| name | harper-schema-graphql |
| description | editing a Harper (HarperDB/Fabri… |
Harper schema.graphql
Overview
Harper defines its database tables and types from GraphQL schema files, loaded
by the graphqlSchema extension (default *.graphql; this project uses
harper-app/schema.graphql). The schema is the source of truth for the data model:
the tables it declares are what resources extend ([[harper-resources]]) and what
the REST/GraphQL surface exposes.
Defining tables
A table is a GraphQL type. Harper-specific directives mark a type as a persisted
table and control exposure, primary keys, indexes, audit timestamps, sealed
records, and relationships. Lisa's Harper Fabric template pins harperdb to the
Harper 4 line (^4.7.29), so use this v4 directive reference for template
projects:
type Dog @table @export(name: "dogs") {
id: Long @primaryKey
name: String @indexed
breed: String
ownerId: Long @indexed
owner: Owner @relationship(from: ownerId)
createdAt: Long @createdTime
updatedAt: Long @updatedTime
}
| Directive | Scope | Syntax | Use |
|---|
@table | Type | type Product @table { ... } | Creates a persisted table named after the type. Optional arguments: table: "products" to override the table name, database: "commerce" to choose a database, expiration: 3600 for TTL-style records, and audit: true to force audit logging. |
@export | Type | type Product @table @export(name: "products") { ... } | Exposes the table as a resource endpoint for REST/MQTT and related surfaces. name is optional; without it the type name is the path segment. |
@sealed | Type | type Product @table @sealed { ... } | Rejects undeclared properties. Omit it when the table intentionally accepts extra record fields. |
@primaryKey | Field | id: Long @primaryKey | Marks the unique table key. If omitted on insert, Harper v4 can auto-generate a UUID for String/ID keys or an auto-incrementing integer for Int/Long/Any keys. Prefer Long or Any for generated numeric keys. |
@indexed | Field | sku: String @indexed | Adds a secondary index used by REST filters, SQL, and NoSQL/search paths. Array fields index each element. For vectors in Harper v4.6+, use embedding: [Float] @indexed(type: "HNSW"). |
@createdTime | Field | createdAt: Long @createdTime | Writes Unix epoch milliseconds when the record is created. |
@updatedTime | Field | updatedAt: Long @updatedTime | Writes Unix epoch milliseconds whenever the record is updated. |
@relationship(from: field) | Field | owner: Owner @relationship(from: ownerId) |
Use v5 docs only when the downstream project has intentionally moved off the Lisa
template's Harper 4 dependency; do not mix v5-only syntax into a harperdb
^4.7.29 project.
How the schema drives the app
- The
graphqlSchema extension creates/migrates tables from the schema on load.
rest: true exposes exported tables/resources as REST endpoints; the GraphQL
surface is generated from the same schema.
- Resources that
extends tables.X depend on X existing in the schema. A rename
or removal in the schema is a breaking change for every resource and verify path
that references it.
Schema evolution
Schema changes are deploy-time data-model changes, not just type edits. Harper's
graphqlSchema extension ensures declared tables and attributes exist when the
component loads, but it does not perform semantic data migrations such as
renaming tables, copying field values, or rewriting existing rows for you.
Classify each change before editing:
| Change | Compatibility | What Harper does | Required agent work |
|---|
| Add optional field | Usually safe | Adds the declared attribute shape; existing rows read as missing/null until written. | Update resources, seeds, and verification that should include the field. |
| Add required/non-null field | Breaking for existing rows and writers | The schema can declare the field, but existing records do not magically gain valid values. | Backfill first or deploy as optional, populate, then tighten in a later deploy. |
Add @indexed | Usually safe, operationally sensitive | Creates/uses a secondary index for the attribute. Large tables may pay rebuild cost. | Verify filtered REST/search paths and note index build risk in the deploy runbook. |
Add @sealed | Breaking when rows or writers use extra properties | Future writes with undeclared properties are rejected. | Audit current data/writers, declare needed fields, or migrate callers before sealing. |
| Rename field | Breaking | Treated as a new field; the old field and its values are not transformed. | Add the new field, copy values with a migration, update code, verify, then remove old usage. |
| Rename type/table | Breaking | Harper v4 does not rename tables; changing the type name creates a new empty table and leaves the old table/data untouched. | Create the new table, copy data, update resources/routes, verify both read/write paths, then retire the old table intentionally. |
| Change field type | Breaking | Existing stored values are not coerced into the new type in a controlled migration. | Add a replacement field/table, transform data with code, verify, then remove old usage. |
| Remove field/type | Breaking | Schema no longer declares it, but dependent resources, routes, queries, seeds, and clients can still reference it. | Delete references first, run a migration/cleanup if needed, and verify old API paths fail or redirect intentionally. |
Migration recipe for production data:
- Add the new schema shape in a backward-compatible way: new table or nullable
replacement field, keeping the old field/table available.
- Write a one-shot migration using a Harper resource method or Operations
API/script that reads old rows, transforms values, and writes the new shape.
Make it idempotent; reruns should skip rows already migrated or compare a
migration marker.
- Deploy to one Fabric environment and run the migration before switching
readers/writers. For replicated Fabric deployments, assume every node may see
the new code/schema at slightly different times; keep old and new reads
compatible until replication and smoke checks are green.
- Update resources, REST/GraphQL queries, data-loader seeds, and verify scripts
in the same PR. A schema PR is incomplete if
bun run verify or the project
smoke path cannot prove the migrated read/write behavior.
- Roll back by reverting code/schema only when the old field/table remains
intact. Once cleanup drops old data or callers, rollback needs a reverse
migration and a restored compatibility path.
Project conventions
schema.graphql is source and lives at the component root that Fabric
packages (harper-app/schema.graphql). Keep it there.
- A schema change is a data-model change. In the same change:
- Update the conceptual schema docs.
- Update any verify/smoke paths that assert joins, relationships, or row counts —
those assertions encode the data model and must track it.
- Update resources and seed/data-loader paths that reference changed tables.
- Document deploy-affecting consequences (new tables, migrations) in the Fabric
runbook. See [[harper-build-and-deploy]].
Verification
After a schema change, boot the app (harper dev harper-app or the project run
command) and confirm tables migrate cleanly and the dependent endpoints respond.
Run any verify path that asserts row counts or joins against the changed model.
Sources