Queues fail at the worst moment. These are the settings that keep Horizon alive under load.
Supervisor Tuning
Balance auto-scaling with maxProcesses per queue so one noisy tenant cannot starve the rest.
Timeouts and Retries
Short timeouts, exponential backoff, and a dead-letter queue for poison jobs.
Monitoring
Alert on queue wait time, not just failed jobs — waiting work is the early signal.