Understanding the code inside Tornado, the asynchronous web server

by dhotson • 15 years ago

151 42 15 years ago

rdtsc7• 15 years ago

> I also feel obliged to talk a bit on nonblocking IO or asynchronous IO (AIO)

Wait, that is not correct. His IO is blocking. He gets a notification when data is ready, from the select/poll/epoll (the asynchronous part), but when IO is actually performed (read, write, recv, etc), the operation it is still happening in the main thread in user space, and it blocks it.

Currently only the file system (I think) has truly asynchronous and non-blocking IO. It is provided by the aio_* set of system calls and is has been sort of an exotic beast (it is not that popular).

Here is a good chart of the possible IO types and their combinations:

http://www.ibm.com/developerworks/linux/library/l-async/

And of course:

http://www.kegel.com/c10k.html

justincormack• 15 years ago

No the network sockets are set to non block and will return E_AGAIN once you have emptied the kernel buffer. Thats still non blocking.

I dont think he is using aio_* which Linux does not really implement usefully.

kingkilr• 15 years ago

Python doesn't provide an API to aio_* because they're basically useless (on Linux at least).

rdtsc7• 15 years ago

No. They would return E_AGAIN after the data has been read or written in blocking mode. If select comes back and says "read is ready on socket 4" then user code does read on socket 4 and gets (say) 4K of data followed by E_AGAIN. Getting that 4K of data happens in a blocking mode. To make it non-blocking you would provide a pointer to your user space buffer to the kernel and it would determine when IO is ready _and_ proceed to put that data in the buffer.

maurycy• 15 years ago

Personally, I would be more interested in understanding the code inside Tornado, a column of air. :-)

ZoFreX3• 15 years ago

> Lets say you have 20 threads. You improved performance 20 times, so the rate is now 4 request per second. Still, way too small. You can keep throwing threads at the problem, but threads are expensive in terms of memory usage and scheduling. I doubt you’ll ever reach hundreds of requests per second this way.

Let's say I'm using Java instead of Python. Lets say I use a lot more threads than 20. I will reach thousands of requests per second this way. There are real problems with that approach, and event-driven approaches have genuine advantages, but can people please stop straight-up lying and saying that you can't just throw a few thousand threads at a problem, because you can, and people do.

wildmXranat• 15 years ago

I agree with you mostly, and I think there is a misunderstanding of the terms. It's possible to serve large amounts of requests with threads as well as with event loops, but there is a distinction between the two on how they consume resources. The author is wrong in assuming threads will not scale, as they will, but a Java web framework using threads will eat up hundreds of megs under heavy load, while an event loop will keep on churning away with progressively longer response times. It's not so much a case of lying as much not knowing when to stop presuming instead of investigating the actual facts.

ZoFreX3• 15 years ago

There are resource issues either way, in both cases you have to store state. With threads this can quickly become expensive if you keep the stack size for all those threads at the default amount (a whopping 48K if memory serves). I would say it's easier to use smaller amounts of memory with event-driven programming, but you still have to keep some data around per-client. Of course, if the situation is very asynchronous (seconds instead of milliseconds) then you don't necessarily need all of this hanging around in RAM.

ericflo• 15 years ago

Yeah, whenever people say threads don't work, point them to Varnish. It uses threads intelligently and everyone agrees it's fast and awesome.

ankrgyl4• 15 years ago

Even with Python (CGI backed by Apache), you can scale with threads. The issue of scaling with threads vs. events is a pretty hot debate, and I think the author sets up this kind of criticism by addressing it poorly. Personally, I fall into the event-driven camp, because it involves the operating system as little as possible (only file descriptors). The C10K link has a great overview of the threaded approach (http://www.kegel.com/c10k.html#threaded). Read through the rest of the page for a discussion on the pros/cons of other approaches.

cdavid• 15 years ago

Another issue with evented frameworks in python is that you have to ensure nothing is blocking anywhere in your code, which becomes harder the more complex your application becomes. People will often mention IO, database, etc... forgetting that it is also an issue if your request handler takes CPU for N ms (e.g. encoding a relatively large payload in json, etc...). everything needs to be written with async in mind.

Languages/frameworks where async IO is an implementation detail (like some web frameworks in haskell) don't have this issue.

