RFC driven development
Usually we know RFCs from IETF. It’s that document that for example starts to present the HTML spec. But on this article I’m talking about RFCs (request for comments) in a broader sense. Like a tool that can be used by product/engineering teams to better plan and be productive.
A great example of a project that has an open process for receiving RFCs is Rust. It’s very interesting how they have an open and collaborative process to change their language and their tools. Could we also use a similar process on typical engineering teams?
RFC process example
If we’re going to use a RFC before starting to implement a feature, we may ask someone to write that first draft. This could be a tech lead. But sometimes, specially if we have a multidisciplinary team, one person may not have all the required context. So we may ask people from several roles to write their own RFC based on their responsibilities.
A backend developer may focus on the data model, database representation, and business logic. A frontender may focus on the pages and flows to implement, on forms and validations and on state management. A quality engineer may focus on what integration tests will be required, how to check the impact of the feature on the rest of the application, on how to test that everything is working and even on how to assess performance requirements.
This initial research allows us to take a look at the code/module that we’ll need to change, to gain context and have a clear picture of what we may need to implement.
Given that the team, and maybe some other stakeholders, will review the RFC, we get a change of validating dependencies and constraints before starting to implement the new shiny feature.
After this work is done, we may also be more confident on providing estimates on the work to be executed.
The idea is not to have everything sorted out and planned to the detail. Is to gather context and knowledge of what needs to be done and be better prepared when coding starts.
Advantages of using RFCs
This process works very well to communicate intent and get feedback before starting to develop new tasks. It allows us to have a clear picture of what’s ahead and what needs to be done. By gathering feedback we may find problems and improvement points before starting the implementation (see Zero bug policy). This may prevent that pull request review where someone just points out that:
- Some code was already implemented elsewhere
- We have some business constraints that clash with the implementation
- The implemented code will fail our performance targets
Because we have more knowledge of what we’ll need to implement, we can estimate it and if we have tight deadlines we can plan for them. Maybe consider some parallel work that we can do, or devise a plan B if things start to get delayed.
We can also share the RFC with people from other roles. Even if they don’t have all the technical context they may find issues and point out improvements. RFCs could be reviewed by all the team members, the product manager, the engineering manager and even designers or marketeers (depending on the feature).
And lets not forget other teams that depend on us. If we have stakeholder teams, they can review what we’re going to do, the interfaces and the flows, and try to raise any concerns.
Downsides of using RFCs
Creating a RFC takes time and may clash with other processes the team might have. For example with SCRUM and poker planning. This is a very different approach. And even if we may adapt it, we may struggle to incorporate it.
There’s also the question of when to do a RFC. Should we do it all the times? Even for smaller things? Sometimes we actually don’t need it. But I’ve already seen a simple bug that escalated to a full week of development and that really asked for a RFC.
Another issue is that although the RFC helps in getting context, we may actually do things differently when we start coding, because we clash with some constraint that we didn’t find before. So at this time, should we go and update the RFC? We start to have a documentation sync issue. Another way to look at this is to see the RFC just as a tool, and not as documentation. So when we start the development the RFC goes to readonly and just stays there for future context.
How much time should we invest in doing a RFC? This naturally depends on the scope of the task to accomplish. But we should a clear picture of how much time we’re going to invest. A couple of hours? A week? A problem may arise when we’re actually taking too much time. We’re doing research and going in circles because of feedback and it’s hard to get closure.
We may also have a RFC that is more easy to write for a role (QA) than another (FE). This means that for example the frontenders may start developing before the backend RFC is completed. Ideally we wouldn’t have this situation, but in an agile fast paced environment we may well face it. This may work very well if everything goes well, but may also yield unexpected problems.
Who reviews the RFC? All the team? This may be a bottleneck. A better approach could be to select specific reviewers per RFC (just like with pull requests). This can make the process slower if we need to wait for feedback.
As with every process, RFC driven development has its pros and cons. It’s not a silver bullet and will not work with all teams. But it can be a very important tool on some scenarios. I’ve been advocating and using this process for some time now and I do like it. But I also understand that it has its pitfalls and that it’s not for everyone.
Discussion and references
Author Pedro Pereira Santos
License CC BY-NC-ND 4.0