I wonder if there are others using ADRs for documenting their decisions.
We used this at my last company. We backported decisions into them as documentation. People sometimes referred to them, but not often. It was hard to take what was a huge list of docs, some of which had been obviated by others, and truly grok them all. But I don’t have a better idea for how to solve the problem. I would say it worked OK, but not great.
We use them at my company. I love how new engineers can’t just come into our codebase and start doing huge refactors. ADRs kind of force them to explain their rationale. We also use them to upvote on whether an architectural change should be implemented or not.
That sounds like you have slightly deviated from the simple ADR format (maybe not)? How do you perform the voting?
We just thumbs up the PR for an ADR if we’re in favor of it. Does that differ from the ADR format?
I see. I guess if it’s accepted then you update the Status in the PR and merge it? There isn’t a single ADR format; there are several ADR formats that are successfully being used by different teams. The format I had in mind was the one introduced by Nygard (linked article). In our case, we generally first discuss the topic and then write the accepted decision in an ADR. So, there isn’t much room for discussion/voting afterwards.
But it’s good to learn about other people using it in different ways (a learning opportunity for me). It’s interesting that you mention ADRs stopping people from refactoring. Do you have any examples you could give?
we generally first discuss the topic and then write the accepted decision in an ADR. So, there isn’t much room for discussion/voting afterwards.
This makes sense. I guess it would cut down on the noise in the PR since, at the point of its creation, it’s already been accepted. Our repos have so many other teams contributing to them that it’s hard to rally everyone to build consensus on an ADR beforehand. I guess we could hold meetings, but having feedback in a PR allows us to go back and look at what discussions led to a decision later on.
How does your team build consensus before opening the PR? Is it through slack or some other chat tool? Or maybe a meeting?
We do that in a meeting. In some cases it may take several meetings before we can make a decision. But I think those are generally fruitful meetings (I don’t see them as waste of time). PRs give people a chance to think about these on their own pace, but when it comes to decision making, I feel like meetings actually save time (provided that people are sufficiently informed).
It’s also possible that our ADRs deal with higher level decisions and is therefore easier to establish some sort of a consensus. From your description (especially the “refactoring” bit) I thought maybe you are using ADRs at a level that is closer to code. I cannot really think of our ADRs stopping people from refactoring the code in any way :)
The meetings approach is interesting. I think we did those for RFCs but I remember they were very very long and unproductive. I think the company size matters when determining the right approach.
How difficult would you expect it to be to go back and produce ADRs for significant decisions in the past that resulted in the current architecture and structure of a small-medium sized project?
It may be difficult if the context behind those decisions are lost, but I imagine it can be doable for a smaller company, assuming the amount is reasonable. And ofc, there’s always a chance the extra overhead may piss off a few engineers 😆
This is the kinda stuff I expect to find in this kind of community! ADRs are a good topic that can help teams act more mature.
And less general career questions and low-level “what technology should I learn” 🤔
We use them. They have been incredibly helpful. The concept has spread to our product team making product decisions!
We started with with ADR, but then we also started using RFCs to discuss designs. Having this sort of minimal documentation really helps us in many aspects.
This is interesting! I think conceptually we’ve tried stuff like this but never had a standard to point to. I think it also helps figuring out when and why a decision was made even if you are an “expert”, as we all forget
I like them a lot. Sometimes I use it as an agenda to sell an idea in a meeting. Othertimes I backport them after a decision has been made. It’s the context that drives a decision that easily gets lost to time.
I’ve been burned when a technical decision hurt us down the road and no one could really remember why we made it to begin with.
I haven’t seen mentioned in the post but there is a template for this here: https://github.com/adr/madr/
Also those things are pretty much mandatory when you write surgical or healthcare devices. They have weird names like SDD (detailed design) or SDS (detailed specifications). There is some info about it here: https://en.wikipedia.org/wiki/Software_design_description
The post is what started the whole ADR wave in the industry (I think there was also a conference talk by him). https://adr.github.io/ has a list of resources (templates, articles, examples, tools, etc.).