Zach Tellman
2015-04-17 21:06:30 UTC
Hey all,
In preparation for Clojure/West, I'm formally releasing the latest Aleph
and the libraries that surround it. Aleph 0.4.0 has been running in
production at Factual for half a year now, and across a variety of services
is handling at peak 600k HTTP requests/sec (spread across 15-20 machines).
Since the landscape of Clojure HTTP servers is pretty crowded these days,
it's worth taking some time to explain how Aleph differs. To be clear,
most Clojure deployments likely use Jetty, and should continue to do so.
However, Aleph has some unique properties:
* It uses the Netty library, which is a high-performance and very
battle-tested network layer for the JVM
* It is the only HTTP server that has *ubiquitous* asynchronous streams
wherever data can be received or sent (all other libraries can only
represent streaming requests using InputStreams, or like http-kit don't
support streaming HTTP requests at all)
* It is the only server that has a WebSocket implementation with any
support for per-connection backpressure. I won't make this post even
longer by going into why this is important, but this will be a central
theme of my talk at Clojure/West next week if you're interested in hearing
more.
* It uses consistent abstractions to represent network connections over a
variety of protocols, which makes it straightforward to use the same
application logic for all of them.
Again, none of these points mean you should immediately drop whatever
you're using and move over to Aleph instead. However, I do feel it
represents the only (current) good option for using core.async or a similar
stream abstraction to represent network data, which is an idea a number of
people seem to be playing with lately. Some examples of this can be found
at http://ideolalia.com/aleph/literate.html.
A full list of the libraries:
aleph - https://github.com/ztellman/aleph - uses the excellent Netty
library to expose HTTP, TCP, and UDP using a consistent asynchronous stream
representation.
manifold - https://github.com/ztellman/manifold - an unopinionated stream
representation designed to cleanly interoperate with other stream
representations (Clojure's seqs, core.async channels, Java's
BlockingQueues, and others). This is the base stream representation for
all network sources and sinks in Aleph.
dirigiste - https://github.com/ztellman/dirigiste - a pure-Java library
that provides instrumented, dynamically sized thread and object pools.
This is used for thread pools in Aleph's HTTP server, and for connection
pools in Aleph's HTTP client.
byte-streams - https://github.com/ztellman/byte-streams - a means of
translating any byte representation into another. Want to turn a
core.async channel that emits byte-arrays into an InputStream, or maybe the
other way around? Look no further. The library's conversion mechanism is
extensible, which is used in Aleph to make Netty's custom byte
representations interoperable with more familiar representations.
byte-transforms - https://github.com/ztellman/byte-transforms - a curated
collection of byte compression, hashing, and encoding mechanisms, which can
work on anything byte-streams can convert.
While all these libraries are used in concert to create Aleph, I've been
very careful to make sure any of them can be used by themselves. If anyone
has questions about them, the best place to get my attention is the Aleph
mailing list: https://groups.google.com/forum/#!forum/aleph-lib.
I will be mentioning some of these libraries at my upcoming Clojure/West
talk (http://clojurewest.org/speakers#ztellman), but I've also set aside an
Unsession for specifically discussing these
libraries: https://github.com/clojurewest/clojurewest2015/wiki/Unsessions.
If you're interested, please add your name to the list.
Zach
In preparation for Clojure/West, I'm formally releasing the latest Aleph
and the libraries that surround it. Aleph 0.4.0 has been running in
production at Factual for half a year now, and across a variety of services
is handling at peak 600k HTTP requests/sec (spread across 15-20 machines).
Since the landscape of Clojure HTTP servers is pretty crowded these days,
it's worth taking some time to explain how Aleph differs. To be clear,
most Clojure deployments likely use Jetty, and should continue to do so.
However, Aleph has some unique properties:
* It uses the Netty library, which is a high-performance and very
battle-tested network layer for the JVM
* It is the only HTTP server that has *ubiquitous* asynchronous streams
wherever data can be received or sent (all other libraries can only
represent streaming requests using InputStreams, or like http-kit don't
support streaming HTTP requests at all)
* It is the only server that has a WebSocket implementation with any
support for per-connection backpressure. I won't make this post even
longer by going into why this is important, but this will be a central
theme of my talk at Clojure/West next week if you're interested in hearing
more.
* It uses consistent abstractions to represent network connections over a
variety of protocols, which makes it straightforward to use the same
application logic for all of them.
Again, none of these points mean you should immediately drop whatever
you're using and move over to Aleph instead. However, I do feel it
represents the only (current) good option for using core.async or a similar
stream abstraction to represent network data, which is an idea a number of
people seem to be playing with lately. Some examples of this can be found
at http://ideolalia.com/aleph/literate.html.
A full list of the libraries:
aleph - https://github.com/ztellman/aleph - uses the excellent Netty
library to expose HTTP, TCP, and UDP using a consistent asynchronous stream
representation.
manifold - https://github.com/ztellman/manifold - an unopinionated stream
representation designed to cleanly interoperate with other stream
representations (Clojure's seqs, core.async channels, Java's
BlockingQueues, and others). This is the base stream representation for
all network sources and sinks in Aleph.
dirigiste - https://github.com/ztellman/dirigiste - a pure-Java library
that provides instrumented, dynamically sized thread and object pools.
This is used for thread pools in Aleph's HTTP server, and for connection
pools in Aleph's HTTP client.
byte-streams - https://github.com/ztellman/byte-streams - a means of
translating any byte representation into another. Want to turn a
core.async channel that emits byte-arrays into an InputStream, or maybe the
other way around? Look no further. The library's conversion mechanism is
extensible, which is used in Aleph to make Netty's custom byte
representations interoperable with more familiar representations.
byte-transforms - https://github.com/ztellman/byte-transforms - a curated
collection of byte compression, hashing, and encoding mechanisms, which can
work on anything byte-streams can convert.
While all these libraries are used in concert to create Aleph, I've been
very careful to make sure any of them can be used by themselves. If anyone
has questions about them, the best place to get my attention is the Aleph
mailing list: https://groups.google.com/forum/#!forum/aleph-lib.
I will be mentioning some of these libraries at my upcoming Clojure/West
talk (http://clojurewest.org/speakers#ztellman), but I've also set aside an
Unsession for specifically discussing these
libraries: https://github.com/clojurewest/clojurewest2015/wiki/Unsessions.
If you're interested, please add your name to the list.
Zach
--
You received this message because you are subscribed to the Google Groups "Aleph" group.
To unsubscribe from this group and stop receiving emails from it, send an email to aleph-lib+***@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
You received this message because you are subscribed to the Google Groups "Aleph" group.
To unsubscribe from this group and stop receiving emails from it, send an email to aleph-lib+***@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.