Pub/Sub Message Primitives for Go
This library contains the basic primitives for developing pub-sub systems.
Messages are published to Topics. Servers subscribe to Messages.
These primitives specify abstract behavior of pub-sub; they do not specify implementation. A Message could exist in an in-memory array, a file, a key/value store like RabbitMQ, or even something like Amazon SNS/SQS or Google Pub/Sub. In order to tap into that backend, a concrete implementation must be written for it.
Here's a list of backends that are currently supported:
| Backend | Link |
|---|---|
| Channel-based | go-msg/mem |
| Amazon AWS (SNS,SQS) | go-aws-msg |
A backend simply represents the infrastructure behind a pub-sub system. This is where Messages live.
Examples could include a key/value store, Google Pub/Sub, or Amazon SNS + SQS.
A Message represents a discrete unit of data. It contains a body and a list of attributes. Attributes can be used to distinguish unique properties of a Message, including how to read the body. More on that in decorator patterns.
A Message is published to a Topic. A Topic writes the body and attributes of a Message to a backend using a MessageWriter. A MessageWriter may only be used for one Message, much like a net/http ResponseWriter
When the MessageWriter is closed, the data that was written to it will be published to that backend and it will no longer be able to be used.
A Server subscribes to Messages from a backend. It's important to note that a Server must know how to convert raw data to a Message - this will be unique to each backend. For example, the way you read message attributes from a file is very different from how you read them from SQS. A Server is always live, so it will continue to block indefinitely while it is waiting for messages until it is shut down.
When a Message is created, the Server passes it to a Receiver for processing. This is similar to how net/http Handler works. A Receiver may return an error if it was unable to process the Message. This will indicate to the Server that the Message must be retried. The specifics to this retry logic will be specific to each backend.
This library was originally conceived because we needed a way to reduce copy-pasted code across our pub-sub systems and we wanted to try out other infrastructures.
These primitives allow us to achieve both of those goals. Want to try out Kafka instead of AWS? No problem! Just write a library that utilizes these primitives and the Kafka SDK.
What these primitives or any implementation of these primitives DO NOT DO is mask or replace all of the functionality of all infrastructures. If you want to use a particular feature of AWS that does not fit with these primitives, that's OK. It might make sense to add that feature to the primitives, it might not. We encourage you to open an issue to discuss such additions.
Aside from the code re-use benefits, there's a number of other features which we believe are useful, including:
-
Concrete implementations can be written once and distributed as libraries.
-
Built-in concurrency controls into Server.
-
Context deadlines and cancellations. This allows for clean shutdowns to prevent data loss.
-
Transaction-based Receivers.