TCP has a standard API called "TCP sockets" and in our environment we don't need to worry about which library implements TCP.
However, HTTP/3 explained says QUIC has no standard API.
So, any program over QUIC depends on the library that implements QUIC.
Porting those programs to other environments can also be a hard work.
Why did QUIC Working Group make this decision?
Why did QUIC Working Group make this decision?
The QUIC working group defines the protocol. It doesn't write APIs. From there, many different groups went off and implemented the protocol independently. Each had different API requirements, so they all evolved a little differently.
You can find a (fairly out date) list of them here: https://github.com/quicwg/base-drafts/wiki/Implementations.
Porting those programs to other environments can also be a hard work.
Several implementations (e.g. msquic) are cross platform (and even cross-language) so you shouldn't have to "port" to other environments.
Related
The base problem is as follows:
Two nodes communicating over a single socket;
Request-reply pattern;
Both nodes are client and server, i.e. node A makes requests to node B and node B makes requests to node A;
This would be easily solved with two sockets but I have only one. You can think of the problem as having to create multiple virtual sockets/channels over a single socket. Do you know of a well-tested messaging library that would support such use case?
In addition:
Support for C++ and Java;
The data serialization is handled by me using Google Protocol Buffers; and
If it is possible to achieve all this using RPC then great, otherwise I'll manage using protobuf.
I'd prefer not to have to write my own library and use something which is well-tested and well-support. I've looked into ZeroMQ but it doesn't seem to support the third requirement (request-reply pattern from A to B and from B to A simultaneously over a single socket). RabbitMQ is another possibility but may not support this requirement either. (I don't have experience with these libraries so maybe I'm wrong...)
(I wonder if I'm asking for too much.)
I don't have a complete answer for you, but I note that what you're asking for is emphatically supported by Cap'n Proto's RPC protocol: calls can be initiated in either direction.
It's not exactly what you want because:
Cap'n Proto has its own serialization format. The usage model is heavily inspired by Protobufs, but it isn't protobufs.
The Java implementation does not yet support RPC (planned, though).
If you care about Windows: the C++ implementation does not yet support RPC on MSVC, mostly due to MSVC being woefully behind on C++11 support. :/
But it might be worth checking out, for inspiration if nothing else.
(Disclosure: I'm the author of Cap'n Proto, and also the author of most of Google's open source Protobuf code.)
What I want to do: I want to add communication capabilities to a couple of applications (soon to be jar libraries for Java) in Scala, and I want to do it in the most painless way, with no Tomcat, wars, paths for GET requests, RPC servers, etc.
What I have done: I've been checking a number of libraries, like Jetty, JAX-RS, Jackson, etc. But then I see the examples and they usually involve many different folders for configuration, WSDL files, etc. Most of the examples lack a main method and I don't have a clear picture about how many additional requirements may they have (e.g. Tomcat).
What I am planning to do: I'm considering to simply open a socket on the "server" to listen, then connect with the "client" and transfer some JSON, in both directions. This should be fairly standard so that I can use other programming languages in compatible ways (e.g. Python).
What I am asking: I would like to know whether there is some library that makes this easier. Not necessarily using raw sockets, but setting up some process communication in just a few lines, maybe not as simple as Node.js, but something similar.
Bonus: It would be cool to
be able to use other programming languages (e.g. Python) by using open standards
have authentication
But I don't really need any of those at this point.
I think you need RPC client/server system, I would suggest to take one of these two:
Finagle - super flexible and powerful RPC client/server from Finagle. You can define your service with Thrift, and it will generate stubs for client/server in scala. With Thrift it should be straightforward to add Python support.
Spray - much smaller library, focused on creating REST services. It's not so powerful as Finagle, however much easier. And REST allows you to use any other clients
Remotely - an elegant RPC system for reasonable people. Interesting and very promising project, however maybe difficult to start with because of extensive Scalaz+Shapeless+Macro usage
Honestly if you want something that is cross-language compatible, simple, straightforward, and concise then you do not want to use plain old sockets!
Check out dropwizard. It is amazing and I use it for small and large projects alike! It is usually configured by no more than a single configuration file. It supports authentication too!
Out of the box it gives you really great inter-process communication over JSON (using Jackson) and much much more. There is also pretty decent Scala support for dropwizard.
