Hi everyone,
In a project involving Firebase and object types like Tickets, Schedules, and Timers, I want to structure my classes such that switching databases (potentially to MySQL) wouldn’t require a complete rewrite.
Approach 1:
- A DatabaseProxy interface with generic methods (e.g., createTicket, createTimer, etc.)
- A FirebaseProxy class implementing the interface, with methods for each object type (e.g., createTicket, createTimer, etc.)
- Manager classes for Tickets, Schedules, and Timers, that primarily use the FirebaseProxy for operations. This provides flexibility for processing input/output, but most of the time the manager classes will just be calling methods on the Proxy directly.
Approach 2:
- A DatabaseProxy interface with the most basic CRUD methods (create, read, update, delete).
- A FirebaseProxy class implementing the interface.
- Manager classes for Tickets, Schedules, and Timers, calling FirebaseProxy with parameters like update(collection, ticket) and implementing createTimer, createTicket, etc.
I like the second approach in theory, but what I’m worried about is whether the separation is too low level. What happens if the database I switch to changes schema such that taking in an object and a collection name isn’t good enough anymore? For example, will there be concerns if I switch between Vector, NoSQL, and SQL?
Any opinions are appreciated!
Definitely want Model and DAO classes with any DB specifics abstracted away at their interface. Business logic goes into the service layer above. It’s not uncommon to have a service layer level of crud that includes logic to create other required entities or perform complex validations. Service and API layers should always be totally agnostic to data layer. A generic “database proxy” like you described should be solved by your ORM and would live below your DAOs.
Sounds like the repository pattern would help here.
I’m doing something similar now where I need to store objects “somewhere”. I have a low level Repository interface to handle persistence that can do the basic CRUD (mainly get/set for my use case). It’s primarily backed by redis, but that same interface has been backed by Postgres, vault, and in-memory caches depending on the need/environment. Works amazingly well.
As a bonus we can create a new Repository to migrate data when needed - such as a redis or postgres upgrade, we build a MigratingRedisRepository that takes in 2 RedisRepository and does the necessary logic of reading from the old and writing to the new.
I think you’re on the right track with a mix of 1&2. Abstract out the data store, it will change some time - and you’ll want to control it for tests too. Let services/managers handle state and delegate down for persistence to wherever that may be.
