prisma-driver-adapter-implementation

Implement SqlDriverAdapterFactory with connect(), then SqlDriverAdapter with queryRaw, executeRaw, executeScript, startTransaction, and dispose. Typecheck against the exact @prisma/driver-adapter-utils version shipped with your target Prisma release.

startTransaction acquires one dedicated connection, begins the transaction, and applies the isolation level. commit() and rollback() are lifecycle cleanup hooks that release the connection once; Prisma issues the actual SQL COMMIT or ROLLBACK via executeRaw, and savepoints are optional methods on the Transaction object.

Yes, through optional createSavepoint, rollbackToSavepoint, and releaseSavepoint methods on the Transaction object. Implement them only where the provider supports savepoints, and never track transaction depth on the shared adapter since parallel transactions make global depth incorrect.

P2039 surfaces when an unmapped driver error reaches Prisma. Preserve originalCode and originalMessage in your DriverAdapterError so the fallback is useful, map known conditions to structured MappedError kinds, and rethrow genuinely unexpected non-driver errors instead of fabricating GenericJs ids.

Implement SqlMigrationAwareDriverAdapterFactory only if connectToShadowDb() can create an isolated shadow database, connect to it, and drop it during disposal or failure cleanup. Never point the shadow adapter at the primary database, and use quoted, cryptographically unique identifiers.