🕑 Estimated reading time: 6mn
By Pierre Trollé, Core React developer @ ManoMano
Imagine an application used on many devices of different shapes, sizes and OSes. The core of this application exposes a state-of-the-art RESTful API. The API delivers the exact same resources irrespective of the requesting device, user or OS, even though all applications will not have the same needs since they do not have the same capabilities or abilities at all. Moreover, APIs tend to increase in openness and genericity of purpose. Thus, they are optimized for the business and their own constraints but no the end users’.
The Backend for Frontend approach, often seen as “BFF”, proposes to proxy the API requests and responses to adapt them to their specific use cases. This way, you could greatly reduce network calls, even compose them. Frontend teams are often closer to the business need, which make them relevant at proposing and defining such approaches. Sometimes, it will not bring you much more than a pass-through but this should not stop you. After all, a fairly common variant of BFFs are self-service APIs under the form of GraphQL.
BFFs are relevant if the number of aggregations can justify it and if the resources come out different. It is clear that the approach causes non-negligible operational and maintenance costs. However, it defines a clear boundary in terms of responsibilities between backend and frontend teams, for better or worse. Frontend teams can mock the backend using the proxy while Backend teams develop the core. In addition, Frontend teams can optimize the experience by grouping calls and only retrieving the absolute minimum from an environment that is close in terms of latency to the actual core.
All of this induces less client-side logic since the BFF absorbs a part of it. The core and the frontend are decoupled thus independent from each other and this renders middlewares a possibility. Nevertheless it remains an abstraction layer from a broader perspective.
How would you handle heavy tasks?
If they are comprised of aggregations and very simple manipulations, BFF, otherwise in the most relevant location between API and the frontend.
What about Serverless?
The cost of Lambda is often hardly justifiable and the ecosystem is very tight in resources.
How does it behave with migrations?
The BFF layer can regroup aggregations and simple tasks but will more often than not just pass-through requests and responses.
By Florian Kauder, Co-Founder @ KBDev
Redux has been initially released in 2015 by Dan Abramov. Its complete code contains just shy of 250 statement lines! It is the spiritual successor of Flux, by Facebook. The idea of it is natural: the application produces Actions that are sent to a Dispatcher which triggers operations. The operations could very well produce Actions, etc.
Dan Abramov has also been inspired by the idea of immutability, Reactive and Elm. Redux posesses Views which generate Actions sending notifications to Reducers which trigger operations related to Stores. A Reducer is similar to a state machine: from a state S, the action A shall lead you in the state S’. Of course, this only works if the implied function, the Reducer, is idempotent and pure. That is, if an input always leads to the same instructions and if they induce no side effect.
One simple advantage of this approach is that, provided the log of actions are conserved, bugs become naturally reproducible. In other words, you can publish, retrieve and replay what happened. Your store stores the current state. It combines and calls reducers and provides them with the latest version of the application state.
What if we want to do asynchronous calls? Well we can with the pattern “Loading - Success - Error”! Simply put, on action, you start by dispatching a Loading event then start your asynchronous operation. On success, you dispatch a Success event, on error or timeout, you dispatch an Error event. You may notice a caveat: a Store must present, at any given point in time, a stable state of the application. Loading events are not stable states since they imply an operation is in progress. To counter that, here are three options: only use actions that do not expose “in progress” states, avoid potentially unstable actions or identify unstable states using their metadata.
Now, what if you need a simple Undo / Redo system? This can be accomplished with two arrays that store the previous and next states. Much simpler, right?
Redux has been created at a time when React was missing most of its critical APIs. Nowadays, with the Context and the Suspense APIs and the Hooks, Redux could very well be absorbed. Most of its issues could be resolved by thinking differently about your code. Though, the reducers are a neat trick that could benefit from being extracted and reused elsewhere!