Best practices for modeling recursive parent relationships in GraphQL

[molostovvs]

Hi everyone!

I'm modeling a hierarchical structure (promotion groups) in GraphQL. The current schema looks like this:

type PromotionGroupLinked {
  id: String!
  name: String!
  parent: PromotionGroupLinked
  promotionCount: Int!
}

So the parent field points back to the same type. In practice, our only clients are our own frontend apps, and they normally query with depth = 1.

My main concern isn't malicious queries but efficiency:

• deep nesting could cause heavy memory usage during serialization
• responses might get unnecessarily large because the same parent chain can be repeated across many nodes

So, it's fine to keep the parent field recursive like this, or would it be more sensible to make parent a separate lightweight type (e.g. just id and name without its own parent)?

How you usually model these kinds of hierarchies in GraphQL and what trade-offs you consider when deciding between a recursive design and a simplified "flat" parent type?

[martinbonnin]

I think there were some explorations of a transport that would normalize the response so that repeated entities are sent only once but I can't find a reference sadly.

My 2 cents would be that, especially with a depth of 1, the default of sending the same repeated entity multiple times is probably good enough (gzip can probably help).

If you want to squeeze the maximum perfomance of your API, this is probably a good use case for a specialized binary protocol (see also #432). I'd love to see this happening some day.

[molostovvs]

Thank you for your reply and time!