fix: force recreate postgres with pgvector before app startup in deploy pipeline

DEPLOY-ISSUES-2.md

@@ -0,0 +1,44 @@
+# Deployment Still Failing — 502 Error
+
+## Date: 2026-05-12 (second attempt)
+
+## Problem
+After push of dc4244f, deploy relaunched BUT:
+- postgres container was NOT recreated with new pgvector/pgvector:pg16 image
+- Still running postgres:16-alpine — no pgvector — type "vector" does not exist
+- docker compose pull postgres added in deploy.yaml is not enough — need docker compose up -d --force-recreate postgres
+
+## Production logs (192.168.1.190)
+- memento-web: Restarting loop
+- memento-postgres: healthy but STILL on postgres:16-alpine
+- Error: type "vector" does not exist
+
+## Root cause
+deploy.yaml does docker compose build then docker compose up -d BUT:
+- postgres has no build: directive — uses image directly
+- docker compose up -d does NOT recreate container if image didnt change locally
+- Even with docker compose pull, existing container may not be recreated
+
+## What must be fixed
+
+### 1. deploy.yaml (.gitea/workflows/deploy.yaml)
+Add BEFORE docker compose up -d:
+- docker compose pull postgres
+- docker compose up -d --force-recreate postgres
+- Wait for postgres healthy before continuing
+Or use: docker compose up -d --force-recreate --build
+
+### 2. Verify docker-compose.yml has correct image
+image: pgvector/pgvector:pg16
+
+### 3. Prisma migration is stuck
+Old migration 20260512120000_pgvector_and_fts_search marked as failed in _prisma_migrations.
+Need to mark it resolved after pgvector is installed.
+
+### 4. Correct deployment order
+1. Pull + recreate postgres with pgvector
+2. Wait postgres healthy
+3. Create vector extension: CREATE EXTENSION IF NOT EXISTS vector
+4. Resolve failed migration in _prisma_migrations
+5. Start memento-web (will run migration)
+6. Start mcp-server