It's an irregular problem caused by the script that rotates the webserver logs. I tracked the problem this weekend. We'll have a fix in place before next weekend. (A short period of downtime for log rotation is unavoidable, but this script problem causes a longer downtime -- sometimes.)