I came up with the same workaround myself; then switched to influxdb, and my monitoring server's io-wait is still at 0% even with twice as many metrics coming in :)