fix: create fresh SQLite engine after fork instead of reusing dead fd

The grandchild process closes all inherited fds (3..maxfd) to detach
from uWSGI. session_factory().get_bind() returned a session bound to
the now-dead SQLite fd. Capture the DB URL before forking, create a
fresh engine in the child.

make_post_sell/GIT_HASH

@@ -1 +1 @@
-d071955
+85b1a2d

make_post_sell/lib/karaoke.py

@@ -164,6 +164,9 @@ def backfill_karaoke_async(shop_id, session_factory, app_settings):
     shop_id_str = str(shop_id)
     lockfile = f"/tmp/karaoke_backfill_{shop_id_str}.lock"
 
+    # Capture the DB URL before forking — the engine's fd dies after fork+close
+    db_url = str(session_factory().get_bind().url)
+
     # Quick check — if lockfile exists and is locked, skip
     try:
         check_fd = open(lockfile, "w")
@@ -207,11 +210,15 @@ def backfill_karaoke_async(shop_id, session_factory, app_settings):
         lock_fd.flush()
 
         import boto3
+        from sqlalchemy import create_engine
         from sqlalchemy.orm import Session as SASession
         from ..models.product import Product, get_media_type
         from ..models.shop import Shop
 
-        session = SASession(bind=session_factory().get_bind())
+        # Create a fresh engine — the parent's SQLite connection fd was closed
+        # after fork (lines 195-199). Reusing session_factory would hit a dead fd.
+        engine = create_engine(db_url)
+        session = SASession(bind=engine)
 
         def _make_s3():
             return boto3.session.Session().client(
@@ -338,6 +345,7 @@ def backfill_karaoke_async(shop_id, session_factory, app_settings):
             log.exception("Background karaoke backfill failed for shop %s", shop_id_str)
         finally:
             session.close()
+            engine.dispose()
     except (IOError, OSError):
         pass  # couldn't acquire lock — another backfill is running
     except Exception: