Wangle – an asynchronous C++ networking and RPC library

https://code.facebook.com/posts/215466732167400/wangle-an-asynchronous-c-networking-and-rpc-library/

by doppp • 10 years ago

164 72 10 years ago

code.facebook.com

x5n12• 10 years ago

In 2016 I would write networking and RPC in go.

swah2• 10 years ago

Me too - and I don't even know C++.

swah2• 10 years ago

I was joking that the main reason for dismissing C++ for this task so quickly was not knowing it, and when and where C++ is used.

How did I came through, to be so unvoted? I'm confused.

tveita• 10 years ago

"networking and rpc" must be written in the languages your server and client are written in, (or something compatible).

Writing a networking library in Go does nothing for Facebook if their services are written in C++.

thrownaway2424• 10 years ago

Because you want to spend 10x as many CPUs?

x5n12• 10 years ago

Mostly because it's easier, performant, and because of the nature of the language it's less likely I would make a mistake doing anything in the language compared to C++ or C.

educar3• 10 years ago

Had to look up the meaning of wangle. "obtain (something that is desired) by persuading others to comply or by manipulating events."

Does anyone know where facebook is coming up with all these words - folly/wangle?

riyadparvez• 10 years ago

Folly is a common word, I think. As non-native English speaker I know the meaning of folly. But I saw wangle for the first time.

educar3• 10 years ago

Yup, I know folly as well (non-native speaker as well). But it doesn't make sense to name a library 'folly' does it?

What I was wondering is if this some self-deprecating humor or these words are popular phrases from some popular book/meme/play.

DougN7• 10 years ago

I was going to correct you and say you meant "wrangle". Turns out they can have the same meaning! Wonder if there are other examples of two words being one letter off that mean the same thing.

cgh• 10 years ago

"Folly" is a relatively common English word. "Wangle" is less common but I still see it from time to time.

What I find more interesting than the naming is FB's willingness to remove features from Folly as they are made redundant by Boost or std. I'm unaccustomed to such a lack of arrogance from a major player.

educar3• 10 years ago

Doesn't removing features break backward compatibility?

My favorite C++ library is Qt; they never remove public API for that reason. In Qt world, we update regularly to get the bug fixes and other features. Last thing I want to do is port my code to some other new API to update the library.

420TOKER• 10 years ago

beats me i wonder the same thing all the time

ot• 10 years ago

"folly" is a play on Facebook Open-source LibrarY. I think using uncommon English words helps with searchability, although of course is not as good as made-up words or acronyms.

fugalh• 10 years ago

When I name things I look for a word that doesn't get drowned out by unrelated hits in search engines, and has some kind of link to what the library does. Straightforward names are great when they are obvious, easy to remember, and not a search disaster. Otherwise clever memorable names make a pretty good substitute.

I picked the name Wangle because it is a synonym to Finagle, and because I couldn't think of a simple self-descriptive name. I think I also had the Edward Lear poem in the back of my subconscious.

We don't have a central naming authority - people just name their projects whatever they want. Sometimes we have to or want to rename things for various reasons but it's mostly similar to the open source world.

floatboth• 10 years ago

https://www.urbandictionary.com/define.php?term=Wangle

There are many meanings, including:

A combination of the words wang + angle. Noun: A curved/bent/crooked penis.

je42• 10 years ago

does anybody know why this library is not based on (boost::)asio ?

vvanders3• 10 years ago

I'll take a shot in the dark and say they probably value their compile times. Every boost library I've ever used has the tendency to pull in the world from a single header and take forever to compile.

lhecker• 10 years ago

ASIO can be used standalone with any C++11 compiler and the compile times are quite good.

If I should make a guess then I'd say that the original author of that part of the library either started before ASIO was standalone and/or simply has a hatred for boost (which would be unwarranted), or he/she was simply more experienced with libevent...

IMHO it's quite sad that they didn't use ASIO though as it's far more extensive.

kinofcain• 10 years ago

They actually are pulling in boost. System thread and filesystem but not asio, according to the cmake file.

The other commenters are probably right that they wanted to integrate with folly's existing async infrastructure.

jpetso• 10 years ago

Companies tend to stick to their own ecosystems, and in this case Facebook's folly library has already bought into libevent, for better or worse. Hopefully once asio gets standardized more libraries will switch over, but until then there are a number of legitimate issues:

* There's a Boost version and a stand-alone version, you might want to use one but not the other: * Boost requires linking against Boost.System, making it not header-only anymore. * Stand-alone isn't as widely used and if you've already bought into Boost, you might not want an extra dependency.

* Asio's API documentation exists, but that's about the best you can say about it.

* The standard version is likely to resemble the library but won't be quite the same. When it gets standardized, you still have to adapt your code. One might reasonably argue it doesn't matter much whether to use one dependency vs. another if neither binary nor source compatibility can be preserved.

* The Networking TS didn't make the standard for C++17 so it'll still be a long time until the C++ world converges around it.

* Chris Kohlhoff's complete absence from both the Boost and stand-alone GitHub project's issue queues raises questions about the state of maintenance, there are lots of pull requests pending without upstream action. (The one email I sent him regarding his executors proposal also never got a reply.) Modifying their own library would be much easier for Facebook than getting asio patches upstream.

