All sites have returned to normal production after restarting the problematic RabbitMQ nodes. We're continuing to monitor.

We have determined that a stored value became corrupted, resulting in delayed processing that caused the communication issue. We have fixed the corrupted value, and production has returned to normal. We are continuing to investigate what caused the initial data corruption, and will take steps to improve our ability to identify issues that might arise in the future.