Kordant/tasks/shieldai-unified-restructure/09-drizzle-schema-migration.md

09. Database — Migrate Full Prisma Schema to Drizzle ORM

meta: id: shieldai-unified-restructure-09 feature: shieldai-unified-restructure priority: P0 depends_on: [] tags: [backend, database, drizzle, schema]

objective:

• Convert the entire 900-line Prisma schema from packages/db/prisma/schema.prisma into Drizzle ORM TypeScript schema files in web/src/server/db/schema.ts and related files. This is the foundational data layer for the unified monolith.

deliverables:

• web/src/server/db/schema.ts — Main schema file with all tables:
  • User & Authentication (User, Account, Session, DeviceToken)
  • Family & Subscription (FamilyGroup, FamilyGroupMember, Subscription)
  • DarkWatch (WatchlistItem, Exposure)
  • Notifications & Alerts (Alert)
  • VoicePrint (VoiceEnrollment, VoiceAnalysis, AnalysisJob, AnalysisResult)
  • SpamShield (SpamFeedback, SpamRule)
  • Audit & Analytics (AuditLog, KPISnapshot)
  • Correlation (NormalizedAlert, CorrelationGroup)
  • Reports (SecurityReport)
  • Marketing (WaitlistEntry, BlogPost)
  • HomeTitle (PropertyWatchlistItem, PropertySnapshot, PropertyChange)
  • RemoveBrokers (InfoBroker, RemovalRequest, BrokerListing)
• web/src/server/db/schema/ — Optional split directory if single file becomes unwieldy:
  • auth.ts, subscription.ts, darkwatch.ts, voiceprint.ts, spamshield.ts, alerts.ts, correlation.ts, reports.ts, marketing.ts, hometitle.ts, removebrokers.ts
• All enums defined as TypeScript const arrays or Drizzle pgEnum
• All indexes, unique constraints, and foreign keys preserved
• Relations defined using Drizzle's relations() helper

steps:

1. Read and analyze packages/db/prisma/schema.prisma completely. Document every model, enum, relation, index, and constraint.
2. Install Drizzle ORM and PostgreSQL driver in web/:
   • drizzle-orm
   • pg (node-postgres) or @neondatabase/serverless if using Neon
   • drizzle-kit for migrations
3. Create web/src/server/db/schema.ts (or split directory).
4. For each Prisma model, create a Drizzle table definition:
   • Map Prisma field types to Drizzle column types:
     • String → varchar, text, uuid
     • Int → integer
     • Float → real
     • Boolean → boolean
     • DateTime → timestamp
     • Json → jsonb
     • String[] → text().array()
   • Preserve @id, @default(uuid()), @unique, @index, @relation
   • Map Prisma enums to Drizzle pgEnum()
5. Define all indexes using Drizzle's .index() and .unique() on table definitions.
6. Define relations using relations() helper for:
   • User → accounts, sessions, familyGroups, subscriptions, alerts, voice enrollments, etc.
   • Subscription → watchlistItems, exposures, alerts, propertyWatchlistItems, removalRequests
   • WatchlistItem → exposures
   • PropertyWatchlistItem → snapshots, changes
   • And all other one-to-many / many-to-one relations
7. Export a unified schema object for use with Drizzle Kit.
8. Verify the schema compiles without TypeScript errors.
9. Generate an ER diagram or schema summary for documentation.

steps:

• Unit: Schema file compiles without TS errors
• Integration: drizzle-kit generate produces migration SQL that matches Prisma schema structure
• Compare: Automated or manual comparison of Prisma schema vs Drizzle schema to ensure no models/fields are missing

acceptance_criteria:

• Every model from Prisma schema has a corresponding Drizzle table definition
• All enums are defined and used correctly
• All primary keys, unique constraints, and indexes are preserved
• All foreign key relations are defined using Drizzle relations
• drizzle-kit generate runs successfully and outputs SQL
• Generated SQL creates tables with correct column types and constraints
• No data loss: the new schema is structurally equivalent to the old one

validation:

• Run cd web && npx drizzle-kit generate and inspect generated SQL
• Compare table count: Prisma schema has X models, Drizzle schema has X tables
• Verify enum values match exactly between Prisma and Drizzle
• Run npx drizzle-kit push against a local PostgreSQL instance and confirm all tables are created

notes:

• This is the most critical backend task. A missing field or incorrect relation will cascade into broken tRPC routers.
• Prisma's implicit many-to-many relations must be explicitly defined as junction tables in Drizzle.
• Prisma's @updatedAt auto-timestamp can be replicated with Drizzle's $onUpdateFn(() => new Date()).
• Keep the old Prisma schema file as reference until task 41.
• Consider using drizzle-zod later (task 11+) to auto-generate validation schemas from Drizzle tables.
• The schema uses PostgreSQL-specific features (arrays, enums, jsonb). Ensure the Drizzle definitions use pgTable, pgEnum, etc.