(Question) Is it possible to set LocalContext for child resolvers?

[abishai]

Hello, I'm playing with KGraphQL library to see if I can port my pure GraphQL controller to it, however I've found no way how to set LocalContext. I use it extensively to provide common state from root resolver to child resolvers. I've found Context class with map, however it has no set method.

I'm speaking about this feature of GraphQL: https://github.com/graphql-java/graphql-java/blob/master/src/main/java/graphql/execution/DataFetcherResult.java#L81

So, I can write

        val result = DataFetcherResult.newResult<>()
        val data = ......
        result.localContext(......)
        return result.data(data).build()

Can I do the same thing in KGraphQL ?

[abishai]

I've written the following extension functions, I am not happy with solution, but it works. Maybe you can provide some official way to do similar thing? I can't make resolvers self-sufficient everywhere as I need a shared state for some of them.

    @Suppress("UNCHECKED_CAST")
    fun Context.set(key: Any, value: Any) {
        val field = Context::class.java.getDeclaredField("map")
        val map = field.also { it.isAccessible = true }.get(this) as MutableMap<Any, Any>
        map[key] = value
    }

    @Suppress("UNCHECKED_CAST")
    fun <T> Context.get(key: Any): T {
        val field = Context::class.java.getDeclaredField("map")
        val map = field.also { it.isAccessible = true }.get(this) as MutableMap<Any, Any>
        return requireNotNull(map[key]) as T
    }

[jeggy]

Currently this is not something that is supported. A feature request for it is mentioned here #129 and will most likely be included in some future versions of KGraphQL.

How I'm solving this is I have a all data needed directly on the type returned from the resolver and if some of the fields should not be available within GraphQL I would use the ignore functionality.

[abishai]

I see, but this way is not optimal for some cases. Well, I'll have to use my hack until proper solution is implemented. Can you just provide setters for Context map? Seems easy and trivial. I can make a PR if you wish :)

[jeggy]

PRs are always welcome 👍

I would guess you want this because you want to pass data from the grandparent nodes or something like this to your resolver and this pattern is not recommended when designing a GraphQL schema. Instead every resolver should always live on it's own.
But that being said, I think we would like to offer developers to work how ever they want, so if you could provide a PR for #129, then that would be great.

[jeggy]

And just to be sure you are not reporting this in the wrong repository. All code and links you provided are for the graphql-java implementation of GraphQL.

KGraphQL and graphql-kotlin/graphql-java are not related.

[abishai]

Well, I'm writing options trading software (for personal use) and I have a user case where I compute my position stats, that in case of options can be rather complex. Every option has underlying asset, that is the same for several options in position; underlying asset has it's own market data, so if I want coherent calculus, I must cache (at least) market data of underlying asset. As math can be complex (fetches data from different microservices, uses iterative algorithms, ....) I'd like to avoid computation of entire object every time and provide only requested data. That's the cardinal idea of graphql, to provide only that is asked. :p So, probably, I'll stick with not recommended way and make a PR..

[jeggy]

Adding support for adding custom data to the GraphQL Context will only help you a little bit as the Context only lives per request.
Why not have some cache living on the side like caffeine.

How I normally solve this issue is to have cache saved using Caffeine alongside a RabbitMQ cluster to invalidate the cache when needed. This way I always have up to date cache ready for use across all GraphQL requests.