view tests/test-commit-interactive-curses.t @ 40021:c537144fdbef

wireprotov2: support response caching

One of the things I've learned from managing VCS servers over the
years is that they are hard to scale. It is well known that some
companies have very beefy (read: very expensive) servers to power
their VCS needs. It is also known that specialized servers for
various VCS exist in order to facilitate scaling servers. (Mercurial
is in this boat.)

One of the aspects that make a VCS server hard to scale is the
high CPU load incurred by constant client clone/pull operations.

To alleviate the scaling pain associated with data retrieval
operations, I want to integrate caching into the Mercurial wire
protocol server as robustly as possible such that servers can
aggressively cache responses and defer as much server load as
possible.

This commit represents the initial implementation of a general
caching layer in wire protocol version 2.

We define a new interface and behavior for a wire protocol cacher
in repository.py. (This is probably where a reviewer should look
first to understand what is going on.)

The bulk of the added code is in wireprotov2server.py, where we
define how a command can opt in to being cached and integrate
caching into command dispatching.

From a very high-level:

* A command can declare itself as cacheable by providing a callable
  that can be used to derive a cache key.
* At dispatch time, if a command is cacheable, we attempt to
  construct a cacher and use it for serving the request and/or
  caching the request.
* The dispatch layer handles the bulk of the business logic for
  caching, making cachers mostly "dumb content stores."
* The mechanism for invalidating cached entries (one of the harder
  parts about caching in general) is by varying the cache key when
  state changes. As such, cachers don't need to be concerned with
  cache invalidation.

Initially, we've hooked up support for caching "manifestdata" and
"filedata" commands. These are the simplest to cache, as they should
be immutable over time. Caching of commands related to changeset
data is a bit harder (because cache validation is impacted by
changes to bookmarks, phases, etc). This will be implemented later.
(Strictly speaking, censoring a file should invalidate caches. I've
added an inline TODO to track this edge case.)

To prove it works, this commit implements a test-only extension
providing in-memory caching backed by an lrucachedict. A new test
showing this extension behaving properly is added. FWIW, the
cacher is ~50 lines of code, demonstrating the relative ease with
which a cache can be added to a server.

While the test cacher is not suitable for production workloads, just
for kicks I performed a clone of just the changeset and manifest data
for the mozilla-unified repository. With a fully warmed cache (of just
the manifest data since changeset data is not cached), server-side
CPU usage dropped from ~73s to ~28s. That's pretty significant and
demonstrates the potential that response caching has on server
scalability!

Differential Revision: https://phab.mercurial-scm.org/D4773

#require tic

Set up a repo

  $ cp $HGRCPATH $HGRCPATH.pretest
  $ cat <<EOF >> $HGRCPATH
  > [ui]
  > interactive = true
  > interface = curses
  > [experimental]
  > crecordtest = testModeCommands
  > EOF

Record with noeol at eof (issue5268)
  $ hg init noeol
  $ cd noeol
  $ printf '0' > a
  $ printf '0\n' > b
  $ hg ci -Aqm initial
  $ printf '1\n0' > a
  $ printf '1\n0\n' > b
  $ cat <<EOF >testModeCommands
  > c
  > EOF
  $ HGEDITOR="\"sh\" \"`pwd`/editor.sh\"" hg commit  -i -m "add hunks" -d "0 0"
  $ cd ..

Normal repo
  $ hg init a
  $ cd a

Committing some changes but stopping on the way

  $ echo "a" > a
  $ hg add a
  $ cat <<EOF >testModeCommands
  > TOGGLE
  > X
  > EOF
  $ hg commit -i  -m "a" -d "0 0"
  no changes to record
  [1]
  $ hg tip
  changeset:   -1:000000000000
  tag:         tip
  user:        
  date:        Thu Jan 01 00:00:00 1970 +0000
  

Committing some changes

  $ cat <<EOF >testModeCommands
  > X
  > EOF
  $ hg commit -i  -m "a" -d "0 0"
  $ hg tip
  changeset:   0:cb9a9f314b8b
  tag:         tip
  user:        test
  date:        Thu Jan 01 00:00:00 1970 +0000
  summary:     a
  
Check that commit -i works with no changes
  $ hg commit -i
  no changes to record
  [1]

Committing only one file

  $ echo "a" >> a
  >>> open('b', 'wb').write(b"1\n2\n3\n4\n5\n6\n7\n8\n9\n10\n") and None
  $ hg add b
  $ cat <<EOF >testModeCommands
  > TOGGLE
  > KEY_DOWN
  > X
  > EOF
  $ hg commit -i  -m "one file" -d "0 0"
  $ hg tip
  changeset:   1:fb2705a663ea
  tag:         tip
  user:        test
  date:        Thu Jan 01 00:00:00 1970 +0000
  summary:     one file
  
