I wrote a blogpost nearly 5 years ago about my programming stack of 2015 and I thought it would be funny to revisit and write an updated one.
The first I mention is IO.js which is node.js, but it was funny at that time node.js development had slowed down and the fork was a critical point. I still use Node.js a lot and I think it is still a solid choice for a programming stack in 2020. It is still evolving and with the new ES Modules that aligns with web browsers, code reuse is a great benefit. The ES Modules are still experimental, but I use ESM which is a great way to bridge until it is ready.
I was very dramatic in 2015 proudly pronouncing it as the year of the container. A lot has happened since and containers are now the defacto and not like in 2015 where it was just new and hot. After I wrote the article I even started a company that would do container hosting, but I learned that being early is both great but a pain in that nobody knew the technology at that time. In 2020 it is still new and many more advanced features have come since, so containers keep on rocking. (Not the company docker anymore, that is a bit of a mess!)
I was talking about multiple databases and MongoDB and graph databases. Looking back I have never had much success with MongoDB, partly because SQL databases have been improved greatly over the past years and have now incorporated many of the great features of NoSQL databases like JSON support and computed fields. I still like that I mentioned SQL and with different problems use different databases, but that is not so true anymore. I have yet to use a Graph database, I haven't had the issues where a SQL database couldn't solve it and where the overhead of adding another database was worth it. We haven't seen many hosted graph databases either from big cloud providers so that is also part of the reason.
I was talking about serverless without using the word serverless, which is quite funny compared to how popular the term has grown over the last 2-3 years.
AWS was a first-mover on this, but their product was simply too complicated and I ran several workshops and gave talks about it, and there were so few people that understood it and were willing to overcome the steep learning curve that it kind of weird.
A lot of people have also mocked the term and said they used serverless products in the past, but the problem was that those products did not allow you to scale up while still only paying a small fixed amount. That is where serverless shines.
So on to my programming stack for next year and probably for the next couple of years! I think it is important to not choose technologies based on hype but based on what potential you see it and how it aligns with your values. A good talk by Bryan Cantrill talks about the different values of programming languages and that it is impossible for a language to encompass all values, https://www.youtube.com/watch?v=2wZ1pCpJUIM
I first noticed Kubernetes while I and James were working on our startup and at that time we dismissed it as it was too complicated, but over time it has become one of the most user-friendly complicated software I know! There are now hosted version on all major cloud providers and kubernetes is now the platform to develop for! Kubernetes is the answer when you want to host something reliable in your company and you can align it with the values of the company, whether that is going fast or going slow and steady!
We are seeing people moving away from Kubernetes again, but that is because everything cloud providers are now doing (Google Cloud Run, AWS Fargate, etc.) is built on top of Kubernetes and one knows that they are not going down the proprietary way. I feel that Kubernetes also made the right abstraction and including extensibility in the Kubernetes API server. That made so that many frameworks were able to build on top of Kubernetes instead of next to, this was critical!
I don't use Kubernetes in smaller projects as Kubernetes is a big setup that both cost money to maintenance and keep up.
2019 was also when Google launched Cloud Run which is essentially Kubernetes underneath but you only pay for what you use. This product is amazing and the value you get is 100x. The product is flexible because it uses the container format and just HTTP, while providing loadbalacing and HTTPS on top. Your website will never go down because of a small compounding mistake, no Cloud Run simply spins it up again on another server, and in reality you don't have to care!
It is also not limiting in the sense of old serverless where the cloud provider was providing the runtime. That had a major downside that you couldn't decide new runtimes and was often stuck on very old versions of example Node.js and some runtimes did simply not exist! That is not a problem with Cloud Run.
Cloud Run also has the scale of Google which means that the product cost very little and it doesn't matter to Google. Opposite to other smaller hosting startups, Google can afford having a lot of customers that don't pay much, they are betting on that you will grow big and stay on the platform.
Being able to store data at low cost has been possible in 2019 and going into 2020. Firestore is a wonderful example of a great database that is hosted and has very few downsides. It is cheap, it can handle scale, it has a lot of features built-in and even a emulator for local development if you need that! Firestore is not at all like its predecessor Firebase although they share a lot of similarities! Firestore is a proper database where Firebase was its own product and focuses more on smaller developer teams. Firestore for example has proper Google Cloud integration with IAM and everything!
