I get this question a lot from our customers when we recommend technologies as part of the new software development proposal. Why not MySQL instead of Postgres? Why not Kotlin instead of Java? Why not Vue.js instead of React? Why not Memcache instead of Redis? Why not Dropwizard instead of Spring Boot? Why not Flutter instead of React Native? And the list goes on. The important point to note is that the alternative is not radically different from the proposed technology. Both options more or less have the same characteristics. There are successful products built using both the technologies. In this post I want to discuss how to answer why not X questions.
In this post I am not covering situations when X is very different from proposed technology. For example, why not Cassandra instead of Postgres? Or why not Aerospike instead of Redis? Or why not Rust instead of Java?
The obvious answer to why we propose certain technologies is that we have expertise in them and it is easier to find engineers in the market for those skills. But, when people ask this question they want to hear strong technical reasons not the usual boring obvious answer. So, the obvious answer does not fly.
I am unable to sleep because of fever so I thought let me write this post. Maybe it is not the best time to write this post but who cares. A couple of weeks back I got into discussion with a customer CTO over libraries over framework. This customer wants us to prefer libraries over framework.This was mainly with regard to Spring Boot vs Dropwizard discussion.
The best definition I have read on the web about framework and library is that a framework calls your code and your code calls the library API. A framework does much more and has strong opinions. Libraries are focussed, solve one problem, and swappable (not entirely true without proper abstractions).
The customer wanted us to use Dropwizard instead of Spring Boot because of the following reasons:
Spring Boot does too much auto-magic. With something like Dropwizard you can have control over how things bind together. You can use manual dependency injection instead of a IoC container doing that magic for you. You can disable auto configuration in Spring Boot. You can also wire beans by hand if you want. But, I agree the default way is to rely on auto-configuration.
Spring Boot vulnerability surface area is higher because it is too easy to add starter jars and bring all the transitive dependencies. I think this will be mostly true with any other approach of building software in Java unless 1) you are going down the stackless way(I don’t think Java platform is there yet) 2) you have good governance on what gets in your dependencies.
Spring Boot executable size is much higher. I compared bare bones Spring Boot(spring-boot-starter-web with Tomcat) and Dropwizard(default maven archetype) executable sizes. As it turns out Spring Boot was 17M and Dropwizard was 19M.
Spring Boot startup time is higher. This depends a lot on what you are doing at startup. The bare bone spring boot app starts in 1.64 seconds whereas the bare bone Dropwizard app took 1.526 seconds to startup.
Spring Boot consumes much more memory. This was true. Spring Boot loaded 7591 classes whereas Dropwizard loaded 6255 classes. Also, heap space consumption of Spring Boot was twice compared to Dropwizard.
Spring Boot apps are difficult to debug. I agree exception stacktraces are too long at times and it takes a minute or two to reach your calling code. But, I personally never had much trouble debugging Spring Boot apps. I mostly rely on good tests and logging to debug stuff.
Lastly, they wanted us to follow a general principle – prefer libraries over frameworks.
The funny part is Spring Boot does not call itself a framework and Dropwizard documentation states Dropwizard straddles the line between a library and a framework.
We went with Dropwizard :). I respect their decision and I think their reasons have merit. I myself have seen too many badly architected/built Spring Boot apps so I am open to trying out a new, simpler, and better alternative.
I have read the Brandon Smith post – Write Libraries, not Frameworks. I also think libraries over frameworks is a good architecture principle that we should strive for. The only problem is when we apply principles blindly without understanding the context.
I think principles like favouring libraries over frameworks are fundamental. For these principles to work you need to have the right context and provide the right environment. I think it will work for you when:
You have a good engineering team that understands the cost of adding libraries. There is no free lunch. It does not work in a typical bottom heavy pyramid team structure where teams are considered feature factories. They will add any library under the roof to deliver features if your mindset is not aligned with the principle.
You have good governance. It is enforced by healthy code reviews, automation (aka fitness functions), architecture knowledge sharing sessions, Microservices production readiness reviews, and architects with the skin in the game.
You spend effort and resources on developer experience to build tooling that makes it easy to scaffold new Microservices with your opinions and choices baked in. I am not sure how it will be any different from a pure framework approach. You will end up building your own Microservices framework with your library choices.
You train your software engineers to buy into this methodology. They will have to unlearn the existing way and learn the new way to build software.
You understand productivity might take a hit till developers understand the new way to build software. Frameworks give you a productivity boost by helping you get started faster and solving common problems for you.
