Seeding the DB

[techsavvyash]

What are the problems/open questions at hand?

• Prisma does not allow for a dynamic selection of tables based on db schema.
• Can we be sure that the schema and models are populated before the user starts to seed?
• Do we allow for dynamic definition of schema.prisma based on the JSON/CSV file?

Approaches we can follow

1. cQube Style

We can follow a cQube style approach with grammar and data files and use those to ingest the data into the db. (Seems like an overkill to me)

2. Code Generations for seed.ts

We can assume that the tables have been defined in the schema.prisma, migrations applied and the prisma client generated. With these assumptions we can have a template file for seed.ts which we can then generate based on some parameters in the JSON/CSV file. This process can be wired up to some command like stencil generate seed-script --file='/path/to/json/file' --type=JSON/CSV

Things to take care of while following this approach would be:

• Make sure the names of the tables in the JSON and models in the schema.prisma are same, since we will be doing plain code generation which will cause errors incase table names are wrong. -- A solution to this is to parse the JSON file against the schema.prisma file.
• In case of CSV files, each table will have a single CSV, then we can enforce that the file name is the same as the model name and also the column headers are same as the field names in schema.prisma
• In case the CSV files have other names and we want to define mappings between the CSV header values and the model field-names we can have a config file which allows for more customisations.
• Reference seed script can be found here

3. Let the user write the seed script

Prisma has defined support for seeding based on writing seed scripts after defining the schema.prisma file, documentation around which can be found here. We can ask the devs to use this directly to seed the db.

[ChakshuGautam]

Sharing notes inline.

• Can we be sure that the schema and models are populated before the user starts to seed?

Yes.

• Do we allow for dynamic definition of schema.prisma based on the JSON/CSV file?

No

With these assumptions we can have a template file for seed.ts which we can then generate based on some parameters in the JSON/CSV file.

Let's nudge users to generate the JSON based on the prisma.<model>.create either with or without relation based on the schema that they created. You can verify the type of the object to make sure it is correct but call prisma.<model>.create to seed.

[ChakshuGautam]

Let's document this and add this to the cli to ensure folks have a cli command to ingest easily and no one is creating custom things.