Personally I'd like to see a world where new C++ projects depend on asio, but it's understandable if others feel differently about it.

thepumpkin19792• 10 years ago

I just tried to compile folly in OSX and it fails to compile with homebrew. https://github.com/facebook/folly/issues/400

Also, Once I manage to compile folly and then wangle, what kind of database drivers can I use in conjunction with this? Would I shoot myself in the foot if I use this with libpq under CPUThreadPoolExecutor?

Same with Rust and dlang, I'm not sure how to apply the freedom to have pluggable executor into something useful in real life, I think this is the main reason Golang is killing it and it keeps getting better.

For now I have my money on Go but can't wait for http://www.ponylang.org/ to become stable.

Edit: Issue reported has been closed, I can now proceed to install wangle https://github.com/facebook/folly/issues/400

Edit 2: Found another issue, this time installing wangle https://github.com/facebook/wangle/issues/37

pcwalton4• 10 years ago

Go uses synchronous I/O backed with an M:N implementation. Rust and D use synchronous I/O backed with a 1:1 implementation. Semantically, there is no difference between the two. The difference is in the implementation, and 1:1 threading is not as slow as you think on Linux.

tptacek• 10 years ago

Why are we talking about Go, Rust, and D on a thread about a new C++ library? I mean, I know the answer, but I wish we weren't.

I ordinarily wouldn't care, but I usually feel like I have something to learn from ambitious networking libraries even when they're written in languages I prefer not to use.

dunkelheit2• 10 years ago

Does that mean that one can use system threads as frivolously as goroutines (e.g. spawning new thread for each incoming request)? Conventional C++ wisdom advices against it. But then it may be informed by some old and bad pthreads implementations.

zzzcpan• 10 years ago

what kind of database drivers can I use in conjunction with this?

Usually synchronous database drivers have to be reimplemented in an asynchronous fashion to work with event loops. It's not very hard though, could be as simple as adding jump tables and context structures into synchronous functions.

Emulating asynchronous execution with a thread is ok too. You do that to access filesystem anyway.

But if your first thought is about accessing a database server - you probably don't need to go asynchronous at all.

malkia• 10 years ago

Emulating asynchronous execution with a thread is ok too

Straying away from the topic, but I used to support a file streaming library for the game studio I worked with, and while this approach worked, some things were not gracefully handled - for example canceling reading in the middle, or somehow giving all your I/O requests to the underlying system, so it can decide what's the best way to read them (aio?).

jhartmann3• 10 years ago

The wangle native way of doing this is just to use an IOThreadPoolExecutor: https://github.com/facebook/wangle/blob/master/wangle/concur... Just make a block of threads to run your db tasks, and you can use a promise/future to return the results back to the async code.

fredemmott• 10 years ago

There's https://github.com/facebook/squangle for MySQL, though it's lacking documentation, and requires a https://github.com/facebook/mysql-5.6 or WebScaleSQL client library

jcoffland• 10 years ago

MariaDB has an asynchronous C API.

sreque• 10 years ago

The problem is the exact opposite. The question you should be asking, is: how can I use my go code in Rust/D/C++? The answer is, not very easy, given that go's runtime is largely incompatible with C.

At least with C++, you can easily write C shim layers and your code becomes compatible with most runtimes in existence. The opposite is not true with Go.

euyyn• 10 years ago

I like that they explained the reasons behind the design decisions of the library. Looking forward to the post about the RPC part!

jhartmann3• 10 years ago

I have used this library pretty heavily in some of my recent work and I love how easy it makes building performant async network servers in C++. Minor nitpick: It would be nice though, if they built shared libraries by default though, last release I took I had to tell cmake to make a shared library. Since this is used by fbthrift as a dependency, I think it really makes sense to build it that way by default. I also use fblualib and it builds with a very old version of some of these libraries probably because the FAIR team didn't want to mess with tweaking the builds so they can work with standalone wangle (parts of the library used to be part of folly).

soconfused1• 10 years ago

would this be useful for game development?

jfasi2• 10 years ago

I can't speak for Wangle, but I've used Google's version (GRPC) to implement the networking transport for an N64 netplay hack[1] I wrote, and it was awesome.

[1] https://github.com/water-works

averageJoeCoder• 10 years ago

I'm wondering, how does C++ fare against Nodejs now, if this library is implemented for creating servers?

ec109685• 10 years ago

Node JS unfortunately doesn't have the notion of separate IO and CPU threads.

rburhum4• 10 years ago

After most compilers started properly supporting the newer C++ standards, it seems there has been an increase in the amount of love that C++ libraries are getting. A lot of people like the to say that [x] is superior to C or C++ - often ignoring that their favorite language's libraries/runtime/compilers/vm is most likely implemented in the very same languages they are putting down.

derefr• 10 years ago

often ignoring that their favorite language's libraries/runtime/compilers/vm is most likely implemented in the very same languages they are putting down

I don't see how that's an argument against a language being "superior to C or C++". You can't create a self-hosting compiler for your language without having already created a non-self-hosting compiler for your language. That means every language is going to have to be "bootstrapped" in terms of something else first. C is the usual candidate just because there are a plethora of useful compiler/runtime/vm-ish libraries people like to rely on for "language prototyping" that expose C bindings—and there are very few other languages than C/C++ that have zero-overhead, zero-friction C binding support.

