Files
canina/backend/prisma/MIGRATION_BASELINE.md
T

29 lines
3.5 KiB
Markdown

# Prisma migration history and Orchestration v4 rollout boundary
The checked-in history is **not a complete baseline of the current Prisma schema**. Do not assume that replaying it into an empty database can recreate the application, or that existing production tables imply corresponding migration records. Never use `prisma db push`, reset, or edit applied SQL to reconcile this.
## Actual tracked history
| Migration | What its SQL actually does |
| --- | --- |
| `20260526145407_init` | Creates users, addresses, wallets, products, product ingredients/symptoms, pets and medical conditions, reminders/completions, health logs, coupons, orders and order items, with their indexes/foreign keys. |
| `20260526160916_add_ui_texts_and_scientific_terms` | Creates `ui_texts` and `scientific_terms`. |
| `20261013210253_add_admin_modules` | Creates only `competitor_price_logs`, `b2b_leads`, `price_suggestions`, and `sms_campaigns`, with indexes and product foreign keys. It does **not** create legacy approvals, department tasks or audit logs. |
| `20261015000000_add_support_conversations` | Creates support conversations/messages and their foreign keys. |
| `20261020000000_add_blog_freshness_authorship` | Adds three columns to an assumed-existing `blogs` table. No earlier tracked migration creates `blogs`; fresh replay is therefore incomplete before v4. |
| `20261022000000_add_orchestration_v4_core` | Additively creates nine `orchestration_*` tables, indexes and internal foreign keys; it neither creates nor copies legacy governance tables. |
There is no checked-in CREATE TABLE migration for `approval_requests`, `department_tasks`, or `audit_logs`. Their models/controllers are not evidence that either live environment has these tables, nor proof of how any historical database acquired them. The broader missing legacy schema must be reconciled separately. This document does not certify live table existence or migration status.
## Required rollout safeguards (not authorization to deploy)
1. Back up each environment and verify restore capability. Capture its actual schema and `_prisma_migrations` history read-only; compare those independently against all tracked SQL, including prerequisites such as `blogs`.
2. Rehearse the exact planned chain on a disposable database restored from a reviewed baseline. A failed empty replay is a legacy-history gap, not evidence that the additive v4 SQL alone is faulty. Resolve the missing baseline through a separately reviewed migration/baseline plan, never by skipping errors.
3. `prisma migrate resolve --applied <name>` only records history; it creates no objects. Use it only for a specifically verified, fully matching existing migration and reviewed environment-specific plan. Never mark the whole chain applied or mark missing tables as applied.
4. After baseline reconciliation and rehearsal pass, apply reviewed pending migrations through CI/CD only using `prisma migrate deploy`. Verify all nine v4 tables, constraints, migration records and application health before considering the read surface ready.
5. Keep v4 mutations, chat and runtime capabilities denied. Retiring legacy mutation routes does not authorize a replacement execution path.
No deployment, live DDL, baseline resolution or migration SQL change is part of this hardening milestone. Schema rollout remains a separate gate, not full-system completion.
Rollback is application-first: stop use of the new module while retaining append-only records. Do not drop tables in automated rollback; removal requires separately reviewed archival migration.