  $ hg cat -r tip a
  a
  $ cat a
  a
  a

Committing only one hunk while aborting edition of hunk

- Untoggle all the hunks, go down to the second file
- unfold it
- go down to second hunk (1 for the first hunk, 1 for the first hunkline, 1 for the second hunk, 1 for the second hunklike)
- toggle the second hunk
- toggle on and off the amend mode (to check that it toggles off)
- edit the hunk and quit the editor immediately with non-zero status
- commit

  $ printf "printf 'editor ran\n'; exit 1" > editor.sh
  $ echo "x" > c
  $ cat b >> c
  $ echo "y" >> c
  $ mv c b
  $ cat <<EOF >testModeCommands
  > A
  > KEY_DOWN
  > f
  > KEY_DOWN
  > KEY_DOWN
  > KEY_DOWN
  > KEY_DOWN
  > TOGGLE
  > a
  > a
  > e
  > X
  > EOF
  $ HGEDITOR="\"sh\" \"`pwd`/editor.sh\"" hg commit -i  -m "one hunk" -d "0 0"
  editor ran
  $ rm editor.sh
  $ hg tip
  changeset:   2:7d10dfe755a8
  tag:         tip
  user:        test
  date:        Thu Jan 01 00:00:00 1970 +0000
  summary:     one hunk
  
  $ hg cat -r tip b
  1
  2
  3
  4
  5
  6
  7
  8
  9
  10
  y
  $ cat b
  x
  1
  2
  3
  4
  5
  6
  7
  8
  9
  10
  y
  $ hg commit -m "other hunks"
  $ hg tip
  changeset:   3:a6735021574d
  tag:         tip
  user:        test
  date:        Thu Jan 01 00:00:00 1970 +0000
  summary:     other hunks
  
  $ hg cat -r tip b
  x
  1
  2
  3
  4
  5
  6
  7
  8
  9
  10
  y

Newly added files can be selected with the curses interface

  $ hg update -C .
  0 files updated, 0 files merged, 0 files removed, 0 files unresolved
  $ echo "hello" > x
  $ hg add x
  $ cat <<EOF >testModeCommands
  > TOGGLE
  > TOGGLE
  > X
  > EOF
  $ hg st
  A x
  ? testModeCommands
  $ hg commit -i  -m "newly added file" -d "0 0"
  $ hg st
  ? testModeCommands

Amend option works
  $ echo "hello world" > x
  $ hg diff -c .
  diff -r a6735021574d -r 2b0e9be4d336 x
  --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
  +++ b/x	Thu Jan 01 00:00:00 1970 +0000
  @@ -0,0 +1,1 @@
  +hello
  $ cat <<EOF >testModeCommands
  > a
  > X
  > EOF
  $ hg commit -i  -m "newly added file" -d "0 0"
  saved backup bundle to $TESTTMP/a/.hg/strip-backup/2b0e9be4d336-3cf0bc8c-amend.hg
  $ hg diff -c .
  diff -r a6735021574d -r c1d239d165ae x
  --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
  +++ b/x	Thu Jan 01 00:00:00 1970 +0000
  @@ -0,0 +1,1 @@
  +hello world

Make file empty
  $ printf "" > x
  $ cat <<EOF >testModeCommands
  > X
  > EOF
  $ hg ci -i -m emptify -d "0 0"
  $ hg update -C '.^' -q

Editing a hunk puts you back on that hunk when done editing (issue5041)
To do that, we change two lines in a file, pretend to edit the second line,
exit, toggle the line selected at the end of the edit and commit.
The first line should be recorded if we were put on the second line at the end
of the edit.

  $ hg update -C .
  0 files updated, 0 files merged, 0 files removed, 0 files unresolved
  $ echo "foo" > x
  $ echo "hello world" >> x
  $ echo "bar" >> x
  $ cat <<EOF >testModeCommands
  > f
  > KEY_DOWN
  > KEY_DOWN
  > KEY_DOWN
  > KEY_DOWN
  > e
  > TOGGLE
  > X
  > EOF
  $ printf "printf 'editor ran\n'; exit 0" > editor.sh
  $ HGEDITOR="\"sh\" \"`pwd`/editor.sh\"" hg commit  -i -m "edit hunk" -d "0 0" -q
  editor ran
  $ hg cat -r . x
  foo
  hello world

Testing the review option. The entire final filtered patch should show
up in the editor and be editable. We will unselect the second file and
the first hunk of the third file. During review, we will decide that
"lower" sounds better than "bottom", and the final commit should
reflect this edition.

