Example of DataLoaders.

[neetkee]

📝 Description

I've asked about using dataLoaders in the related issue, so here is my example of its possible implementation, using an extension of a query handler. What do you think?

Note that I also made ExecutionResult.toGraphQLResponse() function public in this PR.

🔗 Related Issues

#381

[codecov-io]

Codecov Report

Merging #396 into master will not change coverage.
The diff coverage is n/a.

@@            Coverage Diff            @@
##             master     #396   +/-   ##
=========================================
  Coverage     97.62%   97.62%           
  Complexity      324      324           
=========================================
  Files           104      104           
  Lines          1223     1223           
  Branches        198      198           
=========================================
  Hits           1194     1194           
  Misses            9        9           
  Partials         20       20

Impacted Files	Coverage Δ	Complexity Δ
...pediagroup/graphql/spring/model/GraphQLResponse.kt	100% <ø> (ø)	4 <0> (ø)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4e38e11...4592bc6. Read the comment docs.

[dariuszkuc]

Thanks for providing the example! I've added default support to SimpleQueryHandler in #400.

Would you mind updating the example to use latest code?

[neetkee]

I guess we need to create DataLoaderRegistry per request, to avoid passing cached values between requests? I'm not sure how to do it with #400. Maybe we need to autowire some factory, that creates dataLoaderRegistry, I saw a similar way in SPQR lib: DataLoaderRegistryFactory

[dariuszkuc]

Interesting! So data loader registry should be created per request? I've added support for this in #410 by creating factory that allows developers to create new data loader registry instance per query execution. Unsure how useful it is as within graphql-kotlin we generate schemas directly from the code based on the underlying functions and classes (where devs can implement whatever) but I think it is good to have an option to use DataLoader.

[smyrick]

👍 Plus one to this:

Unsure how useful it is as within graphql-kotlin we generate schemas directly from the code based on the underlying functions and classes

In graphql-java land, I can see how this is a helpful caching mechanism since you don't have a class that is run when you execute a resolver/data fetcher. But with our library you will be executing a function that can be in any scope and fetching data from a downstream service can be cached in any way you want. This feature will be something we can support through the factory but we will defer handling of everything else to the server devs.

[neetkee]

@smyrick

But with our library you will be executing a function that can be in any scope and fetching data from a downstream service can be cached in any way you want.

Could you give me some direction on how it can be implemented?

[smyrick]

@neetkee

If you have this class and function as your schema object

class Query {
  fun getUserById(id: Int): User? {
    val result = getUserFromSource(id)
    return result
  }
}

the getUserFromSource could be any function that loads data from an API, database, or a cache system. This means that you could handle the caching layer yourself based on more inputs or configs and not solely based on the DataLoader implementation.

[neetkee]

If I understand correctly, we can use it to load from cache/service/etc by id, one by one, but if we want to batch our request, we need to gather all needed ids, and then execute our query.

[dariuszkuc]

If I understand correctly, we can use it to load from cache/service/etc by id, one by one, but if we want to batch our request, we need to gather all needed ids, and then execute our query.

Yep. It all drills down to how you gather those IDs -> you need to know all of them to batch regardless whether you use data loader or not.

[RIP21]

@smyrick hey. Am I understand that correctly that this thing will be called anytime somebody queries the company field company
lateinit var Company on Employee and ONLY the Employee, and not other classes?
Or I have to branch here or whatever?

Maybe some drawing or ASCII or draw.io graph will help me understand the connection between beans and the whole resolve cycle as I'm unsure :)

[smyrick]

@RIP21 The DataFetcher is the name of the "resolver" for graphql-java. You can see the docs here: https://www.graphql-java.com/documentation/v16/data-fetching/

This is a data fetcher for the entire Company class. GraphQL Kotlin normally would handle creating the different data fetchers for you, but in this case the user wanted have a DataLoader which is a separate GraphQL specific caching mechanism. The DataFetcher is invoked when ever there is a field that returns a Company and, using DataLoaders, they are cache the company info for each employee so they don't have to make multiple calls to get the the duplicate company data for all the employees