revlog: don't flush data file after every added revision
The current behavior of revlogs is to flush the data file when writing
data to it. Tracing system calls revealed that changegroup processing
incurred numerous write(2) calls for values much smaller than the
default buffer size (Python defaults to 4096, but it can be adjusted
based on detected block size at run time by CPython).
The reason we flush revlogs is so readers have all data available.
For example, the current code in revlog.py will re-open the revlog
file (instead of seeking an existing file handle) to read the text
of a revision. This happens when starting a new delta chain when
adding several revisions from changegroups, for example. Yes, this
is likely sub-optimal (we should probably be sharing file descriptors
between readers and writers to avoid the flushing and associated
overhead of re-opening files).
While flushing revlogs is necessary, it appears all callers are
diligent about flushing files before a read is performed (see
buildtext() in _addrevision()), making the flush in
_writeentry() redundant and unncessary. So, we remove it. In practice,
this means we incur a write(2) a) when the buffer is full (typically
4096 bytes) b) when a new delta chain is created rather than after
every added revision. This applies to every revlog, but by volume
it mostly impacts filelogs.
Removing the redundant flush from _writeentry() significantly
reduces the number of write(2) calls during changegroup processing on
my Linux machine. When applying a changegroup of the hg repo based on
my local repo, the total number of write(2) calls during application
of the mercurial/localrepo.py revlogs dropped from 1,320 to 217 with
this patch applied. Total I/O related system calls dropped from 1,577
to 474.
When unbundling a mozilla-central gzipped bundle (264,403 changesets
with 1,492,215 changes to 222,507 files), total write(2) calls
dropped from 1,252,881 to 827,106 and total system calls dropped from
3,601,259 to 3,178,636 - a reduction of 425,775!
While the system call reduction is significant, it appears
to have no impact on wall time on my Linux and Windows machines. Still,
fewer syscalls is fewer syscalls. Surely this can't hurt. If nothing
else, it makes examining remaining system call usage simpler and opens
the door to experimenting with the performance impact of different
buffer sizes.
$ cat >> $HGRCPATH <<EOF
> [extensions]
> rebase=
>
> [alias]
> tlog = log --template "{rev}: '{desc}' {branches}\n"
> tglog = tlog --graph
> EOF
$ hg init a
$ cd a
$ mkdir d
$ echo a > a
$ hg ci -Am A
adding a
$ echo b > d/b
$ hg ci -Am B
adding d/b
$ hg mv d d-renamed
moving d/b to d-renamed/b (glob)
$ hg ci -m 'rename B'
$ hg up -q -C 1
$ hg mv a a-renamed
$ echo x > d/x
$ hg add d/x
$ hg ci -m 'rename A'
created new head
$ hg tglog
@ 3: 'rename A'
|
| o 2: 'rename B'
|/
o 1: 'B'
|
o 0: 'A'
Rename is tracked:
$ hg tlog -p --git -r tip
3: 'rename A'
diff --git a/a b/a-renamed
rename from a
rename to a-renamed
diff --git a/d/x b/d/x
new file mode 100644
--- /dev/null
+++ b/d/x
@@ -0,0 +1,1 @@
+x
Rebase the revision containing the rename:
$ hg rebase -s 3 -d 2
rebasing 3:73a3ee40125d "rename A" (tip)
saved backup bundle to $TESTTMP/a/.hg/strip-backup/73a3ee40125d-1d78ebcf-backup.hg (glob)
$ hg tglog
@ 3: 'rename A'
|
o 2: 'rename B'
|
o 1: 'B'
|
o 0: 'A'
Rename is not lost:
$ hg tlog -p --git -r tip
3: 'rename A'
diff --git a/a b/a-renamed
rename from a
rename to a-renamed
diff --git a/d-renamed/x b/d-renamed/x
new file mode 100644
--- /dev/null
+++ b/d-renamed/x
@@ -0,0 +1,1 @@
+x
Rebased revision does not contain information about b (issue3739)
$ hg log -r 3 --debug
changeset: 3:032a9b75e83bff1dcfb6cbfa4ef50a704bf1b569
tag: tip
phase: draft
parent: 2:220d0626d185f372d9d8f69d9c73b0811d7725f7
parent: -1:0000000000000000000000000000000000000000
manifest: 3:035d66b27a1b06b2d12b46d41a39adb7a200c370
user: test
date: Thu Jan 01 00:00:00 1970 +0000
files+: a-renamed d-renamed/x
files-: a
extra: branch=default
extra: rebase_source=73a3ee40125d6f0f347082e5831ceccb3f005f8a
description:
rename A
$ cd ..
