Migrate from microservices to Elixir/Phoenix: is it possible?

Hello everyone,
I beginned recently to study and work in elixir, and I’m very curious about OTP, scalability and performances.
I’m writing this message to ask you if what I’m thinking makes sense or I am missing something.

By now I’m maintaining a platform made up of 6 microservices, all in Java (Quarkus/Spring), which intercommunicate with Kafka (both for data and events).
Every microservice has its own database; the platform asks data to an external APIs (through 3 “connectors” microservices) and write them into Kafka. Then another microservice reads this data and stores them into a database (with some logic), and exposes them through different queries.
Finally there is another service which acts as authentication server and gateway (with authorization).
All run in a Kubernetes cluster.

What I am thinking is: can I replace all this with a Phoenix application, where every microservice could be a separated app (maybe all under the same umbrella)?

These are possible pros:

  • One database for all (is this an anti-pattern?)
  • Get rid of Kafka (and Avro)
  • Simpler maintenance
  • Easy support to GraphQL (if I choose to make the switch, now it’s all rest)
  • Elixir

Possible cons:

  • Losing of scalability opportunities (with Kafka I can scale services as many as I need: how can I replace this with Phoenix? Do I need this?)
  • Minor decoupling of the system (I cannot “plug” an eventual microservice developed in another language)

Making the question in another way, Elixir applications and OTP could replace a platform made up of some microservices (all developed in the same programming language) which communicate with a message bus?

Thank you for any eventual response :slight_smile:

1 Like

Yes, this is possible. And maybe even more importantly, it might be possible to perform this change incrementally (switching over microservices one by one).

Depending on your needs you might be able to either augment or fully replace Kafka with Broadway.

It is very straightforward to create multiple releases from the same umbrella project, containing a different set of OTP-applications, and connect these together in a ‘distributed Erlang’ cluster. This will allow you to scale different parts of your platform independently (e.g. scaling the amount of webserver nodes independently from the amount of data ingestion nodes etc.), while not having to think hard about how these microservices need to communicate, as transparent everyday message-passing can be used. (Usually e.g. Phoenix.PubSub is used in these kinds of situations which allows for even further decoupling between services.)

Obviously this last part is only possible once everything has been moved over to Elixir. However, when rewriting incrementally you will still be able to remove more and more ‘manual communication’ code as time goes on.