wireproto: add streams to frame-based protocol
Previously, the frame-based protocol was just a series of frames,
with each frame associated with a request ID.
In order to scale the protocol, we'll want to enable the use of
compression. While it is possible to enable compression at the
socket/pipe level, this has its disadvantages. The big one is it
undermines the point of frames being standalone, atomic units that
can be read and written: if you add compression above the framing
protocol, you are back to having a stream-based protocol as opposed
to something frame-based.
So in order to preserve frames, compression needs to occur at
the frame payload level.
Compressing each frame's payload individually will limit compression
ratios because the window size of the compressor will be limited
by the max frame size, which is 32-64kb as currently defined. It
will also add CPU overhead, as it is more efficient for compressors
to operate on fewer, larger blocks of data than more, smaller blocks.
So compressing each frame independently is out.
This means we need to compress each frame's payload as if it is part
of a larger stream.
The simplest approach is to have 1 stream per connection. This
could certainly work. However, it has disadvantages (documented below).
We could also have 1 stream per RPC/command invocation. (This is the
model HTTP/2 goes with.) This also has disadvantages.
The main disadvantage to one global stream is that it has the very
real potential to create CPU bottlenecks doing compression. Networks
are only getting faster and the performance of single CPU cores has
been relatively flat. Newer compression formats like zstandard offer
better CPU cycle efficiency than predecessors like zlib. But it still
all too common to saturate your CPU with compression overhead long
before you saturate the network pipe.
The main disadvantage with streams per request is that you can't
reap the benefits of the compression context for multiple requests.
For example, if you send 1000 RPC requests (or HTTP/2 requests for
that matter), the response to each would have its own compression
context. The overall size of the raw responses would be larger because
compression contexts wouldn't be able to reference data from another
request or response.
The approach for streams as implemented in this commit is to support
N streams per connection and for streams to potentially span requests
and responses. As explained by the added internals docs, this
facilitates servers and clients delegating independent streams and
compression to independent threads / CPU cores. This helps alleviate
the CPU bottleneck of compression. This design also allows compression
contexts to be reused across requests/responses. This can result in
improved compression ratios and less overhead for compressors and
decompressors having to build new contexts.
Another feature that was defined was the ability for individual frames
within a stream to declare whether that individual frame's payload
uses the content encoding (read: compression) defined by the stream.
The idea here is that some servers may serve data from a combination
of caches and dynamic resolution. Data coming from caches may be
pre-compressed. We want to facilitate servers being able to essentially
stream bytes from caches to the wire with minimal overhead. Being
able to mix and match with frames are compressed within a stream
enables these types of advanced server functionality.
This commit defines the new streams mechanism. Basic code for
supporting streams in frames has been added. But that code is
seriously lacking and doesn't fully conform to the defined protocol.
For example, we don't close any streams. And support for content
encoding within streams is not yet implemented. The change was
rather invasive and I didn't think it would be reasonable to implement
the entire feature in a single commit.
For the record, I would have loved to reuse an existing multiplexing
protocol to build the new wire protocol on top of. However, I couldn't
find a protocol that offers the performance and scaling characteristics
that I desired. Namely, it should support multiple compression
contexts to facilitate scaling out to multiple CPU cores and
compression contexts should be able to live longer than single RPC
requests. HTTP/2 *almost* fits the bill. But the semantics of HTTP
message exchange state that streams can only live for a single
request-response. We /could/ tunnel on top of HTTP/2 streams and
frames with HEADER and DATA frames. But there's no guarantee that
HTTP/2 libraries and proxies would allow us to use HTTP/2 streams
and frames without the HTTP message exchange semantics defined in
RFC 7540 Section 8. Other RPC protocols like gRPC tunnel are built
on top of HTTP/2 and thus preserve its semantics of stream per
RPC invocation. Even QUIC does this. We could attempt to invent a
higher-level stream that spans HTTP/2 streams. But this would be
violating HTTP/2 because there is no guarantee that HTTP/2 streams
are routed to the same server. The best we can do - which is what
this protocol does - is shoehorn all request and response data into
a single HTTP message and create streams within. At that point, we've
defined a Content-Type in HTTP parlance. It just so happens our
media type can also work as a standalone, stream-based protocol,
without leaning on HTTP or similar protocol.
