tree: 21caa58ca9aff214c94348694799e6668ec58fb2 [path history] [tgz]
  1. amqp/
  2. electron/
  3. internal/
  4. proton/
  5. readme-go-get.md
  6. README.md
go/pkg/README.md

Qpid Go packages for AMQP

These packages provide Go support for sending and receiving AMQP messages in client or server applications. Reference documentation is available at: https://godoc.org/github.com/apache/qpid-proton

They require the proton-C library and header files to be installed. On many platforms it is available pre-packaged, for example on Fedora

dnf install qpid-proton-c-devel

If you built proton from source, you can set environment variables to find the built libraries and headers as follows:

source <build-directory>/config.sh

If you have installed the library and headers in non-standard directories, then add them to the following environment variables:

LD_LIBRARY_PATH  # directory containing the library
LIBRARY_PATH     # directory containing the library
C_INCLUDE_PATH   # directory containing the proton/ subdirectory with header files

There are 3 packages:

amqp provides functions to convert AMQP messages and data types to and from Go data types. Used by both the proton and electron packages to manage AMQP data.

electron is a simple, concurrent-safe API for sending and receiving messages. It can be used with goroutines and channels to build concurrent AMQP clients and servers.

proton is an event-driven, concurrent-unsafe package that closely follows the proton C API. Most Go programmers will find the electron package easier to use.

See the examples to help you get started.

Feedback is encouraged at:

Why two APIs?

The proton API is a direct mapping of the proton C library into Go. It is usable but not very natural for a Go programmer because it takes an event-driven approach and has no built-in support for concurrent use. electron uses proton internally but provides a more Go-like API that is safe to use from multiple concurrent goroutines.

Go encourages programs to be structured as concurrent goroutines that communicate via channels. Go literature distinguishes between:

  • concurrency: “keeping track of things that could be done in parallel”
  • parallelism: “actually doing things in parallel on multiple CPUs or cores”

A Go program expresses concurrency by starting goroutines for potentially concurrent tasks. The Go runtime schedules the activity of goroutines onto a small number (possibly one) of actual parallel executions.

Even with no hardware parallelism, goroutine concurrency lets the Go runtime order unpredictable events like file descriptors being readable/writable, channels having data, timers firing etc. Go automatically takes care of switching out goroutines that block or sleep so it is normal to write code in terms of blocking calls.

By contrast, event-driven programming is based on polling mechanisms like select, poll or epoll. These also dispatch unpredictably ordered events to a single thread or a small thread pool. However this requires a different style of programming: “event-driven” or “reactive” programming. Go developers call it “inside-out” programming. In an event-driven program blocking is a big problem as it consumes a scarce thread of execution, so actions that take time to complete have to be re-structured in terms of multiple events.

The promise of Go is that you can express your program in concurrent, sequential terms and the Go runtime will turn it inside-out for you. You can start goroutines for all concurrent activities. They can loop forever or block for as long as they need waiting for timers, IO or any unpredictable event. Go will interleave and schedule them efficiently onto the available parallel hardware.

For example: in the electron API, you can send a message and wait for it to be acknowledged in a single function. All the information about the message, why you sent it, and what to do when it is acknowledged can be held in local variables, all the code is in a simple sequence. Other goroutines in your program can be sending and receiving messages concurrently, they are not blocked.

In the proton API, an event handler that sends a message must return immediately, it cannot block the event loop to wait for acknowledgement. Acknowledgement is a separate event, so the code for handling it is in a different event handler. Context information about the message has to be stored in some non-local variable that both functions can find. This makes the code harder to follow.

The proton API is important because it is the foundation for the electron API, and may be useful for programs that need to be close to the original C library for some reason. However the electron API hides the event-driven details behind simple, sequential, concurrent-safe methods that can be called from arbitrary goroutines. Under the covers, data is passed through channels to dedicated proton goroutines so user goroutines can work concurrently with the proton event-loop.

New to Go?

If you are new to Go then these are a good place to start:

Then look at the tools and docs at http://golang.org as you need them.