Allow providing an error rate query rather than a error count query

[LukeDAtkinson]

We are scraping AWS Cloudwatch metrics from Cloudwatch into Prometheus. We want to define SLOs using these metrics.

AWS Cloudwatch reports an error rate as a proportion of total requests. It is possible to calculate the total number of errors by multiplying this by the total number of requests (i.e.

aws_cloudfront_5xx_error_rate_average * increase(aws_cloudfront_requests_sum_count[5m])

). However, when we try to use such a multiplication expression of two metrics in the error field of the ratio indicator, it causes errors in Pyrra. It seems Pyrra expects to be able to parse a single metric expression from this field.

Would it be possible to either add an errorRate field to the ratio indicator or provide a different type of indicator, and implement the necessary calculation of the total errors in Pyrra to handle this case?

[pharaujo]

Hey @LukeDAtkinson, I'm not involved with the project, but my first thought when I read your issue was that creating a recording rule with that expression and then using the resulting series would possibly address your problem? You'd still have to either backfill the calculation to have an accurate SLO, or just wait for the data to be available.