Stumbled across this and had a good laugh. I used to support a large distributed zabbix environment where each zabbix node was back-ended by a mysql database. Constant pain. Everything from general performance problems (ui is slow again? better run another optimize job) to various application problems (hey, replication is failing and the app logs are hemorrhaging various query failures, sweet!).
After switching all of our performance data capture to fixed-size databases (first rrd via pnp4nagios, later graphite back-ended by whisper), IMHO there is no reason to ever suffer storing performance data (or really any other high-turnover, high-write-rate datasets) in a relational database ever again. Constant growth means eternal maintenance work (ref. the optimize jab earlier =D), and working with fixed-size databases end all of that pain.