I'm playing with alerts and I'm running into a limitation
I want to set an alert for when a metric is zero two hours in a row, but the closest thing I can do is set an alert on when the 12h moving average is zero, which is a big hit in terms of latency. I feel like latency is essential for alarms, so there should be more options in terms of frequency, especially at the lower end of the spectrum (right now it's 1 hour or 1 day, which is a factor of 24x)
For context, I want to catch any issues with our sign-up process; however, our user volume is very variable, and every once in a while, in the middle of the night, our sign-up rate goes to zero for an hour. The only time it went to zero for two consecutive hours is when we had a bug, a bug which we caught through other means within 6 hours or so; So a 12 hour latency in the alert would be useless in this case, but a alert that wakes people up every other day in the middle of the night might be even worse...