If you must roll your own then I'd recommend using Jackson for JSON parsing. It's super simple to use and also has great scala support.
If you've got a "controlled" use case where the client and server are on the same LAN and deployed in tandem, I'd (controversially) recommend Java RMI; it's dumb and JVM-specific (and uses a Java-specific protocol), but it's very simple to use.
If you need something more robust and cross-language, I'd recommend Apache Thrift. You write your interfaces in a platform-independent interface definition language, and it's very clear which changes are compatible and which are not; the thrift compiler generates skeleton interfaces for you to use, and then you just write an implementation of that interface and a couple of lines to start the server (as you can see from the example on the homepage). It's also got good support for async implementations if you need the performance. Thrift itself is reasonably standard and cross-platform, with its own binary protocol, or you can use JSON as a transport if you really want to (I'd recommend against that though).
RabbitMQ provides one easy way to do what you want without writing a server and implementing your own persistence, flow control, authentication, etc. You can brew or apt-get install it.
You start up a broker daemon process (i.e. manages message queues)
In the Scala producer, you can use Maven-provided Java API to send JSON strings without any fuss (e.g. no definition languages) to specified queues
Then in your other Scala program, connect to the broker, and listen for messages on the queue, and parse the incoming JSON
Because it is so popular, there are many tutorials online for different patterns you may want to use to distribute the messages, e.g. pub/sub, one-to-one, exactly-once delivery, etc.
I need to implement a distributed XMPP MuC application on the lines of XEP-0289 minus some of the features, in essence I want to have a bare bones implementation of the plugin, my concern is to address fault-tolerance and as of now I do not want to worry about the performance considerations as specified in 289.
I have looked into SleekXmpp as a tool to develop server side plugins, but don't know how comfortable it would be to use it for such an implementation, other options I have looked at are OpenFire , Tigase. I am comfortable with Python/Java and other key features to consider would be good documentation, ease of use etc keeping that in mind I would like to know what would be the preferred path to take for this development.
Any guidance will be appreciated.
you should be able to write a MUC component that includes FMUC (or similar). The general way to do this would be to use a library that supports XEP-0114 components (e.g. SleekXMPP (Python), Swiften (C++)) and implement MUC+FMUC through that. You haven't said what your concerns with SleekXMPP are, but it's a fairly well-respected library in the XMPP community, so seems a fair choice (I'd pick Swiften, but I'm biased as one of the authors).
Your second option (patching the server directly) isn't generally the XMPPish way of adding customisations (as it's vendor-specific), but should also work if you can find someone sufficiently familiar with the server code, or if you're willing to become so.
To achieve fault tolerance (assuming you mean resilience to server failures) you'd need to run your XMPP server clustered, and also cluster your FMUC implementation. With that done, the usual XMPP fail-over using SRV records in DNS should ensure other servers retry connections to another host.
On a side note, the next version of FMUC (XEP-0289) will have some of the features of the current revision stripped out, and a number of improvements made based on deployment experience, so if your work is not time-critical, it might be of benefit to you to read that when it's released. I also note that there exists at least one implementation of FMUC already (Isode's M-Link, on which I work), and there is interest from other vendors, so using the standard protocol might benefit you in terms of not re-inventing the wheel.
I am currently doing my research on building a chat system with more than 10k users connected online. I came across technologies and ways to do it such as jabber(XMPP), websockets, long polling, push. As far as I now, long polling might not work given the number of users. I know there is a lot of ways to accomplish this. I also know that facebook and Google chat systems are developed on XMPP.
I would truly appreciate if anyone could point me to the right direction. I believe all these methods and technologies out there are good depending on the scale of the project. I definitely need performance and scalability.
I've used Socket.io together with NodeJS for such a chat application. It scaled to over 10K concurrent users on moderate servers and there was a lot of room to grow.
This does depend on your limitations, tho.
What kind of hardware are you planning on using?
Which operating system would power your servers?
Which client platforms are you targeting?
Do you have an existing infrastructure you need to fit this into?
Do you have a previously selected programming language?
The existing skill sets your team members have and your team's ability to adopt new platforms and languages if necessary.
Take all of the above into consideration when making your decision.
