Diesel: How Python Does Comet

by z8000 • 17 years ago

44 29 17 years ago

- async I/O using epoll on Linux - for Python 2.6 - uses Python's generators - is not Twisted nor a twister or a Tornado

apgwoz• 17 years ago

Let me be the first to get out of the way all of the complaints:

1. Why didn't they just use Twisted? 2. Why didn't they predict tornado and make it better? 3. Yeah, but (tornado|twisted|my half baked implementation) is still faster at serving 'hello world' than diesel.

---

I for one welcome this new breed of choice in this space. I'm interested in "real-timing" an application I'm currently working on, so having more choices as to which way to implement it is great.

jerf• 17 years ago

Actually, "Why not Twisted?" is still a good question. If they have a good reason, let's hear it. If you want to choose between this and Twisted, that's information you need.

One huge Twisted advantage is that you can have your async comet reactors running in the same process as other Twister reactors, which can be very useful for many applications you might want Comet for. What's the payoff for using this instead?

BTW: These are straight questions, not sarcastic questions. I am perfectly willing to extend benefit of the doubt and assume there are good answers. I just don't know what they are, and I'd like to know, as I am potentially looking to use some stuff like this myself in the near future.

mcav2• 17 years ago

Twisted creates confusing, bloated code.

Every time you have to do asynchronous work, you have to mentally follow the chain of deferreds to ensure you understand exactly what code will be called at what time. Coupled with its unpythonic, deep hierarchical API, Twisted almost literally twists your code into a jumbled blob.

Unless you need the existing protocol-specific APIs twisted provides, stay far, far away. Find something more lightweight, like this, or Eventlet, or something else that lets your code remain concise and clear.

This comes from my experience writing a Twisted server for my startup, and subsequently rewriting it when the Twisted version became far too difficult and frustrating to maintain.

jamwt7• 17 years ago

It's a styles thing.

Twisted is a very capable system, very well written, with all kinds of capabilities that diesel doesn't have and never will. So it's not a capabilities thing.

It's written by a team of guys who are crazy smart, and who know async better than most people on the planet, me included. So it's not a competence thing.

The performance difference between twisted and most other Python async libraries isn't going to be significant. We're all mostly benchmarking epoll or select or whathaveyou, plus a thin layer of frames to get from there to your application code. So it's really not a performance thing either, though hackers love to talk about this anyway.

It's a style thing. Twisted doesn't feel "Pythonic" to me. It doesn't have that succinctness that makes you say "shit, I'm done already?" It's got lots of big-A architecture that's very Correct, but in practice, leaves you wondering why you're being burdened with it when that rainy day when it pays dividends never comes. Purity is allowed to triumph over practicality at every turn.

That's one man's take. I don't claim it's the universal truth, and Twisted has many very happy users. But I suspect I'm not alone in this assessment.

That being said, I will relish the day when someone can do _something_ in the Python async space and not immediately be thrust into the ring as the latest combatant in the twisted-vs-the-new-guy debate. New guys never stop coming, and that's a wonderful thing for the evolution of computing and programming.

z80002• 17 years ago

It does seem a bit messy to read, at least compared to eventlet/gevent which does some nice monkey-patching of the socket API.

japherwocky• 17 years ago

a lot of pythonistas consider monkey-patching to be exceptionally poor taste.

deno• 17 years ago

If it's used to unnecessary change language or API behavior - not for justified performance improvment.

simonw• 17 years ago

Interesting, this is the first Python async/comet library I've seen to take advantage of generator coroutines:

    def echo(remote_addr):
        their_message = yield until_eol()
        yield "you said: %s\r\n" % their_message.strip()

z80002• 17 years ago

There are actually quite a number of (yes, relatively dead) projects that do this or something very similar: eventlet, gevent, cogen, concurrence, weightless, ...

moe6• 17 years ago

Just for the record, eventlet and concurrence are not dead, I even used the latter in production.

They both have a fairly small audience but when you look at their mailing lists you'll see activity.

ddispaltro• 17 years ago

Twisted does it with inlineCallback

axod2• 17 years ago

When will people start to understand that req/s is not the full picture of how well a Comet webserver scales :/

jamwt7• 17 years ago

Hi, author of diesel here.

I can only assume you're probably referring to some of these things:

1. Response latency, esp at 99%-ish mark, under load 2. Memory usage per connection under many idle connections 3. Scalability wrt. data sharing, backing persistent data, replication/redundancy strategies, etc

I'm not sure if you were referring to us, the diesel authors, when you indicated that someone didn't understand something about Comet scaling, but I assure you that diesel does 1 and 2 quite nicely, as do most sensibly-written things based on epoll, kqueue, etc. The benchmark page is with 1k concurrent connections, and it does well with more than that, too.

If you're referring to item 3, I'm afraid diesel doesn't tackle that element of scalability yet. It's more of an I/O library, not really a framework with aspirations of providing high availability. We have some plans and quite a bit of mostly-working code that implements a paxos-based framework for achieving those goals as well, but release of that part of the framework is some months out.

Writing good unit tests for this stuff is a higher priority--unit testing async code is a PITA. We're probably going to need to steal ideas from twisted.trial or something.

Thanks for checking out diesel.

axod2• 17 years ago

Thanks for the response, and good to see some of the other issues understood. It wasn't anything personal, I haven't checked out Diesel fully yet, I just saw the page at http://dieselweb.org/lib/benchmarks/ and thought it could do with exploring all the other scalability issues...

I guess I just have "Here's a graph showing req/s for twisted/tornado/etc etc" overload lately.

It's definitely nice to see more open source options in the Comet arena :)

Oh and http://shoptalkapp.com/ looks very interesting also

PhilChristensen2• 17 years ago

Writing unit tests should be your highest priority. I am far from a TDD advocate, and compared to many I am terrible at maintaining my own tests, but I could never justify using an IO package that contains zero tests.

IO errors are some of the most annoying errors to debug, and the whole point of using a preexisting IO package is to not have to worry about errors in it. I would say you need 100% test coverage of your core library before I could consider using Diesel for a major project.

Just my $0.02...

moe6• 17 years ago

3. Scalability wrt. data sharing, backing persistent data, replication/redundancy strategies, etc

...

We have some plans and quite a bit of mostly-working code that implements a paxos-based framework

Don't. Seriously, don't go that route, it's a huge waste of time. Instead: Keep your server shared-nothing and make it interface with as many different breeds of Message Queues on the backend as possible (AMQP, Stomp etc.).

That's what any non-trivial comet app needs to do anyways and any kind of intelligence inside the comet-server beyond dispatching between the backend-queue and the browser only gets in the way.

The canonical setup is:

    Browser <-> Diesel <-> Message Queue

Diesel should be able to maintain connections to multiple Brokers on the backend in parallel, for failover and load distribution - and that's it.

cvg• 17 years ago

I just think it's funny how they compare it to the 1 week old tornado rather than twisted. Way to jump on the bandwagon. I predict a few more asynch libraries going open source in the not so distant future.

jbellis• 17 years ago

When I think of "comet" I think of "something that gives my apps the ability 'push' to browsers [even if it's really long polling under the hood]." Did I miss the browser integration story here, or are we calling all asynchronous apis "comet" now?

jamwt7• 17 years ago

Yeah, we didn't get a chance to change that headline before the project leaked out there. The next two phases of the project will get higher-level and build out our "real" goal: a bitchin' comet framework.

But you're right, it doesn't have much to do with comet at this point.

Diesel: How Python Does Comet

Related Stories