QR code

What if the Architect is Wrong?

  • Moscow, Russia
  • comments

architect

You most probably know what I think about the architect role on a software project—it’s that of a dictator who makes all technical decisions and who bears the entire responsibility for the final result. I wrote about it and even gave a talk Who is a Software Architect? at BuildStuff in 2016. However, the obvious question you may ask is: What happens if the architect is wrong? Does it mean the entire project is at risk of failure? And isn’t it better to make the whole team responsible for the result, instead of having one single point of failure?

The Color of Money (1986) by Martin Scorsese
The Color of Money (1986) by Martin Scorsese

The question indeed is obvious. Dictatorship is a great management model, provided the dictator is smart. This means, first of all, having the ability 1) to analyze the reality, 2) collect all available differing opinions, and 3) find the best possible option, leaving personal emotions aside. How many people truly can do that? None Very few.

Everybody else will most likely abuse the power and turn into a bad dictator, who doesn’t listen to anyone, doesn’t pay attention to different opinions, and makes technical decisions out of personal feelings. How many software architects like that are out there? All Many.

What is the solution?

How about we get rid of the dictator in the first place and let the team decide what the right architecture is? How about we replace dictatorship with democracy and let software developers vote somehow, so that there will be no single point of failure? They will all be responsible for the product, and when it breaks—they will all be guilty. Right?

Wrong!

“Quality and responsibility mean nothing unless they are attributed personally,” I said in my book Code Ahead. Group responsibility is the most terrible mistake a team can ever make. So, no! No voting and no democracy.

What then?

Imagine a real project, where an architect makes a decision to use MongoDB (a NoSQL database) for persisting user payments. It’s a questionable decision, since, traditionally, relational databases are considered a better option for financial data. However, we know that the architect is a dictator and we are not supposed to argue. We can’t tell the architect that the decision is wrong. Moreover, we should not ask the architect to convince us. The decision is made. It’s done.

What can we do?

We can recall that the only real boss of the architect is requirements. The architect may not listen to us, to developers, to customers, to anyone. But the requirements are the indisputable boss. Did the requirements document mention anything about the choice of the database? Most likely there was nothing about it. So the architect did everything right. The requirements said that payments have to be persisted and they are. The requirements wanted the system to process up to a hundred payments per second and it does. So, where is the mistake?

Well, we just feel that the choice of MongoDB was a bad idea. It’s just gut feeling. But maybe we are wrong and the architect is right?

To make the situation more explicit and resolve the conflict, we have to amend the requirements document. We have to say that something else is required for this particular decision. Let’s say, we add this line:

The choice of each third-party product has to be grounded on a multi-factor analysis of at least four alternatives.

Once this requirement clause is approved, the architect will have to improve the product documentation. There will have to be an analysis of MongoDB, PostgreSQL, Oracle, and some other databases. There will be some selection criteria defined and the architect will provide some numbers to make the choice of MongoDB look reasonable.

Once it’s done, the opinion of the architect will turn into a digital artifact, which may have defects. How many defects are there and how soon we find them becomes a management question, which is relatively easy to resolve. We just need a few additional pairs of eyes to look at this artifact and tell us what’s wrong with it. For example, we can ask someone from the team to review the analysis and tell us what’s wrong. Or we can hire someone from the market, who is very expensive, but a professional in the area of database management.

Once defects are reported, they will have to be resolved somehow by the architect. Either the analysis will have to be improved, or the decision will have to be changed, if the facts start stacking up against it.

The lower the risk tolerance in the project, the more pairs of eyes we need to look at the decisions the architect is making. If the quality is very important or the professional level of the architect is questionable, we need more decisions to be documented and more frequently reviewed. It’s a simple game of numbers. If the architect is completely trusted and the project is not expensive and only short-term—we just let the architect do whatever he or she wants.

On the other hand, if the architect is junior and the project is very important to us, we must demand the majority of technical decisions to be documented and analyzed. We must organize as many reviews of those documents as possible, even inviting independent experts.

Thus, to summarize my points, we must not expect the architect to be an expert capable of solving all problems. On the other hand, we must not replace the architect with a democratic vote. Both ideas are wrong. The right approach is to control the quality of decisions the architect makes via regular reviews.

Do you have such reviews in your project? If not, why not?

sixnines availability badge