SplitHTTP client: Add xmux (multiplex controller) for H3 & H2 #3613

ll11l1lIllIl1lll · 2024-07-29T12:24:54Z

我不确定我有没有正确理解其工作原理，并且我写完之后才发现上游已经更新了一堆提交，可能我合并的时候也哪里改炸了。直觉告诉我这可能有资源泄露问题，至少好像能用，所以我提交了。
这里假定 http.Client 会对每个请求复用连接，我实际使用的时候也看起来像是对请求复用连接。所以 idk.

这个代码其实挺烂的，我稍后再改，或者还是等另一个人去以更好的方式实现吧，至少肯定不是我。

注：好像 Interval 更倾向于定时任务的间隔，如果是硬限制最小间隔可能 Delay 更符合表达习惯。并且可能那 3 个 sc 参数也应该移到 mux

yuhan6665 · 2024-07-29T14:36:40Z

感谢 PR
虽然目前只有 SH 能用但是考虑一下我们要不要把选项并入 https://xtls.github.io/config/outbound.html#muxobject

Fangliding · 2024-07-29T14:39:53Z

em 毕竟原理不一样这个是复用client而不是mux.cool 所以放这还算河里吧

mmmray

just commenting on the implementation. maybe rprx wants more settings, but to me it seems this can resolve some quic instabilities in any case, so it's good in my opinion. thanks!

transport/internet/splithttp/config.proto

mmmray · 2024-07-29T21:24:33Z

transport/internet/splithttp/dialer.go

+	return muxMan.getClient(ctx, dest, streamSettings)
+}
+
+func createHTTPClient(ctx context.Context, dest net.Destination, streamSettings *internet.MemoryStreamConfig) *DefaultDialerClient {


i was confused where this function was used, turns out it is directly pulled in from muxmanager

i suggest to make mux manager unit-testable (and add some tests), so it takes the function as an argument

transport/internet/splithttp/dialer.go

transport/internet/splithttp/mux.go

transport/internet/splithttp/config.proto

RPRX · 2024-07-30T06:06:35Z

Good job！~~这下又不用自己写代码了~~，晚点我 review 一下

dantesieg · 2024-07-30T16:55:16Z

Using xray version including these commits causes crash after a while when using on client, SplitHTTP with h3 alpn. If needed i can send my configs, i used new mux mechanism with h3. Server uses same version taken from Github Actions. I don't know if this helps. I just enabled new mux and tried "prefer_new", "prefer_reuse" and "prefer_existing", didn't change other options.

panic: connection already exists

goroutine 400 [running]:
github.com/quic-go/quic-go.(*connMultiplexer).AddConn(0x400007d0e0, {0x75a9ec4f40?, 0x40000702f8?})
	github.com/quic-go/quic-go@v0.45.1/multiplexer.go:59 +0x198
github.com/quic-go/quic-go.(*Transport).init.func1()
	github.com/quic-go/quic-go@v0.45.1/transport.go:266 +0x3b0
sync.(*Once).doSlow(0xb?, 0x1?)
	sync/once.go:74 +0x100
sync.(*Once).Do(...)
	sync/once.go:65
github.com/quic-go/quic-go.(*Transport).init(0x4000d18700, 0x78?)
	github.com/quic-go/quic-go@v0.45.1/transport.go:225 +0x58
github.com/quic-go/quic-go.(*Transport).dial(0x4000d18700, {0x5db44f6450, 0x40021bdea0}, {0x5db44ed5b0, 0x4000d7d5f0}, {0x0, 0x0}, 0x4000d109c0, 0x4000d1ab40, 0x1)
	github.com/quic-go/quic-go@v0.45.1/transport.go:212 +0x70
github.com/quic-go/quic-go.(*Transport).DialEarly(...)
	github.com/quic-go/quic-go@v0.45.1/transport.go:204
github.com/quic-go/quic-go.DialEarly({0x5db44f6450, 0x40021bdea0}, {0x5db44fc670?, 0x40000702f8?}, {0x5db44ed5b0, 0x4000d7d5f0}, 0x4000d109c0, 0x4000d1ab40)
	github.com/quic-go/quic-go@v0.45.1/client.go:95 +0xfc
github.com/xtls/xray-core/transport/internet/splithttp.createHTTPClient.func2({0x5db44f6450, 0x40021bdea0}, {0x5db3d00321?, 0x5bd7?}, 0x4000d109c0, 0x4000d1ab40)
	github.com/xtls/xray-core/transport/internet/splithttp/dialer.go:130 +0x1b4
github.com/quic-go/quic-go/http3.(*RoundTripper).dial(0x4000251e30, {0x5db44f6450, 0x40021bdea0}, {0x4003602db0, 0x10})
	github.com/quic-go/quic-go@v0.45.1/http3/roundtrip.go:312 +0x224
github.com/quic-go/quic-go/http3.(*RoundTripper).getClient.func1()
	github.com/quic-go/quic-go@v0.45.1/http3/roundtrip.go:249 +0x7c
created by github.com/quic-go/quic-go/http3.(*RoundTripper).getClient in goroutine 397
	github.com/quic-go/quic-go@v0.45.1/http3/roundtrip.go:246 +0x258

