In fact there isn’t such thing as cross-service data sharing at all. Lately I started a series of articles about Service ViewModel Composition. I talked about:
- What problem is services ViewModel Composition designed to solve in What is Services ViewModel Composition, again?
- How Single Item Composition is designed in The Services ViewModel Composition maze
- How ViewModel Lists Composition is designed in Into the darkness of ViewModels Lists Composition
- Introduced some coding in ViewModel Composition: show me the code!
Share all the things!
At a first look it might sound reasonable to use the ViewModel Composition approach to allow services to talk to each other. If
Service X needs some information that are stored in
Service B and
Service H why not allow
X to go to some sort of internal
Composition Engine where request handlers from
H could be deployed to solve the exact problem
X has…? Thank you, but no, thank you.
As my Dutch colleague Dennis van der Stelt would say “Autonomous microservices don’t share data. Period.”. That by the way is also the title of a talk he’s going to give this month at NDC Copenhagen, check it out! (recording is available on YouTube: https://www.youtube.com/watch?v=0TYbHVc2yWI)
Even if it’s technically possible to build such a thing, if we don’t want to end up with a “wonderful” distributed monolith we probably want to avoid it.
Poof, autonomy is gone
Regardless of the method we use to share data across service boundaries we’ll end up with services that are not autonomous. Autonomous services can evolve independently from each other, non-autonomous ones cannot. They are coupled, they won’t be able to evolve without breaking each other, causing evolution to stop or causing friction at development time, deployment time, and run time. But there’s more…
The snowball effect
As soon as we have services querying each other for information we’re immediately introducing temporal coupling. Whatever problem the queried service has, it’s propagated to the querying service. Is it down? The querying service cannot work. Is it slow? The querying service might timeout or fail. And so on, you probably get the point.
As soon as we allow cross-service data sharing, for example by allowing services to request data from another service, we end up with the worst of two worlds: a monolith suddenly turns into a distributed monolith with all distributed system problems on top of a monolithic one. The system immediately loses two key aspects:
- Services are not autonomous anymore
- Services are temporally coupled
Despite being technically possible, just don’t do it. If you feel the need go back to the drawing board, it’s very likely that services boundaries are wrong. Be careful, what at the beginning seems a small harmless snowball can easily turn into an avalanche that halts the entire system.