Batch loading support about graphql-erlang HOT 2 CLOSED

We've wanted to have dataloader-like capabilities for some time, but it is dependent on having full concurrency support in the code base (which currently needs some timeout handling code). Indeed, once that is in, the plan is to have a hook-like functionality where you can register a small function to be run (usually you will use it to send a message).

The concurrency-work is not documented yet, mostly because it is highly experimental and still subject to change. But indeed, the defer_loop function is the place where you want some kind of hook in order to tick and tell a data loader process to execute its batch of work.

from graphql-erlang.

We may need this feature soon because certain nested queries with shared nodes are performing terribly.

The batch start hook should be in Ctx just like error_mod. For extensibility, I think it should be a module just like error. The key can be something like concurrency_mod or batch_mod.
hook:begin_loop(Ctx) should kick the data loader process and defer_hook:end_loop(Ctx) can do some clean up if necessary.

For actual implementation, I'm not sure what's the right thing to do. Here's what I have in mind.
Wrap this call: https://github.com/shopgun/graphql-erlang/blob/8e75a73892a97f43e142fabf04599acebea8eb95/src/graphql_execute.erl#L105

with Hook:begin_loop(Ctx) and Hook:end_loop(Ctx). Maybe a with a try finally to ensure end_loop is always called.
I'm not sure how error in begin_loop and end_loop should be handled, probably just like resolver_crash.

Any inputs on this?

from graphql-erlang.