The work
Scheduled messages, reminders, the missed-activity email, webhook retries and retention sweeps all run on one durable job queue in the database, and messages, presence and calls on one realtime socket and a media server we operate ourselves. Today that is one of each. The work is taking all of it from one server to many: the queue under real load, the socket fan-out across processes through the broker and the shared cache, and the media server beside them.
What we look for
- You have run a queue in production and have opinions about at-least-once delivery.
- You have scaled a WebSocket fan-out past one process and know where it hurts.
- You write the runbook before the incident.