Personally, I've found XMPP to be quite adequate, but a bit bloated for my purposes. YMMV.
You are comparing a fruit basket and three different variety of oranges.
XMPP is the only protocol that you have mentioned that actually is designed to support a chat system (of which many exist). The others are simply asynchronous messaging protocols/techniques. XMPP already supports http based chat via BOSH. Without a doubt, it will also support WebSockets when the specification is finalized. There is actually a draft of this already written, but at this point it appears to be a draft using a draft, so there will probably be few, if any, implementations.
Using XMPP would allow you to build on a proven technology for implementing a chat system and would allow you to choose what transport you want to use "under the hood". You haven't actually said whether you need a http based transport or not, but with XMPP you can use the stock tcp socket based transport or a http based one (BOSH) with the knowledge that it will also support WebSockets in the future.
The other benefit is of course that this is a widely used standard that will allow reuse of existing clients, servers and libraries in pretty much all popular (and not so popular) languages and platforms.
Scalability is not too much of a concern with the numbers you are quoting, as most (maybe all) existing xmpp servers will handle that many users.
Has anybody done or seen a deployment of Apache Thrift in an iPhone app?
I am wondering if is a reasonable solution for a high-volume, low(er)-latency network service for iPhones compared to HTTP.
One noteworthy thing I found is a bug report about running Thrift on the iPhone, which seems to have been fixed. But that doesn't necessarily indicate that it's a done deal.
Thrift and HTTP aren't mutually exclusive. In fact thrift now ships with an HTTP transport implementation to use. It's also a really nice way to auto-generate server/client code that avoids a lot of marshalling/unmarshalling boilerplate while still being really fast. Its internal representation is basically binary JSON, so it's very similar to a RESTful web service (except being easier to code and much, much faster).
So... anyone able to answer the original question? If not, I'll dive in myself with thrift's included Cocoa support and see how it works on the iphone.
Just my two cents..
The accepted answer to this question, is an opinion to not use a technology, not an answer of whether it is possible.
Thrift, is an interface definition language, IDL, like Protobuf and Capt'n'Proto. They permit the definition of a client/server/server protocol which is platform agnostic. JSON and Plist don't provide the same level of type conformance.
Having previously lead an iOS team with 10Ms MAU using Google Protobuf v2.5 on iOS, Android, Windows, and server teams, I can attest that IDLs are great on mobile. Apple uses them for syncing iWork content.
My current team uses Thrift for iOS and Android clients, with a mostly Scala backend. I much prefer it to Protobuf.
We send Thrift payloads over HTTPS and WebSockets. Once you have defined (in Thrift) your our wire communication protocol (i.e. frame structure), it's very easy to evolve your APIs.
However, on iOS in particular there are some implementation issues. The current version of the library is quite poorly packaged, and if you hope to make an Objective-C framework (e.g. for iOS 8+), then you will not be able to out of the box with v0.9.2. This is because the library headers include local imports, (#import "TProtocol.h" instead of #import <Thrift/TProtocol.h>) with no umbrella headers. Worst of all, the Objective-C compiler generates very messy Objective-C classes, also including local imports from the Thrift library.
Some of these issues are pretty damning. It indicates to me that while use of an IDL is very much a good engineering decision, not many iOS teams are using Thrift, unless they're huge with the resources to write their own library.
I've always disliked frameworks that use a common interface definition that builds out both server and client code. It keeps both sides too much in lockstep where in reality server API changes must be very flexible in the versions of clients that are communicating with it.
There are helpful libraries that make JSON or PLIST communication over HTTP pretty easy, and decades of debugging and understanding the HTTP protocol and how to use it well. I would ignore that at your peril.
I have used thrift's objective c bindings for a large iPhone app with a few million users. As one of the posters mentioned we can use Http which gets the best of both worlds. However there is no asynchronous HTTP client for thrift. We had to build an event based wrapper to allow non-blocking I/O calls. The underlying layer still issues one call at a time which hit us in a big way because we have one server call that takes a long time but it does not block UI flow and another really fast one that does block UI flow. If the underlying layer is busy with the slow command our fast command just has to wait. I am trying to build asyc http in c++ which can then be used on the iPhone but that is someways off from being ready.
Thrift as an external API doesn't make sense. Use it internally rock and roll.