Differential Revision: https://phab.mercurial-scm.org/D2907
$ hg init
$ mkdir d1 d1/d11 d2
$ echo d1/a > d1/a
$ echo d1/ba > d1/ba
$ echo d1/a1 > d1/d11/a1
$ echo d1/b > d1/b
$ echo d2/b > d2/b
$ hg add d1/a d1/b d1/ba d1/d11/a1 d2/b
$ hg commit -m "1"
rename a single file
$ hg rename d1/d11/a1 d2/c
$ hg --config ui.portablefilenames=abort rename d1/a d1/con.xml
abort: filename contains 'con', which is reserved on Windows: d1/con.xml
[255]
$ hg sum
parent: 0:9b4b6e7b2c26 tip
1
branch: default
commit: 1 renamed
update: (current)
phases: 1 draft
$ hg status -C
A d2/c
d1/d11/a1
R d1/d11/a1
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d2/c
rename a single file using absolute paths
$ hg rename `pwd`/d1/d11/a1 `pwd`/d2/c
$ hg status -C
A d2/c
d1/d11/a1
R d1/d11/a1
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d2/c
rename --after a single file
$ mv d1/d11/a1 d2/c
$ hg rename --after d1/d11/a1 d2/c
$ hg status -C
A d2/c
d1/d11/a1
R d1/d11/a1
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d2/c
rename --after a single file when src and tgt already tracked
$ mv d1/d11/a1 d2/c
$ hg addrem -s 0
removing d1/d11/a1
adding d2/c
$ hg rename --after d1/d11/a1 d2/c
$ hg status -C
A d2/c
d1/d11/a1
R d1/d11/a1
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d2/c
rename --after a single file to a nonexistent target filename
$ hg rename --after d1/a dummy
d1/a: not recording move - dummy does not exist
move a single file to an existing directory
$ hg rename d1/d11/a1 d2
$ hg status -C
A d2/a1
d1/d11/a1
R d1/d11/a1
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d2/a1
move --after a single file to an existing directory
$ mv d1/d11/a1 d2
$ hg rename --after d1/d11/a1 d2
$ hg status -C
A d2/a1
d1/d11/a1
R d1/d11/a1
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d2/a1
rename a file using a relative path
$ (cd d1/d11; hg rename ../../d2/b e)
$ hg status -C
A d1/d11/e
d2/b
R d2/b
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d1/d11/e
rename --after a file using a relative path
$ (cd d1/d11; mv ../../d2/b e; hg rename --after ../../d2/b e)
$ hg status -C
A d1/d11/e
d2/b
R d2/b
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d1/d11/e
rename directory d1 as d3
$ hg rename d1/ d3
moving d1/a to d3/a
moving d1/b to d3/b
moving d1/ba to d3/ba
moving d1/d11/a1 to d3/d11/a1
$ hg status -C
A d3/a
d1/a
A d3/b
d1/b
A d3/ba
d1/ba
A d3/d11/a1
d1/d11/a1
R d1/a
R d1/b
R d1/ba
R d1/d11/a1
$ hg update -C
4 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d3
rename --after directory d1 as d3
$ mv d1 d3
$ hg rename --after d1 d3
moving d1/a to d3/a
moving d1/b to d3/b
moving d1/ba to d3/ba
moving d1/d11/a1 to d3/d11/a1
$ hg status -C
A d3/a
d1/a
A d3/b
d1/b
A d3/ba
d1/ba
A d3/d11/a1
d1/d11/a1
R d1/a
R d1/b
R d1/ba
R d1/d11/a1
$ hg update -C
4 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d3
move a directory using a relative path
$ (cd d2; mkdir d3; hg rename ../d1/d11 d3)
moving ../d1/d11/a1 to d3/d11/a1
$ hg status -C
A d2/d3/d11/a1
d1/d11/a1
R d1/d11/a1
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d2/d3
move --after a directory using a relative path
$ (cd d2; mkdir d3; mv ../d1/d11 d3; hg rename --after ../d1/d11 d3)
moving ../d1/d11/a1 to d3/d11/a1
$ hg status -C
A d2/d3/d11/a1
d1/d11/a1
R d1/d11/a1
