Metrics from the correct point of view

It’s important to have the correct context in place when monitoring, otherwise your metrics can all report back that everything is totally fine when in reality, it isn’t. This is one reason why I think amazon sees more success chasing after imporoving the huge outliers rather than trying to improve the median. After all, if a customer has zero uptime, what does it matter that the whole system is at 99.999%?