Support for Read Replicas (Postgres, MySQL) or setting Client in read-only mode

@matthewmueller’s update

Problem

Currently we don’t support read-replicas. This is a common need for scaling databases.

Solution

I agree with @dpetrick’s solution below.

One thing I would additionally suggest is to provide a Readonly interface that you can pass through. This Readonly interface would omit the Create, Update, Delete methods.

---

Does prisma2 support the ability to direct queries to master db or read-only replicas based on read/write? This is useful functionality that lots of db drivers support. Hopefully since prisma2 aims to replace the db driver, it would also have this functionality.

(User feedback)

This can be handled in various ways (the list if from the top of my mind):

1. Not natively in Prisma, have actual read/read-write nodes in the DB natively and multiple datasource blocks and multiple generate blocks to generate read-only, read-write photon.

datasource ReadOnlyDB {
  provider = "postgresql"
  url      = "<read only DB credentials>"
}

datasource ReadWriteDB {
  provider = "postgresql"
  url      = "<read-write DB credentials>"
}

generator ReadOnlyPhoton {
  provider = "photonjs"
  datasource = "ReadOnlyDB" // API is not final
}
...

1. Natively in Photon with a read only parameter

generator ReadOnlyPhoton {
  provider = "photonjs"
  readOnly = true // API is not final
}

Other ways are also possible.

Additional context

• https://github.com/prisma/prisma/issues/3389

[dpetrick]

My first intuition on the subject was neither of your options, but the possibility to provide a read-only endpoint directly in the datasource, which would cause Prisma to internally either load balance or direct all read traffic to the read db. Both your options are severely limited in use:

• Option 1) requires you to annotate models with a source… and then you suddenly need 2 annotations for read and write.

• Option 2) would imply that the whole photon client is just throwing out write queries and just allow read queries. Seems odd to me all around, because what’s the point of having this when the approach is pretty much a glorified version of creating 2 photon clients and separating traffic on the client side.

For me, this is the most simple and intuitive version:

datasource PSQL {
  provider = "postgresql"
  url      = "<regular conn-string>"
  read-url = "<read-only conn-string>" // Or whatever we feel like naming it
}

… and Prisma handles the rest.

[bmxpiku]

It would also be very convenient when using AWS Aurora ❤️