@pebula/attribus
v0.0.1
Published
[![Documentation](https://img.shields.io/badge/Documentation-9cf?style=for-the-badge)](https://shlomiassaf.github.io/pebula-node/attribus/docs/getting-started/introduction/) ![npm (scoped)](https://img.shields.io/npm/v/@pebula/attribus?color=green&style
Downloads
4
Maintainers
Readme
@pebula/attribus
AttriBus (Attribute Driven ServiceBus) is a library that provide the tools to create an attribute driven service bus architecture. It simplifies how we use and manage service bus through an declarative API that automate message handling & emitter registration:
:::info AttriBus is a wrapper around the Azure Service Bus JS client library :::
Message Handling
Handling incoming service bus message is now as easy as handling incoming REST requests.
export class ServiceBusController {
@Queue<MethodDecorator>(({
name: 'attribus-queue.demo'
})
async myQueueEntity(context: SbContext) {
}
}
Client Emitters
Defining service bus emitters (Sender
) requires no logic, you can compose a client emitter in seconds:
export class ServiceBusEmitClient {
@QueueEmitter({
name: 'attribus-queue.demo'
})
myQueueEntity: Sender;
}
Easy Setup
The library also helps with service bus entity management through a declarative API for automatic creation of service bus entities based on the entities we define in the controllers / services in our application.
We call it provision.
This allows a code-first approach for service bus management and architecture.
Leverage Advanced Architecture Patterns
By simplifying the work with service bus, AttriBus enables the use of more advanced architecture patterns that you can use to manage the flow of your application.
Using Observables you can now implement a CQRS or Event Sourcing models out of the box. In addition, Observables along with Interceptors allow us to use plugins that encapsulate behaviors for us.
For example, the SbBackoffRetry plugin wraps incoming service bus messages and handles them when their handler throws. It will perform a retry to handle them again but with a delay between each try. The delay is not constant and it is based on an algorithm (linear / exponential) so the effect of periodic stress on you server is reduced.