At Sunday 5:58pm EST, sysadmins received an alert that there was a high load in the message worker. The system was killing processes due to low memory and our monitoring system restarting them automatically. Sysadmins stopped all workers for a few minutes and restarted the machine.
At 6:16pm the system was restored and all messages were sent. We are investigating what has caused the high load.
No message was lost, 53 messages were delayed during the 18min period.