gramps-graphql / rest-helpers

REST helper classes for GrAMPS GraphQL data sources.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

test: Demonstrate issue with dataloader

jlengstorf opened this issue · comments

@timrs2998 commented on Fri Nov 10 2017

@jlengstorf Attached is example code to demonstrate gramps-graphql/gramps-express#47

Problem

The purpose of the dataloader is to enable per-request caching. If I fetch an item by id "1234" twice in the same request, there should only be one request to the backend. However, fetching an item by id twice results in multiple backend calls.

Example

Start the backend server in one terminal:

$ node server.js
Example app listening on port 3000!

Start the GraphQL server using live data:

$ yarn live-data
..
============================================================
    GrAMPS is running in live mode on port 8080

    GraphiQL: http://localhost:8080/graphiql
============================================================

Submit the following GraphQL request:

query {
  hello:YourDataSource(id:"1234") {
    lucky_numbers
  }
  world:YourDataSource(id:"1234") {
    lucky_numbers
  }
}

And since the dataloader should cache the "1234" entity, I expect only one backend call and I expect the lucky_numbers arrays to be the same, even though lucky_numbers is an array of random numbers.

However, in server.js's logs, I can see two requests:

Request number #0
Request number #1

and in the JSON response:

{
  "data": {
    "hello": {
      "lucky_numbers": [
        7844,
        9953,
        1779,
        7452,
        5061,
        8062,
        7354
      ]
    },
    "world": {
      "lucky_numbers": [
        6489,
        6526,
        1814,
        5528,
        8614,
        4666,
        4819
      ]
    }
  }
}

Solution

I believe the problem is that a new dataloader is constructed every time connector.get() is called, rather than once for every GraphQL request. I'm thinking connector's lifecycle needs to change to accommodate this.


@timrs2998 commented on Fri Nov 10 2017

⚠️ don't merge, of course; feel free to close

Thanks for moving this. The default connector probably shouldn't include the dataloader with this behavior. With the new api changes, it's now possible to create dataloaders per request:

export default {
  ..,
  context: req => new Model(new Connector()), // <-- problem solved
}

Maybe it just needs to be documented now.

Have you tested that working?

yes, it's working for us

@jlengstorf

For a different but similar purpose (to solve the N+1 query proliferation problem), I've used this and it works well: https://github.com/calebmer/graphql-resolve-batch

It doesn't address the use case described here but it would be interesting to see if DataLoader can deal with the same cases as graphql-resolve-batch