AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |
Back to Blog
Prometheus cloudwatch exporter example2/18/2024 ![]() Where by default we use the existing behavior, and missing data is treated as missing. # treat_missing_data_as not specified, defaults to treat-as-missing I propose we add an addition per-metric configuration option to MetricRule that may trigger post-processing of each Datapoint in the scrape. For example, an ELB HTTPCode_Backend_5XX alarm would treat missing data as 'notBreaching', when 5xx errors occur infrequently.įor the prometheus/cloudwatch_exporter to fully support the CloudWatch Metrics API, it also needs a mechanism for allowing users to decide how to handle missing data. zero.ĬloudWatch Alarms is one example where users are intended to interpret missing data. ELB HTTPCode_Backend_*) missing data means no occurrences, i.e. In production, 4xx's are often missing, so my SLIs are missing as well.ĬloudWatch expects users to interpret missing data on a per-metric basis, based on "Reporting criteria". When Prometheus calculates (2xx+4xx)/total, the result is often either over 100% or missing data. The cloudwatch_exporter reports incorrect values (either missing or equivalent to previous gauge value), which then interferes with Prometheus operator math. ![]() This will allow me to treat missing Sum(HTTPCode_Backend_4XX) data as 0, which then allows me to perform Prometheus operations on that data.Ĭurrently, I am unable to calculate my availability SLI using CloudWatch ELB metrics. Tl dr - I'd like to add a feature that allows users to user metric config options specify how to treat missing data from CloudWatch.
0 Comments
Read More
Leave a Reply. |