fix: pgvector deployment — idempotent migration, pgvector image, schema sync

- docker-compose.yml: switch postgres:16-alpine to pgvector/pgvector:pg16
- migration: rewrite with IF NOT EXISTS guards, DO block for safe
  text→vector(1536) conversion, handles partial/re-run states
- schema.prisma (both): add @default(now()) on NoteEmbedding.updatedAt,
  sync mcp-server embedding type to Unsupported("vector(1536)")
- deploy.yaml: add docker compose pull postgres before build

.gitea/workflows/deploy.yaml

@@ -125,6 +125,7 @@ jobs:
             git reset --hard origin/main
 
             echo "=== Building ==="
+            docker compose pull postgres
             docker compose build memento-note
             docker compose build mcp-server
 

DEPLOY-ISSUES.md

@@ -0,0 +1,67 @@
+# Déployment Issues — Migration pgvector
+
+## Date: 2026-05-12
+
+## Contexte
+Le commit `03e6a62` (migrate semantic search to pgvector + full-text search) a été pushé sur `main`. Le pipeline Gitea `deploy.yaml` a déployé automatiquement sur `192.168.1.190`. L'application **ne démarre plus** — erreur 502.
+
+## Production Environment
+- **Serveur**: 192.168.1.190 (ops-user, sudo requires password)
+- **docker-compose.yml**: `/opt/memento/docker-compose.yml` (root-owned, ops-user cannot write directly)
+- **PostgreSQL image**: `postgres:16-alpine` — **pgvector NOT available**
+- **Database**: `memento` on `memento-postgres` container
+- **102 embeddings** existent dans la table `NoteEmbedding`
+
+## Problèmes rencontrés
+
+### 1. Extension pgvector manquante
+- L'image PostgreSQL est `postgres:16-alpine` — pas de pgvector
+- Il faut changer pour `pgvector/pgvector:pg16` dans le docker-compose
+- **OPS-USER ne peut pas écrire dans `/opt/memento/docker-compose.yml`** (root-owned)
+- Le `deploy.yaml` Gitea devrait gérer ce changement d'image
+
+### 2. Migration Prisma failed (précédente)
+- `20260510123000_add_notebook_hierarchy_and_trash` était déjà en échec depuis le 10 mai
+- **Résolu manuellement**: `UPDATE _prisma_migrations SET finished_at = NOW() WHERE migration_name = '...' AND finished_at IS NULL;`
+
+### 3. Nouvelle migration pgvector failed
+- `20260512120000_pgvector_and_fts_search` échoue car:
+  - Le type `vector` n'existe pas (pgvector pas installé)
+  - La colonne `updatedAt` sur `NoteEmbedding` n'a pas de default value (102 rows existants)
+- **Partiellement résolu**: `ALTER TABLE "NoteEmbedding" ADD COLUMN "updatedAt" TIMESTAMP(3) NOT NULL DEFAULT NOW();`
+
+### 4. La migration doit:
+- D'abord installer l'extension pgvector: `CREATE EXTENSION IF NOT EXISTS vector;`
+- Puis modifier la colonne `embedding` de `String` (JSON) vers `vector(1536)`
+- Ajouter l'index HNSW
+- Ajouter le FTS tsvector sur `Note`
+- Tout cela doit être fait dans un ordre précis dans le fichier de migration Prisma
+
+## Ce qui doit être corrigé dans le code
+
+### docker-compose.yml
+```yaml
+# AVANT
+image: postgres:16-alpine
+# APRÈS
+image: pgvector/pgvector:pg16
+```
+
+### Migration Prisma (fichier dans prisma/migrations/)
+La migration doit:
+1. `CREATE EXTENSION IF NOT EXISTS vector;` — en raw SQL
+2. Ajouter `updatedAt` avec `DEFAULT NOW()` sur `NoteEmbedding`
+3. Convertir la colonne `embedding` de text/JSON vers `vector(1536)` — avec conversion des données existantes
+4. Créer l'index HNSW
+5. Ajouter la colonne tsvector + trigger sur `Note`
+6. Être **idempotente** — pouvoir tourner plusieurs fois sans erreur
+
+### schema.prisma
+- Vérifier que `updatedAt` a `@default(now())` et `@updatedAt`
+- Vérifier que le type `Unsupported("vector(1536)")` est correct
+
+##État actuel de la prod
+- `memento-web`: **Restarting** en boucle (migration failed → app ne démarre pas)
+- `memento-mcp`: **unhealthy**
+- `memento-postgres`: healthy, mais migration en état incohérent
+- Le site retourne **502 Bad Gateway**

docker-compose.yml

@@ -3,7 +3,7 @@ services:
   # PostgreSQL - Shared Database
   # ============================================
   postgres:
-    image: postgres:16-alpine
+    image: pgvector/pgvector:pg16
     container_name: memento-postgres
     restart: unless-stopped
     environment:

mcp-server/prisma/schema.prisma

@@ -129,11 +129,12 @@ model Note {
 }
 
 model NoteEmbedding {
-  id        String   @id @default(cuid())
-  noteId    String   @unique
-  embedding String
-  createdAt DateTime @default(now())
-  note      Note     @relation(fields: [noteId], references: [id], onDelete: Cascade)
+  id        String                   @id @default(cuid())
+  noteId    String                   @unique
+  embedding Unsupported("vector(1536)")
+  createdAt DateTime                 @default(now())
+  updatedAt DateTime                 @default(now()) @updatedAt
+  note      Note                     @relation(fields: [noteId], references: [id], onDelete: Cascade)
 
   @@index([noteId])
 }

memento-note/prisma/migrations/20260512120000_pgvector_and_fts_search/migration.sql

