Don’t start with a monolith
… when your goal is a microservices architecture
如果你的目标是微服务，就不要从单体应用开始。
In the last few months, I’ve heard repeatedly that the only way to get to a successful microservices architecture is by starting with a monolith first. To paraphrase Simon Brown: If you can’t build a well-structured monolith, what makes you think you can build a well-structured set of microservices? The most recent – and, as usual, very convincing – rendering of this argument comes from Martin Fowler on this very site. As I had a chance to comment on an earlier draft, I had some time to think about this. And I did, especially because I usually find myself in agreement with him, and some others whose views I typically share seemed to agree with him, too.
最近几个月里，我多次听说要成功架构微服务的唯一方法是从单体应用开始。用Simon Brown的话说：如果你不能构建一个架构清晰的单体应用，你怎么能相信你能构建一个架构清晰的微服务呢？最近在Martin Fowler的个人网站上对此话题也有所讨论。
I’m firmly convinced that starting with a monolith is usually exactly the wrong thing to do.
Starting to build a new system is exactly the time when you should be thinking about carving it up into pieces. I strongly disagree with the idea that you can postpone this, as expressed by Sam Newman, again someone I agree with 95% of the time:
I remain convinced that it is much easier to partition an existing, “brownfield” system than to do so up front with a new, greenfield system. You have more to work with. You have code you can examine, you can speak to people who use and maintain the system. You also know what ‘good’ looks like - you have a working system to change, making it easier for you to know when you may have got something wrong or been too aggressive in your decision making process.
— Sam Newman
In the majority of cases, it will be awfully hard, if not outright impossible, to cut up an existing monolith this way. (That doesn’t mean it’s always impossible, but that’s a topic for a future post.) There is some common ground in that I agree you should know the domain you’re building a system for very well before trying to partition it, though: In my view, the ideal scenario is one where you’re building a second version of an existing system.
If you are actually able to build a well-structured monolith, you probably don’t need microservices in the first place. Which is OK! I definitely agree with Martin: You shouldn’t introduce the complexity of additional distribution into your system if you don’t have a very good reason for doing so.
(So what would be a good reason? There are many, but to me the most important one is to allow for fast, independent delivery of individual parts within a larger system. Microservices’ main benefit, in my view, is enabling parallel development by establishing a hard-to-cross boundary between different parts of your system. By doing this, you make it hard – or at least harder – to do the wrong thing: Namely, connecting parts that shouldn’t be connected, and coupling those that need to be connected too tightly. In theory, you don’t need microservices for this if you simply have the discipline to follow clear rules and establish clear boundaries within your monolithic application; in practice, I’ve found this to be the case only very rarely.)
You might be tempted to assume there are a number of nicely separated microservices hiding in your monolith, just waiting to be extracted. In reality, though, it’s extremely hard to avoid creating lots of connections, planned and unplanned. In fact the whole point of the microservices approach is to make it hard to create something like this.
But if you start with a monolith, the parts will become extremely tightly coupled to each other. That’s the very definition of a monolith. The parts will rely on features of the platform they all use. They’ll communicate based on abstractions that are shared because they all use the same libraries. They’ll communicate using means that are only available when they are hosted in the same process. And these are only the technical aspects! Far worse than that, the parts will (almost) freely share domain objects, rely on the same, shared persistence model, assume database transactions are readily available so that there’s no need for compensation … Even the very fact that it’s easy to refactor things and move them around – all in the convenience of your IDE’s view of a single project – is what makes it extremely hard to cut things apart again. It’s extremely hard to split up an existing monolith into separate pieces.
I strongly believe – and experience from a number of our recent projects confirms this – that when you start out, you should think about the subsystems you build, and build them as independently of each other as possible. Of course you should only do this if you believe your system is large enough to warrant this. If it’s just you and one of your co-workers building something over the course of a few weeks, it’s entirely possible that you don’t.
But starting with an approach where you carve up your system into smaller parts, and treat each of them as a clearly separated, individual system with its own development, deployment, and delivery cycle, and (the possibility of) its own internal architecture, is a very powerful concept that can help you to deliver a system in the first place.
So is there any actual experience to back this up? Yes, there are a few systems we’ve been involved with recently that showed this concept to work – provided you tolerate the fact that what I’m talking about is more likely bigger than your typical microservice. The most prominent one is Otto.de, about which I did a talk together with their tech lead (and which you can read about in this nice write-up by one of their lead architects). But there are quite a few others as well, and I remain convinced it’s a good idea to start building a system using this approach – given you know the domain you’re building for really, really well.
But there is another lesson to be learned from this discussion, in my view – and it’s a more general one: Beware of architectural recipes that are too simple and too obvious. This one – start by carving up your domain into separate, independent parts – is no exception. Sometimes a monolith is preferable, sometime it’s not. If you decide to build things using a microservices approach, you need to be aware that while it will be a lot easier to make localized decisions in each individual part, it will be much harder to change the very boundaries that enable this. Refactoring in the small becomes easier, refactoring in the large becomes much harder.
As is always the case with architecture discussions, there is no way to get around the fact that you need to make that decision on your own, in each and every individual case.