In my toolbelt, I still keep SQL which is still wonderful with its flexibility and its commonness. You can run it really cheaply in the Cloud. With $7 you can run it in Google Cloud, but the downsides for side-projects is that you pay even if you don't use it. Firestore is a winner here, having idle data stored does not cost much. But SQL means a lot of things, SQL can also be used with Google Cloud BigQuery which can handle Terabytes of data and can be queried with SQL language easily. Learning SQL and keeping it fresh is a well worth trade.
It is amazing how popular REST is and still something new can come along and take its place. GraphQL has for me been an amazing shift and I think it is because I am both a creator and consumer of data endpoints. If you only look from one of the sides REST looks simpler to both, but what can't be seen is all the extra overhead of logic to achieve the same result as one can with GraphQL. With GraphQL it is possible to write one HTTP request and get back data for what would have taken 10s of requests and a lot of code.
GraphQL is mocked by looking similar to a lot of other old RPC/SOAP frameworks, the difference is in all the tooling that has been made and the learnings of those frameworks and also with the grown interest of types. All the necessary logic has been moved back to the backend and the front-end developers are making features faster than ever with GraphQL. I have yet to see many developers use GraphQL fragments, but that is something I hope to see in 2020, I am taking advantage of it.
I still really like the ideas of containers because of the predictability you get by having something being reproduced over and over again in the same environment. It helps save time when everyone speaks the same language. The container tools have not improved as much on Mac and Windows as you would like and everything is far from running in containers yet, but on the Cloud side it is well established. Everything is still based around the Docker standard but open source standards are slowly popping up like Containerd and podman for running them. The best developer experience is still Docker and I still like how easy it is, 1-click install and that has taken a lot of work for them to get there.
I don't use docker-compose as much anymore and I think that is because I have become better at using CI/CD and writing tests. I used to use docker-compose spinning up a whole environment to allow an end-to-end test. That has changed with the projects that I have worked on lately not needed that and the cost of maintaining it (yes it breaks) has not been worth it.
I am excited about the continuing progress of containers and so is containers still a big part of my programming stack of 2020.
This tool is the newest one in my stack. I have had problems for a long time with writing bigger apps and keeping build time low, while not making it over complicated to others inside a company. That is what bazel can help you with, keeping your development cycles fast and ensuring high collaborations across the company. It does that by intelligent caching output using a global cache for all developers and allowing developers to define relationships between projects that is being built. Bazel then know when a library has changed and what services have to be tested again! Bazel is a tool that Google has internally built and it recently reached version 1. It was not a dramatic change but a signal that it is ready for real use. It feels a bit like Docker in the early days with rough edges and documentation that are hard to figure out, but a tool like Bazel is sure to become the standard in development. As software is eating the world, making reliable software becomes more important and more difficult to do, believing that we can't automate us of that problem would be a mistake.
It is funny looking back as some of the things were not obvious at the time. These posts also helps saving the thoughts of a point in time in my programming which is interesting to look back at. I for one could not have predicted how big the technologies have become now, but I did see some clear signals back then. Docker in 2015 felt awesome and I was really into it, but for many people it was still something that had to be reflected on and considered. That is understandable, changing technology is costly, but sometimes the value is far greater than what it cost to adopt.
One thing I didn't write in the article at that time was Docker Swarm, which was like Kubernetes but much simpler. Back then I was sure Docker Swarm would win as it was the built-in cluster tool and it was much easier, but I turned out to be wrong. If Docker Swarm was the default today, Docker the company had probably been a big success, but when the other big companies saw that early on, most of them bet on Kubernetes, as it had a more democratic approached and so it won. Being wrong is not fun and a risk.
Choosing popular technologies is not a goal, also called hype-driven development, but knowing when a technology will give you a clear advantage is a definite plus in skills! I would still use these tools even if nobody where using them next year, but using something that is popular will get you a lot of value. When more people are using it all the small and annoying problems around it will be fixed.
GraphQL is still new, Bazel is still new, Cloud Run is still new, Firestore is still new, but it is clear that the modern programmer has a pretty good life in 2020 in terms of productivity. Everything is cheaper, easier than ever and only good ideas and executing is the problems left!
|Previous post:||Making the best community software|
|Newer post:||How to keep secrets secret in the cloud|