LearZhou · 2024-08-02T04:28:45Z

我这边大致的测试中：prefer_new更友好

在alpn是h2时若使用prefer_reuse，反而会变慢，不如关闭mux

RPRX · 2024-08-07T04:47:07Z

等 #3624 和 #3643 合并后 rebase 一下我再看看吧

RPRX · 2024-08-10T22:27:04Z

@ll11l1lIllIl1lll 可以 rebase 一下了，基于 main

RPRX · 2024-08-11T10:17:05Z

先改了下标题，晚点我比照着 #3560 (comment) 看下代码

mmmray · 2024-08-11T18:39:38Z

general comment: i think there are already too many things called mux. most are related to the v2ray universe. there is some dangerous overlap with http with "h2mux" in sing-box. this is a mux but the rfcs don't call it that. can it be called "connection pool settings" or something else?

RPRX · 2024-08-16T10:23:08Z

首先 mode 不需要那么多别名，其次基础模式是二选一不是共存，prefer_reuse 就对应 concurrency，prefer_new 就对应另一个

好像没看到限制“一个连接最多被累计复用多少次”的代码

#3560 (comment) 下面说的 send rate 和 number of bytes sent/rcvd 等先不加，~~只是提一下，毕竟 PR 里也没有~~

RPRX · 2024-08-16T10:38:50Z

general comment: i think there are already too many things called mux. most are related to the v2ray universe. there is some dangerous overlap with http with "h2mux" in sing-box. this is a mux but the rfcs don't call it that. can it be called "connection pool settings" or something else?

可以叫 XMC：Xray Multiplex Controller，因为这套机制完善后还要加给 Xray 的 Mux、gRPC、H2 等，~~虽然可能会把 H2 删掉~~

RPRX · 2024-08-16T11:01:26Z

@Fangliding xPaddingBytes 的文档

RPRX · 2024-08-16T11:03:55Z

@Fangliding 以及 "Cache-Control: no-store"

mmmray · 2024-08-16T11:36:38Z

it's too awkward to update the document when the release is not out. starting some updates here XTLS/Xray-docs-next#558

Fangliding · 2024-08-16T13:21:10Z

@Fangliding 以及 "Cache-Control: no-store"

#3652 的改动是默认的没有config选项((((

RPRX · 2024-08-23T08:12:44Z

To konsclufka：~~出来干活了~~

RPRX · 2024-08-28T05:32:51Z

@mmmray ~~又要 rebase 了~~

APT-ZERO · 2024-08-28T19:58:41Z

Is this options for bypassing single connection speed limit of ISPs?
Is it possible to add same multi connection controller to gRPC?

mmmray · 2024-08-28T21:53:41Z

I'll take over this PR. I might add some more things if there is time. I hope to have it ready by sunday.

@APT-ZERO The answer is, yes can be added to grpc and a bunch of others, but it's not planned right now, and it can bypass speed limits like that, but I think if everybody uses it it will become meaningless (especially if you think of Irancell). That's all I'll say here. I suggest to go to github discussions or telegram for general discussion, I think this PR thread is already way too long and hard to navigate ~~but i'm often the first one to complain about such things~~

RPRX · 2024-09-15T03:48:17Z

改个名，既然 mux 是给原本没 mux 的东西加 mux，这个可以叫 demux，即 @mmmray 所说的 "un-mux"

MaxUses 改名 ConnectionReuseTimes

mmmray · 2024-09-15T15:48:06Z

done, but the "demux" is too strange and different from mux.cool settings.