$ hg init b
$ cd b
$ echo a > a
$ hg ci -Am A
adding a
$ echo b > b
$ hg ci -Am B
adding b
$ hg cp b b-copied
$ hg ci -Am 'copy B'
$ hg up -q -C 1
$ hg cp a a-copied
$ hg ci -m 'copy A'
created new head
$ hg tglog
@ 3: 'copy A'
|
| o 2: 'copy B'
|/
o 1: 'B'
|
o 0: 'A'
Copy is tracked:
$ hg tlog -p --git -r tip
3: 'copy A'
diff --git a/a b/a-copied
copy from a
copy to a-copied
Rebase the revision containing the copy:
$ hg rebase -s 3 -d 2
rebasing 3:0a8162ff18a8 "copy A" (tip)
saved backup bundle to $TESTTMP/b/.hg/strip-backup/0a8162ff18a8-dd06302a-backup.hg (glob)
$ hg tglog
@ 3: 'copy A'
|
o 2: 'copy B'
|
o 1: 'B'
|
o 0: 'A'
Copy is not lost:
$ hg tlog -p --git -r tip
3: 'copy A'
diff --git a/a b/a-copied
copy from a
copy to a-copied
Rebased revision does not contain information about b (issue3739)
$ hg log -r 3 --debug
changeset: 3:98f6e6dbf45ab54079c2237fbd11066a5c41a11d
tag: tip
phase: draft
parent: 2:39e588434882ff77d01229d169cdc77f29e8855e
parent: -1:0000000000000000000000000000000000000000
manifest: 3:2232f329d66fffe3930d43479ae624f66322b04d
user: test
date: Thu Jan 01 00:00:00 1970 +0000
files+: a-copied
extra: branch=default
extra: rebase_source=0a8162ff18a8900df8df8ef7ac0046955205613e
description:
copy A
$ cd ..
Test rebase across repeating renames:
$ hg init repo
$ cd repo
$ echo testing > file1.txt
$ hg add file1.txt
$ hg ci -m "Adding file1"
$ hg rename file1.txt file2.txt
$ hg ci -m "Rename file1 to file2"
$ echo Unrelated change > unrelated.txt
$ hg add unrelated.txt
$ hg ci -m "Unrelated change"
$ hg rename file2.txt file1.txt
$ hg ci -m "Rename file2 back to file1"
$ hg update -r -2
1 files updated, 0 files merged, 1 files removed, 0 files unresolved
$ echo Another unrelated change >> unrelated.txt
$ hg ci -m "Another unrelated change"
created new head
$ hg tglog
@ 4: 'Another unrelated change'
|
| o 3: 'Rename file2 back to file1'
|/
o 2: 'Unrelated change'
|
o 1: 'Rename file1 to file2'
|
o 0: 'Adding file1'
$ hg rebase -s 4 -d 3
rebasing 4:b918d683b091 "Another unrelated change" (tip)
saved backup bundle to $TESTTMP/repo/.hg/strip-backup/b918d683b091-3024bc57-backup.hg (glob)
$ hg diff --stat -c .
unrelated.txt | 1 +
1 files changed, 1 insertions(+), 0 deletions(-)
$ cd ..
Verify that copies get preserved (issue4192).
$ hg init copy-gets-preserved
$ cd copy-gets-preserved
$ echo a > a
$ hg add a
$ hg commit --message "File a created"
$ hg copy a b
$ echo b > b
$ hg commit --message "File b created as copy of a and modified"
$ hg copy b c
$ echo c > c
$ hg commit --message "File c created as copy of b and modified"
$ hg copy c d
$ echo d > d
$ hg commit --message "File d created as copy of c and modified"
Note that there are four entries in the log for d
$ hg tglog --follow d
@ 3: 'File d created as copy of c and modified'
|
o 2: 'File c created as copy of b and modified'
|
o 1: 'File b created as copy of a and modified'
|
o 0: 'File a created'
Update back to before we performed copies, and inject an unrelated change.
$ hg update 0
0 files updated, 0 files merged, 3 files removed, 0 files unresolved
$ echo unrelated > unrelated
$ hg add unrelated
$ hg commit --message "Unrelated file created"
created new head
$ hg update 4
0 files updated, 0 files merged, 0 files removed, 0 files unresolved
Rebase the copies on top of the unrelated change.
$ hg rebase --source 1 --dest 4
rebasing 1:79d255d24ad2 "File b created as copy of a and modified"
rebasing 2:327f772bc074 "File c created as copy of b and modified"
rebasing 3:421b7e82bb85 "File d created as copy of c and modified"
saved backup bundle to $TESTTMP/copy-gets-preserved/.hg/strip-backup/79d255d24ad2-a2265555-backup.hg (glob)
$ hg update 4
3 files updated, 0 files merged, 0 files removed, 0 files unresolved
There should still be four entries in the log for d
$ hg tglog --follow d
@ 4: 'File d created as copy of c and modified'
|
o 3: 'File c created as copy of b and modified'
|
o 2: 'File b created as copy of a and modified'
|
o 0: 'File a created'
Same steps as above, but with --collapse on rebase to make sure the
copy records collapse correctly.
$ hg co 1
0 files updated, 0 files merged, 3 files removed, 0 files unresolved
$ echo more >> unrelated
$ hg ci -m 'unrelated commit is unrelated'
created new head
$ hg rebase -s 2 --dest 5 --collapse
rebasing 2:68bf06433839 "File b created as copy of a and modified"
note: rebase of 2:68bf06433839 created no changes to commit
rebasing 3:af74b229bc02 "File c created as copy of b and modified"
merging b and c to c
note: rebase of 3:af74b229bc02 created no changes to commit
rebasing 4:dbb9ba033561 "File d created as copy of c and modified"
merging c and d to d
note: rebase of 4:dbb9ba033561 created no changes to commit
saved backup bundle to $TESTTMP/copy-gets-preserved/.hg/strip-backup/68bf06433839-dde37595-backup.hg (glob)
$ hg co tip
3 files updated, 0 files merged, 0 files removed, 0 files unresolved
This should show both revision 3 and 0 since 'd' was transitively a
copy of 'a'.
$ hg tglog --follow d
@ 3: 'Collapsed revision
| * File b created as copy of a and modified
| * File c created as copy of b and modified
| * File d created as copy of c and modified'
o 0: 'File a created'
$ cd ..