My problem with all these different methods of rate-limiting is that there is no One True Way, even within a single system.
You may want to only get a single message for one kind of error condition, but for some others you may want periodic messages, and for some you might even want to have the messages get escalated to some other person or use some other method of notification.
What we really need is a completely separate, simple, configurable notification system, with an API, web interface and command-line tools. If done properly, any other system that needed notification-sending could use it.
Unfortunately, to be really effective, it needs to know the current status of "things", which is getting into the realm of monitoring. I don't see the various monitoring systems changing to use this hypothetical notification system, so the only way I can see to make this work would be to include adapters for various monitoring systems which would continuously import state. Then the notifications could be triggered off it. That would be a lot to do, hoping that it would be used.
