bundle2: increase payload part chunk size to 32kb
Bundle2 payload parts are framed chunks. Esentially, we obtain
data in equal size chunks of size `preferedchunksize` and emit those
to a generator. That generator is fed into a compressor (which can
be the no-op compressor, which just re-emits the generator). And
the output from the compressor likely goes to a file descriptor
or socket.
What this means is that small chunk sizes create more Python objects
and Python function calls than larger chunk sizes. And as we know,
Python object and function call overhead in performance sensitive
code matters (at least with CPython).
This commit increases the bundle2 part payload chunk size from 4k
to 32k. Practically speaking, this means that the chunks we feed
into a compressor (implemented in C code) or feed directly into a
file handle or socket write() are larger. It's possible the chunks
might be larger than what the receiver can handle in one logical
operation. But at that point, we're in C code, which is much more
efficient at dealing with splitting up the chunk and making multiple
function calls than Python is.
A downside to larger chunks is that the receiver has to wait for that
much data to arrive (either raw or from a decompressor) before it
can process the chunk. But 32kb still feels like a small buffer to
have to wait for. And in many cases, the client will convert from
8 read(4096) to 1 read(32768). That's happening in Python land. So
we cut down on the number of Python objects and function calls,
making the client faster as well. I don't think there are any
significant concerns to increasing the payload chunk size to 32kb.
The impact of this change on performance significant. Using `curl`
to obtain a stream clone bundle2 payload from a server on localhost
serving the mozilla-unified repository:
before: 20.78 user; 7.71 system; 80.5 MB/s
after: 13.90 user; 3.51 system; 132 MB/s
legacy: 9.72 user; 8.16 system; 132 MB/s
bundle2 stream clone generation is still more resource intensive than
legacy stream clone (that's likely because of the use of a
util.chunkbuffer). But the throughput is the same. We might
be in territory we're this is effectively a benchmark of the
networking stack or Python's syscall throughput.
From the client perspective, `hg clone -U --stream`:
before: 33.50 user; 7.95 system; 53.3 MB/s
after: 22.82 user; 7.33 system; 72.7 MB/s
legacy: 29.96 user; 7.94 system; 58.0 MB/s
And for `hg clone --stream` with a working directory update of
~230k files:
after: 119.55 user; 26.47 system; 0:57.08 wall
legacy: 126.98 user; 26.94 system; 1:05.56 wall
So, it appears that bundle2's stream clone is now definitively faster
than legacy stream clone!
Differential Revision: https://phab.mercurial-scm.org/D1932
$ cat >> $HGRCPATH <<EOF
> [extensions]
> convert=
> [convert]
> hg.saverev=False
> EOF
$ hg init orig
$ cd orig
$ echo foo > foo
$ echo bar > bar
$ hg ci -qAm 'add foo bar' -d '0 0'
$ echo >> foo
$ hg ci -m 'change foo' -d '1 0'
$ hg up -qC 0
$ hg copy --after --force foo bar
$ hg copy foo baz
$ hg ci -m 'make bar and baz copies of foo' -d '2 0'
created new head
Test that template can print all file copies (issue4362)
$ hg log -r . --template "{file_copies % ' File: {file_copy}\n'}"
File: bar (foo)
File: baz (foo)
$ hg bookmark premerge1
$ hg merge -r 1
merging baz and foo to baz
1 files updated, 1 files merged, 0 files removed, 0 files unresolved
(branch merge, don't forget to commit)
$ hg ci -m 'merge local copy' -d '3 0'
$ hg up -C 1
1 files updated, 0 files merged, 1 files removed, 0 files unresolved
(leaving bookmark premerge1)
$ hg bookmark premerge2
$ hg merge 2
merging foo and baz to baz
1 files updated, 1 files merged, 0 files removed, 0 files unresolved
(branch merge, don't forget to commit)
$ hg ci -m 'merge remote copy' -d '4 0'
created new head
Make and delete some tags
$ hg tag that
$ hg tag --remove that
$ hg tag this
#if execbit
$ chmod +x baz
#else
$ echo some other change to make sure we get a rev 5 > baz
#endif
$ hg ci -m 'mark baz executable' -d '5 0'
$ cd ..
