fix: pgvector deployment — idempotent migration, pgvector image, schema sync
All checks were successful
Deploy to Production / Build and Deploy (push) Successful in 2m21s

- 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
This commit is contained in:
Antigravity
2026-05-12 07:28:03 +00:00
parent 03e6a62b80
commit dc4244f2ad
6 changed files with 127 additions and 25 deletions

View File

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

67
DEPLOY-ISSUES.md Normal file
View File

@@ -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**

View File

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

View File

@@ -131,8 +131,9 @@ model Note {
model NoteEmbedding { model NoteEmbedding {
id String @id @default(cuid()) id String @id @default(cuid())
noteId String @unique noteId String @unique
embedding String embedding Unsupported("vector(1536)")
createdAt DateTime @default(now()) createdAt DateTime @default(now())
updatedAt DateTime @default(now()) @updatedAt
note Note @relation(fields: [noteId], references: [id], onDelete: Cascade) note Note @relation(fields: [noteId], references: [id], onDelete: Cascade)
@@index([noteId]) @@index([noteId])

View File

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

View File

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