You use automated checks to continuously prove that software is not deviating from the principle. You can write a build tool task that fails the build if executable size reaches a threshold. You can write tests that fail the build if people use certain libraries. All of this is possible. You need to invest the time of the right engineers to make this happen.
I don’t know whether this will work in our environment or context. It will depend on if we can walk the talk. I have seen too many times all the good things thrown under the bus when business puts pressure for features.
Looks like medicine(or writing this post) has done its job. I started writing at 2:28am and now at 3:39am my fever is down and I am feeling better.
All software systems we build use some sort of configuration files. These configuration files change depending on the environment in which your service/system is deployed. They allow us build a single deployable unit that can be deployed in multiple environments without any code change. We just change our configuration file depending on the environment and provide our service path to an external location where the configuration file exists. And, our service uses the configuration file to bootstrap itself.
Configuration files become unwieldy if not managed well. Incorrect configuration values is one of the major reasons for system downtime. Most teams don’t write tests for their configurations so a lot of times bugs are discovered in higher environments.
I was also seeing the same problem in one of my projects. There was a lack of clarity on which configuration properties change between different environments and which remains the same. Also, in a local and lower environment I don’t mind database credentials in my configuration files but for a higher environment I don’t want them to be present in the code.
Most of us are building systems that sit on top of other third party systems. This is common in the FinTech ecosystem that I am currently involved with. Most Neo-banks are built on modern core banking systems like Mambu, Thought Machine, etc. Core banking systems are not the only third party systems you need to build a modern bank. You also need a CRM, lending management system, payment switches, AML/fraud prevention system, Engagement platform, CMS, KYC, and a few others. Once you have selected all the ecosystem partners you have to do custom software development to build new and innovative customer journeys and integrate these systems into a working neo-bank. In this post, I will talk about important factors you should consider when architecting systems that are powered by third party systems.
The product I am building/architecting at work these days uses Monorepo for all our Microservices. Our Microservices are primarily built using Java 17 and Spring Boot 2.6.x. For frontend and platform code(Terraform, Helm charts, configuration files, etc) we have different Git repositories. We use Gradle 7.3 as our build tool. We also make use of shared libraries for code reuse. I know people suggest you should avoid using shared libraries in Microservices but as I discussed in an earlier post I think there are valid reasons to use shared libraries in Microservices.
I prefer Monorepo for three main reasons:
Better visibility and control.
Atomic code refactoring across Microservices. This is common in the initial phase of development.
One of the advantages of Microservices architecture is that it enables components to have deployment independence. Based on my consulting and software development experience deployment independence is often overlooked and very few teams achieve it. Deployment independence is important since it brings true agility and reduces communication overhead between different teams and services.
Shared libraries make Microservices tightly coupled and introduce hard dependencies. Since, now a team making a change has to ensure that it does not break another service that depends on the shared library. This requires communication between multiple teams. Also, change in a shared library leads to all the services that depend on it to be redeployed. This leads to long build, release, and deployment times. We might have to consider the deployment order of services as well. All this leads to more synchronization and communication between teams. So, it is recommended that in Microservices architecture teams should avoid using shared libraries.
One of the common Web API design anti-patterns that I see in the field is the exposure of database model in the API contract. If you are building a Java Spring Boot JPA application then it means exposing JPA entities as Web API’s request and response objects. The primary reason this happens is because most teams are not following contract first model of API design. They start from code and database schema and then they create API contract from them.
This is not the first time I have seen this anti-pattern being applied by development teams. I have seen this often so I thought let me document it so that in future I can share this post. The advantages of document such lessons/patterns/practices are:
I can be thorough in my explanation. Writing helps me understand if my point is valid. Writing is thinking for me.
While explaining to a developer I might forget a key point.
Give the development team time to reflect upon the feedback by themselves.
Discussion after going over the post might be more productive.
I can keep updating this post.
Following are the reasons that I think we should avoid exposing database model as an API contract.
Once again, I stumbled upon the documenting architecture decisions post by Michael Nygard. This time, a particular line from his post got me thinking and made me dig a little deeper into the subject. The line said, “ ‘architecturally significant’ decisions [are] those that affect the structure, non-functional characteristics, dependencies, interfaces, or construction techniques.” The reason this piece of statement is important is because most of the time architects are not clear about what they should document.
The author suggests that we must document architecturally significant decisions. He divides them into five categories. In the post, he does not give examples against those categories, so, I am taking the liberty to add mine.
Below is my interpretation of these categories. It could be different from the author’s intent.