardlink-whitelisted
  $ nlinksdir r4/.hg/undo.backup.dirstate.bck r4/.hg/dirstate
  1 r4/.hg/dirstate
  2 r4/.hg/undo.backup.dirstate.bck
#endif

Test hardlinking outside hg:

  $ mkdir x
  $ echo foo > x/a

  $ linkcp x y
  $ echo bar >> y/a

No diff if hardlink:

  $ diff x/a y/a

Test mq hardlinking:

  $ echo "[extensions]" >> $HGRCPATH
  $ echo "mq=" >> $HGRCPATH

  $ hg init a
  $ cd a

  $ hg qimport -n foo - << EOF
  > # HG changeset patch
  > # Date 1 0
  > diff -r 2588a8b53d66 a
  > --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
  > +++ b/a	Wed Jul 23 15:54:29 2008 +0200
  > @@ -0,0 +1,1 @@
  > +a
  > EOF
  adding foo to series file

  $ hg qpush
  applying foo
  now at: foo

  $ cd ..
  $ linkcp a b
  $ cd b

  $ hg qimport -n bar - << EOF
  > # HG changeset patch
  > # Date 2 0
  > diff -r 2588a8b53d66 a
  > --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
  > +++ b/b	Wed Jul 23 15:54:29 2008 +0200
  > @@ -0,0 +1,1 @@
  > +b
  > EOF
  adding bar to series file

  $ hg qpush
  applying bar
  now at: bar

  $ cat .hg/patches/status
  430ed4828a74fa4047bc816a25500f7472ab4bfe:foo
  4e7abb4840c46a910f6d7b4d3c3fc7e5209e684c:bar

  $ cat .hg/patches/series
  foo
  bar

  $ cat ../a/.hg/patches/status
  430ed4828a74fa4047bc816a25500f7472ab4bfe:foo

  $ cat ../a/.hg/patches/series
  foo

Test tags hardlinking:

  $ hg qdel -r qbase:qtip
  patch foo finalized without changeset message
  patch bar finalized without changeset message

  $ hg tag -l lfoo
  $ hg tag foo

  $ cd ..
  $ linkcp b c
  $ cd c

  $ hg tag -l -r 0 lbar
  $ hg tag -r 0 bar

  $ cat .hgtags
  4e7abb4840c46a910f6d7b4d3c3fc7e5209e684c foo
  430ed4828a74fa4047bc816a25500f7472ab4bfe bar

  $ cat .hg/localtags
  4e7abb4840c46a910f6d7b4d3c3fc7e5209e684c lfoo
  430ed4828a74fa4047bc816a25500f7472ab4bfe lbar

  $ cat ../b/.hgtags
  4e7abb4840c46a910f6d7b4d3c3fc7e5209e684c foo

  $ cat ../b/.hg/localtags
  4e7abb4840c46a910f6d7b4d3c3fc7e5209e684c lfoo

  $ cd ..