ZoFreX3• 15 years ago

I'm not entirely in either camp, I don't think event-driven programming is worse or even bad. I'm just sick of false arguments in either direction! (Especially as a Java developer who tried to learn nio only to find out that using lots of threads was fine all along... not a pleasant journey!)

jrockway2• 15 years ago

Well, actually, it involves the operating system rather heavily. select, epoll, kqueue, etc., etc., are not user-space things, after all, and if the OS implementation is bad, you are going to run into trouble.

bdarnell2• 15 years ago

Agreed. If requests per second is the metric you're optimizing for, then threads are a perfectly fine solution (see e.g. Paul Tyma's presentations about how threads are superior to event-driven approaches in Java). The strongest case for event-driven servers is for cases like long polling, where a connection will sit idle for a minute or more and the number of connected idle clients becomes an important metric.

thejbf• 15 years ago

Yes, but I can still spawn thousands of event-based Scala actors on the same machine. The actual problem with author: He believes running long and costly operations on an HTTP request is OK which is never and ever right. If you need to wait 5 seconds to perform a task, make it asynchronous.

barrkel3• 15 years ago

Something I'd like people to understand, at a very deep level: event-based async programming, and blocking thread-based programming, are fundamentally the same. The problem of maintaining your state in between event triggers is explicit in the first style; it's implicit, in the thread stack and CPU instruction pointer, in the second style. In the second style, it's the OS which is running the event loop, and it dispatches events by resuming a continuation - a continuation that starts by "returning" from the blocking call.

Much of the translation from blocking style to event-based style is moving the work of dispatching and looping from kernel to userland. Other ancillary benefits, like reduced address space usage by blocked threads, are in principle also achievable in a threading model - e.g. by storing stack frames on the heap and being more aggressive about collecting them (assuming GC).

Other benefits of async - such as overlapping work - are also fairly trivially possible with threading, though less deterministic.

_delirium• 15 years ago

That's essentially the thesis of this 1978 paper, though they group evented programming under the broader "message-passing" style: http://www.sics.se/~adam/pt/duality78.pdf

jrockway2• 15 years ago

But I'll bet that a node.js HTTP server handling 10,000 slow downloads is going to use a lot less memory than 10,000 Java+OS threads handling those same downloads.

ZeSmith• 15 years ago

Reminds me of this: http://teddziuba.com/2009/09/twisted-vs-tornado-youre-both.h...

jparise• 15 years ago

Some of this is a bit dated. It looks like it was written in late 2009, judging by the first comment. For example, IO loop timers are now stored using a heapq[1] instead of a sorted list, and many of the other issues have been similarly addressed over the last year and a half.

Still, this is a good general overview of Tornado's internals.

[1] https://github.com/facebook/tornado/commit/b6c4d6d20196fa4fe...

bdarnell2• 15 years ago

The heapq change was just last week and has not yet been in any release, so that's a perfectly reasonable omission. :) But yes, this is about a pre-1.0 release, so I would encourage anyone interested in exploring tornado internals to check out a more recent release.

dad• 15 years ago

There was a very good talk about this at #nodeconf this week by Tom from Joyent. The basic summary was that threaded network servers, like Apache, say, approach the speed problem by "pre-allocating" resources, a chunk for each thread/instance. The issue is that there are a finite number of resources on each machine (RAM mostly), and blocked threads are eating up a slice of those resources even when doing no work. This sets the upper bound of number of simultaneous connections that can be handled.

In an event based system the overhead for each connection is at least three orders of magnitude lower, sometimes four or five (hope I'm remembering this right). This translates into _considerable_ increase in number of connections that can be handled simultaneously, not quite the equivalent number of orders of magnitude due to other aspects of the system becoming more of the bottle-necks, but still dramatic.

This was a data-driven observation, not conjecture or assertion. I listened in during a break while the nodejs guys argued about approaches to getting TLS working better - they were worrying about the 1MB overhead for a TLS connection because that's a significant percentage of a connection handler. Think about that for a minute in the context of an apache threaded instance. 1MB matters? wow!

I'm new to this area and this was very interesting stuff and seemed related to the various discussions below about a threaded system can do what an event based system can do.

Understanding the code inside Tornado, the asynchronous web server

Related Stories