Rust is one of those languages—and it does indeed look like things are going in the direction of new languages getting prototyped/bootstrapped with a Rust compiler/runtime/vm instead.

santaclaus2• 10 years ago

Rust is one of those languages—and it does indeed look like things are going in the direction of new languages getting prototyped/bootstrapped with a Rust compiler/runtime/vm instead.

I'm curious what examples you have in mind here? LLVM seems to be the default go-to compiler infrastructure, at the moment.

pjmlp6• 10 years ago

After most compilers started properly supporting the newer C++ standards, it seems there has been an increase in the amount of love that C++ libraries are getting. Sadly there are many more compilers out there than just gcc, clang, icc and msvc++.

Many of those aren't still C++11 compliant, specially in mainframes, classical commercial UNIXes, embedded real time OS and certified compilers.

But it is true, thanks to the productivity of latest standards, there has been an increase in C++ visibility, specially thanks to Microsoft.

People like to bash their C++'s compliance, but they are the commercial vendor with the more up to date support, even than icc last time I checked.

Also they were the main sponsors for Back to native conferences and eventually making it together with CppCon.

https://blogs.msdn.microsoft.com/vcblog/2014/04/03/cppcon-th...

Also Windows Phone is the only mobile OS where C++ is a first class language.

. A lot of people like the to say that [x] is superior to C or C++ - often ignoring that their favorite language's libraries/runtime/compilers/vm is most likely implemented in the very same languages they are putting down.

A lot of people like to point out that fact, ignoring that many compiler writers tend to use C or C++ out of convenience for existing tools instead of bootraping the full eco-system around a new programming language.

Many use the same route as Go, using one of those languages and when the toolchain gets mature, those parts get re-writen.

All in all, I do like the language, now if we could get rid of those C underpinnings...

santaclaus2• 10 years ago

Intel's C++11 support continues to be quite buggy. I'm still running into with issues with Intel's support for auto type deduction and for uniform initialization in non-obscure use scenarios.

vvanders3• 10 years ago

Windows phone, first class C++?

Man the times have changed. I remember when they made native code verboten in the Windows Mobile -> Windows Phone transition. Killed off a whole product line I was working on at the time.

chetanahuja2• 10 years ago

"Also Windows Phone is the only mobile OS where C++ is a first class language."

Umm... I don't know how you define "first class language" but our SDK (networking stack) compiles C++ code right in the same XCode project with ObjC code without any sort of translation layers etc (a la JNI in Java) between the two.

vardump• 10 years ago

So Wangle is (mostly) a protocol agnostic platform for building asynchronous network clients and services in C++.

What transport layers are supported and can it be extended?

What I can gather, it only supports IPv4/IPv6 TCP or UDP transports. TLS is also supported.

Notably, local pipes seem to be unsupported. Those can be sometimes useful for performance and security/isolation reasons.

jfasi2• 10 years ago

The first thing that comes to mind after reading this is GRPC (https://github.com/grpc/grpc). That library has the advantage of supporting just about every language, not just C++.

eva1984• 10 years ago

Good that they mentioned Netty as inspiration

chetanahuja2• 10 years ago

"Wangle is heavily influenced by the JVM-based libraries Netty,"

Stopped reading there. Have experience with using Netty in a java server application and engineers dread being assigned a bug or enhancement in that codebase. There's a tendency throw around the word "over-engineered" lightly but netty is a real, live and inexplicably popular example of exactly what that word means.

ausjke• 10 years ago

Good to see C++ is gaining more visibilities these days, considering the Oracle/Java combination I hope C++ can take away some JAVA dominance, at least nobody is going to sue you over C++ APIs.

Failed to compile with CMake 2.8

    CMake Error at /usr/share/cmake-2.8/Modules FindPackageHandleStandardArgs.cmake:91 (MESSAGE):
    REQUIRED_ARGS (missing: FOLLY_INCLUDE_DIR  OLLY_LIBRARIES)
    Call Stack (most recent call first):
   /usr/share/cmake-2.8/Module/FindPackageHandleStandardArgs.cmake:252(_FPHSA_FAILURE_MESSAGE)
   cmake/FindFolly.cmake:23 (FIND_PACKAGE_HANDLE_STANDARD_ARGS)
   CMakeLists.txt:17 (find_package)

sametmax• 10 years ago

Also, check out crossbar.io and the WAMP protocol, offering transparent, clean and yet simple PUB/SUB & RPC accross languages, including C++, Python, PHP and JS (node and yes, browsers).

Matthias2472• 10 years ago

WAMP is a protocol, Wangle is a framework that you can use to implement protocols (like WAMP), crossbar is a broker (an application that implements the protocol using a [python] framework. So it's not a direct comparison.

Wangle – an asynchronous C++ networking and RPC library

Related Stories