In my mind "adding mux to config" means "configuring mux explicitly", not "adding mux". What about changing the default value of mux to concurrency=1 or just some value lower than infinity?

RPRX · 2024-09-15T16:34:32Z

~~遇事不决 X~~，改名 xmux，不叫 HttpDemux 或 SplitHTTPMux，未来这套机制可以 port 给 gRPC，以及 Xray 现有的 mux

mmmray · 2024-09-15T16:38:55Z

bold proposal, mux settings tries to use the "transport-native" mux, then falls back to mux.cool if there is no such thing. prevent people from using mux.cool with h3, which is probably never a good idea (?)

mmmray · 2024-09-15T16:43:27Z

~~the config parameter is done, some structs are still named specific to splithttp, but it can be fixed in the PR where this is ported to other transports.~~ fine, this was a bit too lazy

RPRX · 2024-09-15T16:43:34Z

SplitHTTPMux->Xmux

Fangliding · 2024-09-15T16:48:30Z

依我看最好别把逻辑搞太复杂 check来check去的直接在文档警告别启用mux.cool就是了 grpc也有类似的警告或者这样多层套娃没准对打乱指纹也有用 split这个直接叫muxcontroller或者类似的东西（如标题） xmux这个名字留给以后真port到mux

RPRX · 2024-09-15T16:52:35Z

In this case, the mux manager will open 8 connections first, then open more connections if the concurrency limit has been hit on all connections.

我明天再想一下这个，应该还要重命名

RPRX · 2024-09-16T03:15:30Z

maxConnections
maxConcurrency
cMaxReuseTimes
cMaxLifeTimeMs

前两个只能二选一，不能出现突破预期的行为，比如 maxConnections 不能被突破，除非受到后两个限制要开一条新连接

RPRX · 2024-09-16T03:17:46Z

前两个只能二选一，均有值则报错，均无值则不限制复用，但后两个限制仍然可以生效

mmmray · 2024-09-16T11:25:55Z

fineee, done. why restrict it though? it feels to me that this combination is fairly intuitive, although useless.

RPRX · 2024-09-16T11:45:55Z

@mmmray json 忘改了，还有字段顺序调整一下

fineee, done. why restrict it though? it feels to me that this combination is fairly intuitive, although useless.

我看了下之前的代码也没有预连接（没有 minConnections），那么 maxConnections 实际上会变得 meaningless

RPRX · 2024-09-16T12:03:04Z

关于默认值，现在相当于 maxConnections 默认为 1，其它均为 0

或者像 mux 一样改为默认由 maxConcurrency 主导，以及其它两项要不要有默认值，@mmmray ~~你怎么看~~

mmmray · 2024-09-16T12:10:27Z

Actually, the way the current code works, minConnections is more correct. It will open connections until the target has been reached. But not "pre": instead everytime a sub-connection is opened, a new connection is opened. ah, because it's not "pre" you want it not to be called "min". Ok

so i think 0 and 1 are mostly logically equivalent.

concurrency naming (without min/max) was mostly so that it can be equivalent to mux.cool settings.

I don't really know how one would dominate the other, since they are now mutually exclusive.

mmmray · 2024-09-16T12:17:52Z

Re-reading the comment, I think you're saying that maxConnections=1 default makes it impossible to set a config like "mux": {"maxConcurrency": 10}, but it's not true. You can set this kind of thing fine and it will behave exactly like mux.cool, because maxConnections can still be exceeded through setting maxConcurrency. The only thing I changed in the config is that it's not possible to explicitly set both parameters at the same time.

RPRX · 2024-09-16T12:32:22Z

我觉得如果写了 maxConnections，按字面意思它就不该被突破，否则会造成困惑，anyway 我们先不设任何默认值吧

mmmray · 2024-09-16T12:36:51Z

Then i think the current state is good. maxConnections is broken if maxConcurrency is set to a value greater than 0. The user is not permitted to do that.

The other case is {"maxConnections": 10, "cMaxReuseTimes": 1}, then it's possible to exceed the concurrency limit temporarily, but I think that's ok.

