fix(operator): two-phase migrations — cold SQL then warm AGE/graph

Split migrations into two phases:
- schema/migrations/     — standard SQL (tables, indexes, permissions)
- schema/migrations-warm/ — AGE/Cypher DDL (needs running graph engine)

Move 058 (precreate_graph_labels) to migrations-warm. After cold
migrations, restart postgres so AGE is fully initialized, verify it's
healthy, then apply warm migrations.

Also:
- Migration runner continues past failures instead of stopping
- configure.py ai-provider handles missing catalog table gracefully
- Runner supports --warm flag to select migration directory

Author: aaronsb

operator/configure.py

@@ -143,16 +143,20 @@ def cmd_ai_provider(self, args):
                     print(f"📝 Current: {current['provider']} / {current['model_name']}")
 
                 # Validate model exists in catalog (ADR-800) — warn but don't block
-                cur.execute(
-                    """SELECT id, enabled FROM kg_api.provider_model_catalog
-                       WHERE provider = %s AND model_id = %s AND category = 'extraction'""",
-                    (provider, model),
-                )
-                catalog_row = cur.fetchone()
-                if catalog_row is None:
-                    print(f"⚠️  Model '{model}' not in catalog for {provider}. Run: models refresh {provider}")
-                elif not catalog_row['enabled']:
-                    print(f"⚠️  Model '{model}' is in catalog but not enabled. Run: models enable {catalog_row['id']}")
+                try:
+                    cur.execute(
+                        """SELECT id, enabled FROM kg_api.provider_model_catalog
+                           WHERE provider = %s AND model_id = %s AND category = 'extraction'""",
+                        (provider, model),
+                    )
+                    catalog_row = cur.fetchone()
+                    if catalog_row is None:
+                        print(f"⚠️  Model '{model}' not in catalog for {provider}. Run: models refresh {provider}")
+                    elif not catalog_row['enabled']:
+                        print(f"⚠️  Model '{model}' is in catalog but not enabled. Run: models enable {catalog_row['id']}")
+                except Exception:
+                    # Table may not exist yet (migrations pending)
+                    conn.rollback()
 
                 # Insert/update configuration
                 cur.execute(

operator/database/migrate-db.sh

@@ -60,6 +60,7 @@ fi
 DRY_RUN=false
 AUTO_CONFIRM=false
 VERBOSE=false
+WARM_MODE=false
 
 for arg in "$@"; do
     case $arg in
@@ -72,20 +73,25 @@ for arg in "$@"; do
         -v|--verbose)
             VERBOSE=true
             ;;
+        --warm)
+            WARM_MODE=true
+            ;;
         -h|--help)
             echo "Usage: $0 [OPTIONS]"
             echo ""
             echo "Options:"
             echo "  --dry-run       Show what would be applied without making changes"
             echo "  -y, --yes       Skip confirmation prompt"
             echo "  -v, --verbose   Show detailed migration SQL"
+            echo "  --warm          Apply warm migrations (require running AGE/graph engine)"
             echo "  -h, --help      Show this help message"
             echo ""
             echo "Examples:"
             echo "  $0                    # Interactive mode with confirmation"
             echo "  $0 --dry-run          # Preview pending migrations"
             echo "  $0 -y                 # Apply migrations without confirmation"
             echo "  $0 -y --verbose       # Apply with detailed output"
+            echo "  $0 -y --warm          # Apply warm migrations (after app start)"
             exit 0
             ;;
         *)
@@ -96,6 +102,11 @@ for arg in "$@"; do
     esac
 done
 
+# Select migration directory based on mode
+if [ "$WARM_MODE" = true ]; then
+    MIGRATIONS_DIR="schema/migrations-warm"
+fi
+
 # Header
 if [ "$DRY_RUN" = true ]; then
     echo -e "${CYAN}📦 Database Migration Runner (DRY RUN)${NC}"
@@ -296,13 +307,15 @@ for pending in "${PENDING_MIGRATIONS[@]}"; do
         FAILED_COUNT=$((FAILED_COUNT + 1))
 
         # Show error details
-        echo -e "${RED}Error details:${NC}"
+        echo -e "${YELLOW}Error details:${NC}"
         echo "$MIGRATION_OUTPUT" | grep -i "ERROR:\|ROLLBACK\|LINE" | sed 's/^/  /'
 
-        echo ""
-        echo -e "${RED}✗ Migration $VERSION failed - stopping${NC}"
-        echo -e "${YELLOW}Fix the migration and try again${NC}"
-        exit 1
+        # Continue to next migration instead of stopping.
+        # Failed migration won't be recorded (ON_ERROR_STOP prevents the
+        # INSERT INTO schema_migrations from running), so it will be
+        # retried on the next run when the database may be fully ready.
+        echo -e "${YELLOW}  ⚠️  Skipping — will retry on next migration run${NC}"
+        continue
     fi
 
     echo -e "${GREEN}  ✅ Migration $VERSION applied successfully${NC}"

operator/lib/start-infra.sh

@@ -124,6 +124,21 @@ else
 fi
 cd "$DOCKER_DIR"
 
+# Restart postgres so AGE extension is fully initialized before warm migrations
+echo -e "${BLUE}  Restarting PostgreSQL for AGE initialization...${NC}"
+docker restart $POSTGRES_CONTAINER >/dev/null 2>&1
+for i in {1..15}; do
+    if docker exec $POSTGRES_CONTAINER psql -U ${POSTGRES_USER:-admin} -d ${POSTGRES_DB:-knowledge_graph} -tAc "SELECT 1" >/dev/null 2>&1; then
+        echo -e "${GREEN}✓ PostgreSQL restarted${NC}"
+        break
+    fi
+    if [ $i -eq 15 ]; then
+        echo -e "${RED}✗ PostgreSQL did not come back after restart${NC}"
+        exit 1
+    fi
+    sleep 2
+done
+
 # Check AGE extension
 AGE_EXT=$(docker exec $POSTGRES_CONTAINER psql -U ${POSTGRES_USER:-admin} -d ${POSTGRES_DB:-knowledge_graph} -tAc "SELECT COUNT(*) FROM pg_extension WHERE extname='age'" 2>/dev/null || echo "0")
 if [ "$AGE_EXT" -gt 0 ]; then
@@ -132,6 +147,18 @@ else
     echo -e "${RED}✗ AGE extension not found${NC}"
 fi
 
+# Apply warm migrations (require running AGE/graph engine)
+cd "$PROJECT_ROOT"
+if [ -f "$PROJECT_ROOT/operator/database/migrate-db.sh" ] && [ -d "schema/migrations-warm" ]; then
+    WARM_COUNT=$(ls schema/migrations-warm/*.sql 2>/dev/null | wc -l)
+    if [ "$WARM_COUNT" -gt 0 ]; then
+        echo -e "${BLUE}  Applying warm migrations (AGE/graph)...${NC}"
+        "$PROJECT_ROOT/operator/database/migrate-db.sh" -y --warm 2>&1 | grep -E "✓|✅|→|⚠️|✗" || true
+        echo -e "${GREEN}✓ Warm migrations applied${NC}"
+    fi
+fi
+cd "$DOCKER_DIR"
+
 # Show applied migrations
 MIGRATION_LIST=$(docker exec $POSTGRES_CONTAINER psql -U ${POSTGRES_USER:-admin} -d ${POSTGRES_DB:-knowledge_graph} -tAc "SELECT version || ' - ' || name FROM public.schema_migrations ORDER BY version" 2>/dev/null || echo "")