Data Migrations

It's great that Prisma allows us to easily represent our database models and their relationships in a schema.prisma file. And we can easily make our database just look like our schema by running npx prisma db push. But what if we make a "breaking schema change" that requires us to migrate our data?
Let's talk about what I mean by "breaking schema change". Let's say we have a sandwich ordering shop, like in the example from the first exercise. Here's the order model:
model Order {
  id         String      @id @default(uuid())
  createdAt  DateTime    @default(now())
  updatedAt  DateTime    @updatedAt
  status     String
  userId     String
  user       User        @relation(fields: [userId], references: [id])
  orderItems OrderItem[]
}
What if we decided to add a Reviews model and have reviews on the order? That's a pretty straightforward change using the knowledge we've accumulated so far. Simply create the Reviews model and add it to the order model:
model Order {
  id         String      @id @default(uuid())
  createdAt  DateTime    @default(now())
  updatedAt  DateTime    @updatedAt
  status     String
  userId     String
  user       User        @relation(fields: [userId], references: [id])
  orderItems OrderItem[]
  reviews    Review[]
}

model Review {
  id        String   @id @default(uuid())
  createdAt DateTime @default(now())
  updatedAt DateTime @updatedAt
  rating    Int
  text      String?
  orderId   String
  order     Order    @relation(fields: [orderId], references: [id])
}
We can npx prisma db push this change with no issues.
But let's look at the User model:
model User {
  id        String   @id @default(uuid())
  email     String   @unique
  username  String
  name      String?
  createdAt DateTime @default(now())
  updatedAt DateTime @updatedAt
  orders    Order[]
}
What if we decided we wanted to make the name required? Or what if we decided we want to make the username unique? These are "breaking schema changes" which necessitate a data migration.
Managing these data migrations is tricky business and often involves custom SQL or an intrusive ORM that may not quite fit with what you need done so you end up having to workaround the limitations.
Prisma takes the middle ground approach. Instead of completely hiding the details of the migration, Prisma generates the SQL for the migration and saves it to disk. From there you are free to alter it as needed.
This SQL file should be committed to your git repository and deployed alongside your app. That way you can track database changes over time. Prisma will also keep track of which migrations have been applied to your database so you don't have to worry about that. (It creates a very small table in your database for this). So when you deploy your app, you simply run npx prisma migrate deploy and Prisma will take care of the rest.

Example

I've found this to be helpful in my own work. As a personal example, on kentcdodds.com I have user accounts and originally I had my user roles field optional. I later realized this didn't work well for me, so I decided to switch it to a required field. I couldn't just npx prisma db push this change because I already had users who didn't have an assigned role.
So I created a migration to make the role field required and default to MEMBER. But then I modified the SQL to set the role to MEMBER for all users and then updated the role field for my own membership to ADMIN. Here's the migration file:
-- CreateEnum
CREATE TYPE "Role" AS ENUM ('ADMIN', 'MEMBER');

-- AlterTable
ALTER TABLE "User" ADD COLUMN     "role" "Role" DEFAULT E'MEMBER';

-- Manually written stuff:

-- Update all users to be members:
update "User" set role = E'MEMBER';

-- update me@kentcdodds.com to be ADMIN:
update "User" set role = E'ADMIN' where email = 'me@kentcdodds.com';

-- make role required
ALTER TABLE "User" ALTER COLUMN "role" SET NOT NULL;
And because I had prisma migrations set up to run whenever I deploy new code changes I got everything updated and working as expected.
However, I could have had some issues if I had users signing up while I was deploying this change. I would have had to do some extra work to make sure that I didn't lose any data.
πŸ¦‰ If you're curious what a migration directory looks like after a few migrations, check the migrations directory of kentcdodds.com.

Zero Downtime Migrations

At scale, dealing with migrations can be a little tricky. With lots of data and big changes, migrations can take a lot of time. We don't want to pull down our website while we're deploying a migration. But we also don't want to deploy a version of our app that doesn't work with the data in the database.
This is where the practice of "widen then narrow" comes in. The idea is that you make your database schema more permissive, deploy your app, then make your database schema more restrictive. For more on this subject, check out the Epic Stack database migrations docs.

When to use Migrate vs Push

The short answer is you use push when you're experimenting with schema changes and migrate when you're ready to commit to a schema change.
The longer answer is: Read the Prisma docs about it πŸ˜‡