$ hg convert --datesort orig new 2>&1 | grep -v 'subversion python bindings could not be loaded'
initializing destination new repository
scanning source...
sorting...
converting...
8 add foo bar
7 change foo
6 make bar and baz copies of foo
5 merge local copy
4 merge remote copy
3 Added tag that for changeset 88586c4e9f02
2 Removed tag that
1 Added tag this for changeset c56a7f387039
0 mark baz executable
updating bookmarks
$ cd new
$ hg out ../orig
comparing with ../orig
searching for changes
no changes found
[1]
#if execbit
$ hg bookmarks
premerge1 3:973ef48a98a4
premerge2 8:91d107c423ba
#else
Different hash because no x bit
$ hg bookmarks
premerge1 3:973ef48a98a4
premerge2 8:3537b15eaaca
#endif
Test that redoing a convert results in an identical graph
$ cd ../
$ rm new/.hg/shamap
$ hg convert --datesort orig new 2>&1 | grep -v 'subversion python bindings could not be loaded'
scanning source...
sorting...
converting...
8 add foo bar
7 change foo
6 make bar and baz copies of foo
5 merge local copy
4 merge remote copy
3 Added tag that for changeset 88586c4e9f02
2 Removed tag that
1 Added tag this for changeset c56a7f387039
0 mark baz executable
updating bookmarks
$ hg -R new log -G -T '{rev} {desc}'
o 8 mark baz executable
|
o 7 Added tag this for changeset c56a7f387039
|
o 6 Removed tag that
|
o 5 Added tag that for changeset 88586c4e9f02
|
o 4 merge remote copy
|\
+---o 3 merge local copy
| |/
| o 2 make bar and baz copies of foo
| |
o | 1 change foo
|/
o 0 add foo bar
check shamap LF and CRLF handling
$ cat > rewrite.py <<EOF
> import sys
> # Interlace LF and CRLF
> lines = [(l.rstrip() + ((i % 2) and '\n' or '\r\n'))
> for i, l in enumerate(file(sys.argv[1]))]
> file(sys.argv[1], 'wb').write(''.join(lines))
> EOF
$ $PYTHON rewrite.py new/.hg/shamap
$ cd orig
$ hg up -qC 1
$ echo foo >> foo
$ hg ci -qm 'change foo again'
$ hg up -qC 2
$ echo foo >> foo
$ hg ci -qm 'change foo again again'
$ cd ..
$ hg convert --datesort orig new 2>&1 | grep -v 'subversion python bindings could not be loaded'
scanning source...
sorting...
converting...
1 change foo again again
0 change foo again
updating bookmarks
init broken repository
$ hg init broken
$ cd broken
$ echo a >> a
$ echo b >> b
$ hg ci -qAm init
$ echo a >> a
$ echo b >> b
$ hg copy b c
$ hg ci -qAm changeall
$ hg up -qC 0
$ echo bc >> b
$ hg ci -m changebagain
created new head
$ HGMERGE=internal:local hg -q merge
$ hg ci -m merge
$ hg mv b d
$ hg ci -m moveb
break it
$ rm .hg/store/data/b.*
$ cd ..
$ hg --config convert.hg.ignoreerrors=True convert broken fixed
initializing destination fixed repository
scanning source...
sorting...
converting...
4 init
ignoring: data/b.i@1e88685f5dde: no match found
3 changeall
2 changebagain
1 merge
0 moveb
$ hg -R fixed verify
checking changesets
checking manifests
crosschecking files in changesets and manifests
checking files
3 files, 5 changesets, 5 total revisions
manifest -r 0
$ hg -R fixed manifest -r 0
a
manifest -r tip
$ hg -R fixed manifest -r tip
a
c
d