wireproto: compress data from a generator
Currently, the "getbundle" wire protocol command obtains a generator of
data, converts it to a util.chunkbuffer, then converts it back to a
generator via the protocol's groupchunks() implementation. For the SSH
protocol, groupchunks() simply reads 4kb chunks then write()s the
data to a file descriptor. For the HTTP protocol, groupchunks() reads
32kb chunks, feeds those into a zlib compressor, emits compressed data
as it is available, and that is sent to the WSGI layer, where it is
likely turned into HTTP chunked transfer chunks as is or further
buffered and turned into a larger chunk.
For both the SSH and HTTP protocols, there is inefficiency from using
util.chunkbuffer.
For SSH, emitting consistent 4kb chunks sounds nice. However, the file
descriptor it is writing to is almost certainly buffered. That means
that a Python .write() probably doesn't translate into exactly what is
written to the I/O layer.
For HTTP, we're going through an intermediate layer to zlib compress
data. So all util.chunkbuffer is doing is ensuring that the chunks we
feed into the zlib compressor are of uniform size. This means more CPU
time in Python buffering and emitting chunks in util.chunkbuffer but
fewer function calls to zlib.
This patch introduces and implements a new wire protocol abstract
method: compresschunks(). It is like groupchunks() except it operates
on a generator instead of something with a .read(). The SSH
implementation simply proxies chunks. The HTTP implementation uses
zlib compression.
To avoid duplicate code, the HTTP groupchunks() has been reimplemented
in terms of compresschunks().
To prove this all works, the "getbundle" wire protocol command has been
switched to compresschunks(). This removes the util.chunkbuffer from
that command. Now, data essentially streams straight from the
changegroup emitter to the wire, possibly through a zlib compressor.
Generators all the way, baby.
There were slim to no performance changes on the server as measured
with the mozilla-central repository. This is likely because CPU
time is dominated by reading revlogs, producing the changegroup, and
zlib compressing the output stream. Still, this brings us a little
closer to our ideal of using generators everywhere.
$ cat >> $HGRCPATH <<EOF
> [extensions]
> rebase=
> mq=
>
> [mq]
> plain=true
>
> [alias]
> tglog = log -G --template "{rev}: '{desc}' tags: {tags}\n"
> EOF
$ hg init a
$ cd a
$ hg qinit -c
$ echo c1 > f
$ hg add f
$ hg ci -m C1
$ echo r1 > f
$ hg ci -m R1
$ hg up -q 0
$ hg qnew f.patch -d '1 0'
$ echo mq1 > f
$ hg qref -m P0
$ hg qnew f2.patch
$ echo mq2 > f
$ hg qref -m P1 -d '2 0'
$ hg tglog
@ 3: 'P1' tags: f2.patch qtip tip
|
o 2: 'P0' tags: f.patch qbase
|
| o 1: 'R1' tags:
|/
o 0: 'C1' tags: qparent
Rebase - try to rebase on an applied mq patch:
$ hg rebase -s 1 -d 3
abort: cannot rebase onto an applied mq patch
[255]
Rebase - same thing, but mq patch is default dest:
$ hg up -q 1
$ hg rebase
abort: cannot rebase onto an applied mq patch
[255]
$ hg up -q qtip
Rebase - generate a conflict:
$ hg rebase -s 2 -d 1
rebasing 2:3504f44bffc0 "P0" (f.patch qbase)
merging f
warning: conflicts while merging f! (edit, then use 'hg resolve --mark')
unresolved conflicts (see hg resolve, then hg rebase --continue)
[1]
Fix the 1st conflict:
$ echo mq1r1 > f
$ hg resolve -m f
(no more unresolved files)
continue: hg rebase --continue
$ hg rebase -c
rebasing 2:3504f44bffc0 "P0" (f.patch qbase)
rebasing 3:929394423cd3 "P1" (f2.patch qtip tip)
merging f
warning: conflicts while merging f! (edit, then use 'hg resolve --mark')
unresolved conflicts (see hg resolve, then hg rebase --continue)
[1]
Fix the 2nd conflict:
$ echo mq1r1mq2 > f
$ hg resolve -m f
(no more unresolved files)
continue: hg rebase --continue
$ hg rebase -c
already rebased 2:3504f44bffc0 "P0" (f.patch qbase) as ebe9914c0d1c
rebasing 3:929394423cd3 "P1" (f2.patch qtip)
saved backup bundle to $TESTTMP/a/.hg/strip-backup/3504f44bffc0-30595b40-backup.hg (glob)
$ hg tglog
@ 3: 'P1' tags: f2.patch qtip tip
|
o 2: 'P0' tags: f.patch qbase
|
o 1: 'R1' tags: qparent
|
o 0: 'C1' tags:
$ hg up -q qbase
$ cat f
mq1r1
$ cat .hg/patches/f.patch
# HG changeset patch
# User test
# Date 1 0
# Thu Jan 01 00:00:01 1970 +0000
# Node ID ebe9914c0d1c3f60096e952fa4dbb3d377dea3ab
# Parent bac9ed9960d8992bcad75864a879fa76cadaf1b0
P0