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d2/d3
move directory d1/d11 to an existing directory d2 (removes empty d1)
$ hg rename d1/d11/ d2
moving d1/d11/a1 to d2/d11/a1
$ hg status -C
A d2/d11/a1
d1/d11/a1
R d1/d11/a1
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d2/d11
move directories d1 and d2 to a new directory d3
$ mkdir d3
$ hg rename d1 d2 d3
moving d1/a to d3/d1/a
moving d1/b to d3/d1/b
moving d1/ba to d3/d1/ba
moving d1/d11/a1 to d3/d1/d11/a1
moving d2/b to d3/d2/b
$ hg status -C
A d3/d1/a
d1/a
A d3/d1/b
d1/b
A d3/d1/ba
d1/ba
A d3/d1/d11/a1
d1/d11/a1
A d3/d2/b
d2/b
R d1/a
R d1/b
R d1/ba
R d1/d11/a1
R d2/b
$ hg update -C
5 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d3
move --after directories d1 and d2 to a new directory d3
$ mkdir d3
$ mv d1 d2 d3
$ hg rename --after d1 d2 d3
moving d1/a to d3/d1/a
moving d1/b to d3/d1/b
moving d1/ba to d3/d1/ba
moving d1/d11/a1 to d3/d1/d11/a1
moving d2/b to d3/d2/b
$ hg status -C
A d3/d1/a
d1/a
A d3/d1/b
d1/b
A d3/d1/ba
d1/ba
A d3/d1/d11/a1
d1/d11/a1
A d3/d2/b
d2/b
R d1/a
R d1/b
R d1/ba
R d1/d11/a1
R d2/b
$ hg update -C
5 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d3
move everything under directory d1 to existing directory d2, do not
overwrite existing files (d2/b)
$ hg rename d1/* d2
d2/b: not overwriting - file already committed
(hg rename --force to replace the file by recording a rename)
moving d1/d11/a1 to d2/d11/a1
$ hg status -C
A d2/a
d1/a
A d2/ba
d1/ba
A d2/d11/a1
d1/d11/a1
R d1/a
R d1/ba
R d1/d11/a1
$ diff -u d1/b d2/b
--- d1/b * (glob)
+++ d2/b * (glob)
@@ * (glob)
-d1/b
+d2/b
[1]
$ hg update -C
3 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d2/a d2/ba d2/d11/a1
attempt to move one file into a non-existent directory
$ hg rename d1/a dx/
abort: destination dx/ is not a directory
[255]
$ hg status -C
$ hg update -C
0 files updated, 0 files merged, 0 files removed, 0 files unresolved
attempt to move potentially more than one file into a non-existent directory
$ hg rename 'glob:d1/**' dx
abort: with multiple sources, destination must be an existing directory
[255]
move every file under d1 to d2/d21
$ mkdir d2/d21
$ hg rename 'glob:d1/**' d2/d21
moving d1/a to d2/d21/a
moving d1/b to d2/d21/b
moving d1/ba to d2/d21/ba
moving d1/d11/a1 to d2/d21/a1
$ hg status -C
A d2/d21/a
d1/a
A d2/d21/a1
d1/d11/a1
A d2/d21/b
d1/b
A d2/d21/ba
d1/ba
R d1/a
R d1/b
R d1/ba
R d1/d11/a1
$ hg update -C
4 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d2/d21
move --after some files under d1 to d2/d21
$ mkdir d2/d21
$ mv d1/a d1/d11/a1 d2/d21
$ hg rename --after 'glob:d1/**' d2/d21
moving d1/a to d2/d21/a
d1/b: not recording move - d2/d21/b does not exist
d1/ba: not recording move - d2/d21/ba does not exist
moving d1/d11/a1 to d2/d21/a1
$ hg status -C
A d2/d21/a
d1/a
A d2/d21/a1
d1/d11/a1
R d1/a
R d1/d11/a1
$ hg update -C
2 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d2/d21
move every file under d1 starting with an 'a' to d2/d21 (regexp)
$ mkdir d2/d21
$ hg rename 're:d1/([^a][^/]*/)*a.*' d2/d21
moving d1/a to d2/d21/a
moving d1/d11/a1 to d2/d21/a1
$ hg status -C
A d2/d21/a
d1/a
A d2/d21/a1
d1/d11/a1
R d1/a
R d1/d11/a1
$ hg update -C
2 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d2/d21
attempt to overwrite an existing file
$ echo "ca" > d1/ca
$ hg rename d1/ba d1/ca
