I own an Acura TL 2006. It’s a great car. Every day I derive great pleasure driving it to work. It has a tight sporty suspension, precise steering, comfortable leather seats and an awesome audio system.
At the same time, I know better than to take it off-road. Its high performance tires are optimized for asphalt traction and low rolling resistance, not gravel or soil. It does not have enough clearance for rocks, or 4×4 drive required for rough terrain. If I did take it off-road, I could erroneously conclude that it is an awful car, which I know not to be true. I would have simply used it for something it was never designed to do.
I used this example to explain the concern I have seeing the evolution of the industry’s relationship with the micro-service architecture. It was just a matter of time people until people started taking their micro-service Acuras off-road and then writing how they are awful cars.
Original success stories
Architectures and approaches normally turn into trends because enough use cases exist to corroborate their genuine usefulness when solving a particular problem or a class of problems. Otherwise, only architecture astronauts would care. In the case of micro-services before they were trendy, enough companies built monoliths beyond their manageability. They had a real problem on their hands – a large application that fundamentally clashed with the modern ways of scaling, managing and evolving large systems in the cloud. Through some trial and error, they reinvented their properties as a loose collections of micro-services with independent scalability, life cycle and data concerns. Netfix, Groupon, Paypal, SoundCloud are just a small sample of companies running micro-services in production with success.
It is important to remember this because the trendiness of micro-services threatens to compel developers to try them out in contexts where they are not meant to be used, resulting in the projects overturned in the mud. This is bad news for all of us who derive genuine benefits from such an architecture.
Things to avoid
It is therefore good to try to arrive at a useful list of use cases where micro-services are not a good choice. It will keep us more honest, keep the micro-service hype at bay and prevent some failures that would sour people to an otherwise sound technical approach:
- Don’t start with micro-services – this one is a no-brainer. Micro-services attempt to solve problems of scale. When you start, your app is tiny. Even if it is not, it is just you or maybe you and couple more developers. You know it intimately and can rewrite it over a weekend. The app is small enough that you can easily reason about it. There is a reason why we use the word ‘monolith’ – it implies a rock big enough that it can kill you if it falls on you. When you start, your app is more like a pebble. It takes certain amount of time and effort by a growing number of developers to even approach monolith (and therefore micro-service) territory.
- Don’t even think about micro-services without DevOps – micro-services cause an explosion of moving parts. It is insane to attempt it without serious deployment and monitoring automation. You should be able to push a button and get your app deployed. In fact, you should not even do anything – committing code should get your app deployed through the commit hooks that trigger the delivery pipelines (at least in development – you still need some manual checks and balances for deploying into production).
- Try not to manage your own infrastructure – micro-services often introduce multiple databases, message brokers, data caches and similar services that all need to be maintained, clustered and kept in top shape. It really helps if your first attempt at micro-services is free from such concerns. A PaaS such as Cloud Foundry or Heroku will allow you to be functional faster and with less headache than with an IaaS, providing that your micro-services are PaaS-friendly.
- Don’t create too many micro-services – each new micro-service adds overhead. Cumulative overhead may outstrip the benefits of the architecture if you go crazy. It is better to err on the side of larger services and only split when they end up containing parts with conflicting demands for scaling, life cycle and/or data. Making them too small will simply transfer complexity away from the micro-services and into the service integration task.
- Don’t share micro-services between systems – I listed this final point here for completeness, but it is so important that it requires to be broken into its own section.
On micro-service sharing
I have seen many a fiery debate about the difference between micro-services and SOA. There are many similarities (it is hard to argue that micro-service architecture, or MSA is revisiting SOA principles). More recently I have formed a fairly strong opinion that a key differentiation between MSA and SOA is that of ambition.
When you go back and read about the lofty goals of SOA proponents, it is easy to notice that the aim was much higher. MSA success stories didn’t attempt to reinvent the world around catalogs of reusable services, systems that are discovering those services through registries, etc. At the beginning of every MSA success story is a team that grew their simple application too fast without refactoring along the way and hit the maintainability wall.
If you carefully read ‘monolith to micro-services’ blog posts, you will notice that the end result is the same thing. Groupon team has not created a ‘catalog of social coupon services to be assembled into coupon applications’ – they rebuilt Groupon Web site. They broke the monolith into small pieces and rebuilt it again. As far as their end users are concerned, the monolith is still there – the site was rebuilt in mid-air.
Since I think that micro-services are pragmatic and sane revisiting of SOA, it is apt to assume that creating reusable micro-services is low on the list of priorities. Yes, a micro-service needs to be individually deployable and be flexible enough that it can be bound to other services dynamically (minimally through some kind of a configuration on startup). You need to be able to deploy each service to multiple logical ‘spaces’ (DEV, QA, STAGING, PROD). But each logical micro-service instance is part of a single distributed monolith, re-imagined in a cloud-friendly way.
From a monolith to a – distributed monolith?
Where am I going with all this? I am a bit concerned that the industry noise will ruin micro-services by taking them outside their comfort zone. Too many people are taking them to the areas where they shouldn’t, and I don’t want the inevitable backlash to overshoot. Micro-services are a solution for the Big Ball of Mud architecture, but the alternative micro-service system is still a big ball. This ball made up of many small balls, is cleaner and easier to manage, deploy, scale and evolve, and can be inflated bigger than the old ball without exploding, but it is fundamentally the same thing.
Any attempts at nano-services, trying to deploy micro-services manually, using them because they are trendy without real need, or re-using them between multiple systems will result in a disappointment we don’t really need at the moment.
Are micro-services SOA? No, and please let’s keep it that way.
© Dejan Glozic, 2015
One of the more interesting ideas I’ve come across in similar articles relates to Conway’s Law :
Simply stated: “organizations which design systems … are constrained to produce designs which are copies of the communication structures of these organizations”
It makes sense to me that for microservices to really pay off you should be an org where developer autonomy and asynchrony can work.
Maybe we can give it a clickbaitey title, like “Microservices and micromanagement are mutually exclusive”
Oh, awesome, I like it very much. But you came up with it so I will steal it in talks but you have the dibs on blogging with that title :-).
I think you will do the topic more justice, so feel free to use the title.. as long as you give me some credit for it =P
It seems like an interesting idea for a companion article for your dysfunctional teams post.
Thank you for ssharing this