diff -r bac9ed9960d8 -r ebe9914c0d1c f
--- a/f Thu Jan 01 00:00:00 1970 +0000
+++ b/f Thu Jan 01 00:00:01 1970 +0000
@@ -1,1 +1,1 @@
-r1
+mq1r1
Update to qtip:
$ hg up -q qtip
$ cat f
mq1r1mq2
$ cat .hg/patches/f2.patch
# HG changeset patch
# User test
# Date 2 0
# Thu Jan 01 00:00:02 1970 +0000
# Node ID 462012cf340c97d44d62377c985a423f6bb82f07
# Parent ebe9914c0d1c3f60096e952fa4dbb3d377dea3ab
P1
diff -r ebe9914c0d1c -r 462012cf340c f
--- a/f Thu Jan 01 00:00:01 1970 +0000
+++ b/f Thu Jan 01 00:00:02 1970 +0000
@@ -1,1 +1,1 @@
-mq1r1
+mq1r1mq2
Adding one git-style patch and one normal:
$ hg qpop -a
popping f2.patch
popping f.patch
patch queue now empty
$ rm -fr .hg/patches
$ hg qinit -c
$ hg up -q 0
$ hg qnew --git f_git.patch -d '3 0'
$ echo mq1 > p
$ hg add p
$ hg qref --git -m 'P0 (git)'
$ hg qnew f.patch -d '4 0'
$ echo mq2 > p
$ hg qref -m P1
$ hg qci -m 'save patch state'
$ hg qseries -s
f_git.patch: P0 (git)
f.patch: P1
$ hg -R .hg/patches manifest
.hgignore
f.patch
f_git.patch
series
$ cat .hg/patches/f_git.patch
Date: 3 0
P0 (git)
diff --git a/p b/p
new file mode 100644
--- /dev/null
+++ b/p
@@ -0,0 +1,1 @@
+mq1
$ cat .hg/patches/f.patch
Date: 4 0
P1
diff -r ???????????? p (glob)
--- a/p ??? ??? ?? ??:??:?? ???? ????? (glob)
+++ b/p ??? ??? ?? ??:??:?? ???? ????? (glob)
@@ -1,1 +1,1 @@
-mq1
+mq2
Rebase the applied mq patches:
$ hg rebase -s 2 -d 1
rebasing 2:0c587ffcb480 "P0 (git)" (f_git.patch qbase)
rebasing 3:c7f18665e4bc "P1" (f.patch qtip tip)
saved backup bundle to $TESTTMP/a/.hg/strip-backup/0c587ffcb480-0ea5695f-backup.hg (glob)
$ hg qci -m 'save patch state'
$ hg qseries -s
f_git.patch: P0 (git)
f.patch: P1
$ hg -R .hg/patches manifest
.hgignore
f.patch
f_git.patch
series
$ cat .hg/patches/f_git.patch
# HG changeset patch
# User test
# Date 3 0
# Thu Jan 01 00:00:03 1970 +0000
# Node ID 12d9f6a3bbe560dee50c7c454d434add7fb8e837
# Parent bac9ed9960d8992bcad75864a879fa76cadaf1b0
P0 (git)
diff --git a/p b/p
new file mode 100644
--- /dev/null
+++ b/p
@@ -0,0 +1,1 @@
+mq1
$ cat .hg/patches/f.patch
# HG changeset patch
# User test
# Date 4 0
# Thu Jan 01 00:00:04 1970 +0000
# Node ID c77a2661c64c60d82f63c4f7aefd95b3a948a557
# Parent 12d9f6a3bbe560dee50c7c454d434add7fb8e837
P1
diff -r 12d9f6a3bbe5 -r c77a2661c64c p
--- a/p Thu Jan 01 00:00:03 1970 +0000
+++ b/p Thu Jan 01 00:00:04 1970 +0000
@@ -1,1 +1,1 @@
-mq1
+mq2
$ cd ..
Rebase with guards
$ hg init foo
$ cd foo
$ echo a > a
$ hg ci -Am a
adding a
Create mq repo with guarded patches foo and bar and empty patch:
$ hg qinit
$ echo guarded > guarded
$ hg add guarded
$ hg qnew guarded
$ hg qnew empty-important -m 'important commit message' -d '1 0'
$ echo bar > bar
$ hg add bar
$ hg qnew bar -d '2 0'
$ echo foo > foo
$ hg add foo
$ hg qnew foo
$ hg qpop -a
popping foo
popping bar
popping empty-important
popping guarded
patch queue now empty
$ hg qguard guarded +guarded
$ hg qguard bar +baz
$ hg qguard foo +baz
$ hg qselect baz
number of unguarded, unapplied patches has changed from 1 to 3
$ hg qpush bar
applying empty-important
patch empty-important is empty
applying bar
now at: bar
$ hg qguard -l
guarded: +guarded
empty-important: unguarded
bar: +baz
foo: +baz
$ hg tglog
@ 2: 'imported patch bar' tags: bar qtip tip
|
o 1: 'important commit message' tags: empty-important qbase
|
o 0: 'a' tags: qparent
Create new head to rebase bar onto:
$ hg up -C 0
0 files updated, 0 files merged, 1 files removed, 0 files unresolved
$ echo b > b
$ hg add b
$ hg ci -m b
created new head
$ hg up -C 2
1 files updated, 0 files merged, 1 files removed, 0 files unresolved
$ echo a >> a
$ hg qref
$ hg tglog
@ 3: '[mq]: bar' tags: bar qtip tip
|
| o 2: 'b' tags:
| |
o | 1: 'important commit message' tags: empty-important qbase
|/
o 0: 'a' tags: qparent
Rebase bar (make sure series order is preserved and empty-important also is
removed from the series):
$ hg qseries
guarded
empty-important
bar
foo
$ [ -f .hg/patches/empty-important ]
$ hg -q rebase -d 2
note: rebase of 1:0aaf4c3af7eb created no changes to commit
$ hg qseries
guarded
bar
foo
$ [ -f .hg/patches/empty-important ]
[1]
$ hg qguard -l
guarded: +guarded
bar: +baz
foo: +baz
$ hg tglog
@ 2: '[mq]: bar' tags: bar qbase qtip tip
|
o 1: 'b' tags: qparent
|
o 0: 'a' tags:
$ cd ..