@@ -1,38 +1,71 @@
 -- Phase 1: Enable pgvector extension
 CREATE EXTENSION IF NOT EXISTS vector;
 
--- Phase 2: Add native vector column to NoteEmbedding
--- Convert existing JSON-string embeddings to native vector(1536)
-ALTER TABLE "NoteEmbedding" ADD COLUMN "vec" vector(1536);
+-- Phase 2: Convert embedding column from text/JSON to vector(1536) if needed
+DO $$
+DECLARE
+    _udt text;
+    _vec_tmp_exists boolean;
+BEGIN
+    SELECT udt_name INTO _udt
+    FROM information_schema.columns
+    WHERE table_schema = 'public'
+      AND table_name = 'NoteEmbedding'
+      AND column_name = 'embedding';
 
--- Migrate existing data: parse JSON arrays into pgvector format
-UPDATE "NoteEmbedding"
-SET "vec" = ("embedding"::jsonb)::text::vector(1536)
-WHERE "embedding" IS NOT NULL;
+    IF _udt IS NOT NULL AND _udt != 'vector' THEN
+        SELECT EXISTS (
+            SELECT 1 FROM information_schema.columns
+            WHERE table_schema = 'public'
+              AND table_name = 'NoteEmbedding'
+              AND column_name = '_vec_tmp'
+        ) INTO _vec_tmp_exists;
 
--- Drop old string column, rename new one
-ALTER TABLE "NoteEmbedding" DROP COLUMN "embedding";
-ALTER TABLE "NoteEmbedding" RENAME COLUMN "vec" TO "embedding";
+        IF NOT _vec_tmp_exists THEN
+            ALTER TABLE "NoteEmbedding" ADD COLUMN "_vec_tmp" vector(1536);
+        END IF;
 
--- Add updatedAt column for tracking reindex freshness
-ALTER TABLE "NoteEmbedding" ADD COLUMN "updatedAt" TIMESTAMP NOT NULL DEFAULT now();
+        UPDATE "NoteEmbedding"
+        SET "_vec_tmp" = ("embedding"::jsonb)::text::vector(1536)
+        WHERE "embedding" IS NOT NULL
+          AND "_vec_tmp" IS NULL;
+
+        ALTER TABLE "NoteEmbedding" DROP COLUMN "embedding";
+        ALTER TABLE "NoteEmbedding" RENAME COLUMN "_vec_tmp" TO "embedding";
+    ELSIF _udt IS NULL THEN
+        SELECT EXISTS (
+            SELECT 1 FROM information_schema.columns
+            WHERE table_schema = 'public'
+              AND table_name = 'NoteEmbedding'
+              AND column_name = '_vec_tmp'
+        ) INTO _vec_tmp_exists;
+
+        IF _vec_tmp_exists THEN
+            ALTER TABLE "NoteEmbedding" RENAME COLUMN "_vec_tmp" TO "embedding";
+        END IF;
+    END IF;
+END $$;
+
+-- Add updatedAt with DEFAULT NOW() if not present
+ALTER TABLE "NoteEmbedding" ADD COLUMN IF NOT EXISTS "updatedAt" TIMESTAMP(3) NOT NULL DEFAULT NOW();
 
 -- HNSW index for fast approximate nearest neighbor search (cosine distance)
-CREATE INDEX "NoteEmbedding_embedding_hnsw_idx" ON "NoteEmbedding"
+CREATE INDEX IF NOT EXISTS "NoteEmbedding_embedding_hnsw_idx" ON "NoteEmbedding"
   USING hnsw ("embedding" vector_cosine_ops)
   WITH (m = 16, ef_construction = 64);
 
 -- Phase 3: Add full-text search tsvector column to Note
-ALTER TABLE "Note" ADD COLUMN "tsv" tsvector;
+ALTER TABLE "Note" ADD COLUMN IF NOT EXISTS "tsv" tsvector;
 
--- Populate tsv from existing title + content
+-- Populate tsv where still NULL
 UPDATE "Note"
 SET "tsv" =
   setweight(to_tsvector('simple', COALESCE("title", '')), 'A') ||
-  setweight(to_tsvector('simple', COALESCE("content", '')), 'B');
+  setweight(to_tsvector('simple', COALESCE("content", '')), 'B')
+WHERE "tsv" IS NULL;
 
 -- GIN index for fast FTS queries
-CREATE INDEX "Note_tsv_gin_idx" ON "Note" USING gin ("tsv");
+CREATE INDEX IF NOT EXISTS "Note_tsv_gin_idx" ON "Note" USING gin ("tsv");
 
 -- Trigger function to auto-update tsv on INSERT or UPDATE of title/content
 CREATE OR REPLACE FUNCTION "note_tsv_trigger"() RETURNS trigger AS $$
@@ -44,7 +77,7 @@ BEGIN
 END;
 $$ LANGUAGE plpgsql;
 
--- Attach trigger
+-- Attach trigger (DROP IF EXISTS + CREATE is idempotent)
 DROP TRIGGER IF EXISTS "note_tsv_update" ON "Note";
 CREATE TRIGGER "note_tsv_update"
   BEFORE INSERT OR UPDATE OF "title", "content" ON "Note"

memento-note/prisma/schema.prisma

@@ -302,7 +302,7 @@ model NoteEmbedding {
   noteId    String   @unique
   embedding Unsupported("vector(1536)")
   createdAt DateTime @default(now())
-  updatedAt DateTime @updatedAt
+  updatedAt DateTime @default(now()) @updatedAt
   note      Note     @relation(fields: [noteId], references: [id], onDelete: Cascade)
 
   @@index([noteId])