d1/ca: not overwriting - file exists
(hg rename --after to record the rename)
$ hg status -C
? d1/ca
$ hg update -C
0 files updated, 0 files merged, 0 files removed, 0 files unresolved
forced overwrite of an existing file
$ echo "ca" > d1/ca
$ hg rename --force d1/ba d1/ca
$ hg status -C
A d1/ca
d1/ba
R d1/ba
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d1/ca
attempt to overwrite an existing broken symlink
#if symlink
$ ln -s ba d1/ca
$ hg rename --traceback d1/ba d1/ca
d1/ca: not overwriting - file exists
(hg rename --after to record the rename)
$ hg status -C
? d1/ca
$ hg update -C
0 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d1/ca
replace a symlink with a file
$ ln -s ba d1/ca
$ hg rename --force d1/ba d1/ca
$ hg status -C
A d1/ca
d1/ba
R d1/ba
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d1/ca
#endif
do not copy more than one source file to the same destination file
$ mkdir d3
$ hg rename d1/* d2/* d3
moving d1/d11/a1 to d3/d11/a1
d3/b: not overwriting - d2/b collides with d1/b
$ hg status -C
A d3/a
d1/a
A d3/b
d1/b
A d3/ba
d1/ba
A d3/d11/a1
d1/d11/a1
R d1/a
R d1/b
R d1/ba
R d1/d11/a1
$ hg update -C
4 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d3
move a whole subtree with "hg rename ."
$ mkdir d3
$ (cd d1; hg rename . ../d3)
moving a to ../d3/d1/a
moving b to ../d3/d1/b
moving ba to ../d3/d1/ba
moving d11/a1 to ../d3/d1/d11/a1
$ hg status -C
A d3/d1/a
d1/a
A d3/d1/b
d1/b
A d3/d1/ba
d1/ba
A d3/d1/d11/a1
d1/d11/a1
R d1/a
R d1/b
R d1/ba
R d1/d11/a1
$ hg update -C
4 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d3
move a whole subtree with "hg rename --after ."
$ mkdir d3
$ mv d1/* d3
$ (cd d1; hg rename --after . ../d3)
moving a to ../d3/a
moving b to ../d3/b
moving ba to ../d3/ba
moving d11/a1 to ../d3/d11/a1
$ hg status -C
A d3/a
d1/a
A d3/b
d1/b
A d3/ba
d1/ba
A d3/d11/a1
d1/d11/a1
R d1/a
R d1/b
R d1/ba
R d1/d11/a1
$ hg update -C
4 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d3
move the parent tree with "hg rename .."
$ (cd d1/d11; hg rename .. ../../d3)
moving ../a to ../../d3/a
moving ../b to ../../d3/b
moving ../ba to ../../d3/ba
moving a1 to ../../d3/d11/a1
$ hg status -C
A d3/a
d1/a
A d3/b
d1/b
A d3/ba
d1/ba
A d3/d11/a1
d1/d11/a1
R d1/a
R d1/b
R d1/ba
R d1/d11/a1
$ hg update -C
4 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d3
skip removed files
$ hg remove d1/b
$ hg rename d1 d3
moving d1/a to d3/a
moving d1/ba to d3/ba
moving d1/d11/a1 to d3/d11/a1
$ hg status -C
A d3/a
d1/a
A d3/ba
d1/ba
A d3/d11/a1
d1/d11/a1
R d1/a
R d1/b
R d1/ba
R d1/d11/a1
$ hg update -C
4 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm -rf d3
transitive rename
$ hg rename d1/b d1/bb
$ hg rename d1/bb d1/bc
$ hg status -C
A d1/bc
d1/b
R d1/b
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d1/bc
transitive rename --after
$ hg rename d1/b d1/bb
$ mv d1/bb d1/bc
$ hg rename --after d1/bb d1/bc
$ hg status -C
A d1/bc
d1/b
R d1/b
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d1/bc
$ echo "# idempotent renames (d1/b -> d1/bb followed by d1/bb -> d1/b)"
# idempotent renames (d1/b -> d1/bb followed by d1/bb -> d1/b)
$ hg rename d1/b d1/bb
$ echo "some stuff added to d1/bb" >> d1/bb
$ hg rename d1/bb d1/b
$ hg status -C
M d1/b
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
