Remaining error budget for rolling 30 days seems incorrect

[alexandrospanayi]

Hey everyone.

I've noticed something off with the rolling 30 day vs. start of the month remaining error budget metric. For example, today is August 25th. The start of the month is 1st of August, and rolling 30 days is around July 27th.

This doesn't seem to make much sense to me. The cut off between the start of the month and rolling 30 day is around 3-4 days. There is simply no way the difference between the rolling 30 day and start of the month accoutns for -371% burned error budget.

As you can see from the SLI, the patterns of going below our objective is predictable.

I feel like either I'm not fully understanding the logic behind this or there's a bug in how the Prometheus Rule is recording the metric.

[alexandrospanayi]

Just another example here.

I mean this definitely makes no sense right? How can the rolling 30 day budget be 100% remaining if there's dips in the SLI that the start of the month error budget is accounting for?