> We investigated using Prisma v2 as a way of auto-creating a GraphQL API that directly interfaces with a PostgreSQL database, but we immediately pivoted to other solutions as soon as we discovered that the Prisma Client is really spinning up a behind-the-scenes GraphQL rust server itself to access the db. The performance of fetching a mildly complicated query (3 joins) ended up being more than three times slower than Hasura or PostGraphile.

Do you know what Hasura is doing differently? I assumed it worked the same way as Prisma does, with an intermediate server.

Hasura is a gql server too. I can’t say how it’s different from prisma, but it works by running as a frontend process to your Postgres dB. You issue gql, it converts to sql and then back to gql on the way out.

Some of the things it does differently though. They declare access permissions in the dB (Hasura tables, managed through their api) and they work those into the queries. They do this by building a json object with any of your variables in and then reference those within the query. This allows for a neat trick where they can run the same query for different clients by starting with multiple rows (one per client) containing their respective variables. This is particularly great for subscriptions. They can poll the dB, to grab all data for all connected clients in a single hit, and then if anything has changed, they can notify the external clients about it.

Hasura is a GraphQL layer for the database. It introspects a PostgreSQL database and creates a GraphQL API which exposes access/operations to the database.

Prisma is a toolkit that consists of a database client (Prisma Client), a migrations tool (Prisma Migrate), and a database IDE/UI (Prisma Studio). The generated Prisma Client is in TypeScript and can be imported into a Node.js application.

While Prisma can be used to build a GraphQL API that connects to a database, Prisma is completely agnostic to the GraphQL tools you use. (https://www.prisma.io/docs/understand-prisma/prisma-in-your-...)

I'm not sure about PostGraphile, but Hasura I know transforms a graphql request into a SQL request that can return the JSON response expected as part of the GraphQL spec. I believe that's the reason that Hasura only works with PostgreSQL, since it has the necessary SQL functions to generate arbitrary JSON objects.

> The performance of fetching a mildly complicated query (3 joins) ended up being more than three times slower than Hasura or PostGraphile.

Please forgive my ignorance, but these type of conversations always intrigue me and make me want to learn more. As a software engineer, at what point are you running these performance benchmarks, and how are you doing them exactly? How have you had time to make these decisions and pivot to other technologies?