Socket.IO requires the sticky sessions, so when it’s runned in Docker Swarm environment, the Swarm load balancer can forward the request to any node where the service is launched and there will be the 400 errors. The solution is to limit the number of nodes for Socket.IO service to one or to use another load balancer like Traefik or HAProxy.

If you have some long background tasks, sometimes it’s useful to notify a user about the progress. In this article i put an example for flask, celery and socket.io. In short form it’s described in flask-socketio documentation.