fix(celery): Reset DB connection pools for forked worker processes (#13350)

* Reset sqlalchemy connection pool on celery process fork

* Fix race condition with async chart loading state

* pylint: ignore

* prettier

superset/tasks/celery_app.py

@@ -19,13 +19,16 @@
 This is the main entrypoint used by Celery workers. As such,
 it needs to call create_app() in order to initialize things properly
 """
+from typing import Any
+
+from celery.signals import worker_process_init
 
 # Superset framework imports
 from superset import create_app
-from superset.extensions import celery_app
+from superset.extensions import celery_app, db
 
 # Init the Flask app / configure everything
-create_app()
+flask_app = create_app()
 
 # Need to import late, as the celery_app will have been setup by "create_app()"
 # pylint: disable=wrong-import-position, unused-import
@@ -33,3 +36,10 @@ from . import cache, schedules, scheduler  # isort:skip
 
 # Export the celery app globally for Celery (as run on the cmd line) to find
 app = celery_app
+
+
+@worker_process_init.connect
+def reset_db_connection_pool(**kwargs: Any) -> None:  # pylint: disable=unused-argument
+    with flask_app.app_context():
+        # https://docs.sqlalchemy.org/en/14/core/connections.html#engine-disposal
+        db.engine.dispose()