RPRX · 2024-09-16T12:43:18Z

终于合了这个 PR，感谢各位的努力！

RPRX · 2024-09-17T08:43:45Z

#3823 (comment)

) XTLS#3613 (comment) Closes XTLS#3560 (comment) --------- Co-authored-by: mmmray <142015632+mmmray@users.noreply.github.com>

mmmray reviewed Jul 29, 2024

View reviewed changes

RPRX force-pushed the main branch from b11fcf0 to 7e24239 Compare August 2, 2024 21:48

SplitHTTP Client: Multiplexing Config

bcb0cb5

ll11l1lIllIl1lll force-pushed the ya-sh-mux branch from 201d620 to bcb0cb5 Compare August 11, 2024 01:30

RPRX changed the title ~~SpclithttpClient: a confusing mux~~ SplitHTTP client: Add multiplex controller for H3 & H2 Aug 11, 2024

mmmray mentioned this pull request Aug 11, 2024

splithttp client with http3 connectivity issue #3673

Closed

3 tasks

mmmray force-pushed the main branch from f3dc545 to f0547bc Compare August 12, 2024 10:59

Fangliding force-pushed the main branch from abb654d to 8da6954 Compare August 15, 2024 11:49

RPRX force-pushed the main branch from 8da6954 to b612da2 Compare August 16, 2024 00:59

mmmray added 2 commits August 29, 2024 15:52

Merge remote-tracking branch 'origin/main' into ya-sh-mux

68057b7

regenerate pb.go

df4da35

renaming

3c31c7a

x moment

7808859

more x

01aef1b

another rename, prohibit connections and concurrency

b9ffc93

fix json parsing, fix order

bc7cf58

RPRX changed the title ~~SplitHTTP client: Add multiplex controller for H3 & H2~~ SplitHTTP client: Add xmux (multiplex controller) for H3 & H2 Sep 16, 2024

RPRX merged commit b1c6471 into XTLS:main Sep 16, 2024
36 checks passed

av1934413 approved these changes Sep 21, 2024

View reviewed changes

SplitHTTP client: Add xmux (multiplex controller) for H3 & H2 #3613

SplitHTTP client: Add xmux (multiplex controller) for H3 & H2 #3613

Conversation

ll11l1lIllIl1lll commented Jul 29, 2024 • edited Loading

yuhan6665 commented Jul 29, 2024

Fangliding commented Jul 29, 2024

mmmray left a comment

Choose a reason for hiding this comment

mmmray Jul 29, 2024

Choose a reason for hiding this comment

RPRX commented Jul 30, 2024

dantesieg commented Jul 30, 2024

LearZhou commented Aug 2, 2024 • edited Loading

RPRX commented Aug 7, 2024

RPRX commented Aug 10, 2024

RPRX commented Aug 11, 2024

mmmray commented Aug 11, 2024

RPRX commented Aug 16, 2024

RPRX commented Aug 16, 2024

RPRX commented Aug 16, 2024

RPRX commented Aug 16, 2024

mmmray commented Aug 16, 2024 • edited Loading

Fangliding commented Aug 16, 2024 • edited Loading

RPRX commented Aug 23, 2024

RPRX commented Aug 28, 2024

APT-ZERO commented Aug 28, 2024

mmmray commented Aug 28, 2024

RPRX commented Sep 15, 2024

mmmray commented Sep 15, 2024

RPRX commented Sep 15, 2024

mmmray commented Sep 15, 2024

mmmray commented Sep 15, 2024 • edited Loading

RPRX commented Sep 15, 2024

Fangliding commented Sep 15, 2024 • edited Loading

RPRX commented Sep 15, 2024

RPRX commented Sep 16, 2024

RPRX commented Sep 16, 2024

mmmray commented Sep 16, 2024

RPRX commented Sep 16, 2024 • edited Loading

RPRX commented Sep 16, 2024 • edited Loading

mmmray commented Sep 16, 2024 • edited Loading

mmmray commented Sep 16, 2024

RPRX commented Sep 16, 2024

mmmray commented Sep 16, 2024

RPRX commented Sep 16, 2024

RPRX commented Sep 17, 2024

ll11l1lIllIl1lll commented Jul 29, 2024 •

edited

Loading

LearZhou commented Aug 2, 2024 •

edited

Loading

mmmray commented Aug 16, 2024 •

edited

Loading

Fangliding commented Aug 16, 2024 •

edited

Loading

mmmray commented Sep 15, 2024 •

edited

Loading

Fangliding commented Sep 15, 2024 •

edited

Loading

RPRX commented Sep 16, 2024 •

edited

Loading

RPRX commented Sep 16, 2024 •

edited

Loading

mmmray commented Sep 16, 2024 •

edited

Loading