Prisma

you will have to do this yourself via a custom resolver

I see. So, I would need to complete a secondary request to retrieve that information? One of the reasons for switching to GraphQL is the idea of only making one request versus a Rest API with several. Please school me if I am wrong.

I'm not sure myself, i don't think doing a custom resolver adds another request. I'd love to know more about the overhead of htat myself but it's been hard to get an answer.

maybe <@U0RQY0KK5> can chime in at some point about this

I have an issue open as im having some issues with the performance of custom resolvers
<https://github.com/graphcool/graphql-server-example/issues/85>

I can see the various `Connection` schemas that are added, have read through the docs here: <https://www.prismagraphql.com/docs/reference/prisma-api/queries-ahwee4zaey>, and can see that they are interested in adding this stuff here: <https://github.com/graphcool/graphcool-framework/issues/416>, but there are no current work-arounds that I have seen that can be accomplished without a custom *Query*.

yeah that's my finding too, which is why i am suggestion a custom resolver

for example, i have an app that needs avgRatings

```
 avgRating: {
    fragment: `fragment AvgRating on Place { id }`,
    resolve: async ({ id }, args, ctx: Context, info) =&gt; {
      const place = await ctx.db.query.place(
        { where: { id } },
        `{ reviews { stars } }`,
      )
      const reviews = place.reviews
      const avgRating = reviews.reduce(
        (
          prevValue: number,
          currValue: Review,
          currIndex: number,
          reviews: Review[],
        ): number =&gt; {
          return prevValue + currValue.stars / reviews.length
        },
        0,
      )
      return avgRating
    },
},
```

but i still don't understand the overhead of a custom resolver

and how it impacts the query's performance

I can see in your issue that the spike was pretty dramatic.

yeah.. it's worrying me because i don't understand the casue

im hoping it's just because Prisma db is hosted on a shared cluster

When you were building a list of *places*, how did you call that custom resolver? I must not understand its implementation.

<@U8MKD6D43> I did nothing special that is not done in that example repo.