  $ hg update -C .
  1 files updated, 0 files merged, 0 files removed, 0 files unresolved
  $ echo "top" > c
  $ cat x >> c
  $ echo "bottom" >> c
  $ mv c x
  $ echo "third a" >> a
  $ echo "we will unselect this" >> b

  $ cat > editor.sh <<EOF
  > cat "\$1"
  > cat "\$1" | sed s/bottom/lower/ > tmp
  > mv tmp "\$1"
  > EOF
  $ cat > testModeCommands <<EOF
  > KEY_DOWN
  > TOGGLE
  > KEY_DOWN
  > f
  > KEY_DOWN
  > TOGGLE
  > R
  > EOF

  $ HGEDITOR="\"sh\" \"`pwd`/editor.sh\"" hg commit  -i -m "review hunks" -d "0 0"
  # To remove '-' lines, make them ' ' lines (context).
  # To remove '+' lines, delete them.
  # Lines starting with # will be removed from the patch.
  #
  # If the patch applies cleanly, the edited patch will immediately
  # be finalised. If it does not apply cleanly, rejects files will be
  # generated. You can use those when you try again.
  diff --git a/a b/a
  --- a/a
  +++ b/a
  @@ -1,2 +1,3 @@
   a
   a
  +third a
  diff --git a/x b/x
  --- a/x
  +++ b/x
  @@ -1,2 +1,3 @@
   foo
   hello world
  +bottom

  $ hg cat -r . a
  a
  a
  third a

  $ hg cat -r . b
  x
  1
  2
  3
  4
  5
  6
  7
  8
  9
  10
  y

  $ hg cat -r . x
  foo
  hello world
  lower

Check spacemovesdown

  $ cat <<EOF >> $HGRCPATH
  > [experimental]
  > spacemovesdown = true
  > EOF
  $ cat <<EOF >testModeCommands
  > TOGGLE
  > TOGGLE
  > X
  > EOF
  $ hg status -q
  M b
  M x
  $ hg commit -i -m "nothing to commit?" -d "0 0"
  no changes to record
  [1]

Check ui.interface logic for the chunkselector

The default interface is text
  $ cp $HGRCPATH.pretest $HGRCPATH
  $ chunkselectorinterface() {
  > "$PYTHON" <<EOF
  > from mercurial import hg, ui;\
  > repo = hg.repository(ui.ui.load(), ".");\
  > print(repo.ui.interface("chunkselector"))
  > EOF
  > }
  $ chunkselectorinterface
  text

If only the default is set, we'll use that for the feature, too
  $ cp $HGRCPATH.pretest $HGRCPATH
  $ cat <<EOF >> $HGRCPATH
  > [ui]
  > interface = curses
  > EOF
  $ chunkselectorinterface
  curses

If TERM=dumb, we use text, even if the config says curses
  $ chunkselectorinterface
  curses
  $ TERM=dumb chunkselectorinterface
  text
(Something is keeping TERM=dumb in the environment unless I do this, it's not
scoped to just that previous command like in many shells)
  $ TERM=xterm chunkselectorinterface
  curses

It is possible to override the default interface with a feature specific
interface
  $ cp $HGRCPATH.pretest $HGRCPATH
  $ cat <<EOF >> $HGRCPATH
  > [ui]
  > interface = text
  > interface.chunkselector = curses
  > EOF

  $ chunkselectorinterface
  curses

  $ cp $HGRCPATH.pretest $HGRCPATH
  $ cat <<EOF >> $HGRCPATH
  > [ui]
  > interface = curses
  > interface.chunkselector = text
  > EOF

  $ chunkselectorinterface
  text

If a bad interface name is given, we use the default value (with a nice
error message to suggest that the configuration needs to be fixed)

  $ cp $HGRCPATH.pretest $HGRCPATH
  $ cat <<EOF >> $HGRCPATH
  > [ui]
  > interface = blah
  > EOF
  $ chunkselectorinterface
  invalid value for ui.interface: blah (using text)
  text

  $ cp $HGRCPATH.pretest $HGRCPATH
  $ cat <<EOF >> $HGRCPATH
  > [ui]
  > interface = curses
  > interface.chunkselector = blah
  > EOF
  $ chunkselectorinterface
  invalid value for ui.interface.chunkselector: blah (using curses)
  curses

  $ cp $HGRCPATH.pretest $HGRCPATH
  $ cat <<EOF >> $HGRCPATH
  > [ui]
  > interface = blah
  > interface.chunkselector = curses
  > EOF
  $ chunkselectorinterface
  invalid value for ui.interface: blah
  curses

  $ cp $HGRCPATH.pretest $HGRCPATH
  $ cat <<EOF >> $HGRCPATH
  > [ui]
  > interface = blah
  > interface.chunkselector = blah
  > EOF
  $ chunkselectorinterface
  invalid value for ui.interface: blah
  invalid value for ui.interface.chunkselector: blah (using text)
  text