overwriting with renames (issue1959)
$ hg rename d1/a d1/c
$ hg rename d1/b d1/a
$ hg status -C
M d1/a
d1/b
A d1/c
d1/a
R d1/b
$ hg diff --git
diff --git a/d1/a b/d1/a
--- a/d1/a
+++ b/d1/a
@@ -1,1 +1,1 @@
-d1/a
+d1/b
diff --git a/d1/b b/d1/b
deleted file mode 100644
--- a/d1/b
+++ /dev/null
@@ -1,1 +0,0 @@
-d1/b
diff --git a/d1/a b/d1/c
copy from d1/a
copy to d1/c
$ hg update -C
2 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm d1/c # The file was marked as added, so 'hg update' action was 'forget'
check illegal path components
$ hg rename d1/d11/a1 .hg/foo
abort: path contains illegal component: .hg/foo
[255]
$ hg status -C
$ hg rename d1/d11/a1 ../foo
abort: ../foo not under root '$TESTTMP'
[255]
$ hg status -C
$ mv d1/d11/a1 .hg/foo
$ hg rename --after d1/d11/a1 .hg/foo
abort: path contains illegal component: .hg/foo
[255]
$ hg status -C
! d1/d11/a1
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm .hg/foo
$ hg rename d1/d11/a1 .hg
abort: path contains illegal component: .hg/a1
[255]
$ hg --config extensions.largefiles= rename d1/d11/a1 .hg
The fsmonitor extension is incompatible with the largefiles extension and has been disabled. (fsmonitor !)
abort: path contains illegal component: .hg/a1
[255]
$ hg status -C
$ hg rename d1/d11/a1 ..
abort: ../a1 not under root '$TESTTMP'
[255]
$ hg --config extensions.largefiles= rename d1/d11/a1 ..
The fsmonitor extension is incompatible with the largefiles extension and has been disabled. (fsmonitor !)
abort: ../a1 not under root '$TESTTMP'
[255]
$ hg status -C
$ mv d1/d11/a1 .hg
$ hg rename --after d1/d11/a1 .hg
abort: path contains illegal component: .hg/a1
[255]
$ hg status -C
! d1/d11/a1
$ hg update -C
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ rm .hg/a1
$ (cd d1/d11; hg rename ../../d2/b ../../.hg/foo)
abort: path contains illegal component: .hg/foo
[255]
$ hg status -C
$ (cd d1/d11; hg rename ../../d2/b ../../../foo)
abort: ../../../foo not under root '$TESTTMP'
[255]
$ hg status -C
check that stat information such as mtime is preserved on rename - it's unclear
whether the `touch` and `stat` commands are portable, so we mimic them using
python. Not all platforms support precision of even one-second granularity, so
we allow a rather generous fudge factor here; 1234567890 is 2009, and the
primary thing we care about is that it's not the machine's current time;
hopefully it's really unlikely for a machine to have such a broken clock that
this test fails. :)
$ mkdir mtime
Create the file (as empty), then update its mtime and atime to be 1234567890.
>>> import os
>>> filename = "mtime/f"
>>> mtime = 1234567890
>>> open(filename, "w").close()
>>> os.utime(filename, (mtime, mtime))
$ hg ci -qAm 'add mtime dir'
"hg cp" does not preserve the mtime, so it should be newer than the 2009
timestamp.
$ hg cp -q mtime mtime_cp
>>> from __future__ import print_function
>>> import os
>>> filename = "mtime_cp/f"
>>> print(os.stat(filename).st_mtime < 1234567999)
False
"hg mv" preserves the mtime, so it should be ~equal to the 2009 timestamp
(modulo some fudge factor due to not every system supporting 1s-level
precision).
$ hg mv -q mtime mtime_mv
>>> from __future__ import print_function
>>> import os
>>> filename = "mtime_mv/f"
>>> print(os.stat(filename).st_mtime < 1234567999)
True