Prometheus concepts in Cloud Observability

Cloud Observability accepts Prometheus Counters, Gauges, and Histograms through the OpenTelemetry Collector. However, there may be differences in how these metric instruments and measurements are displayed in Cloud Observability as opposed to Prometheus. This document collects these differences for reference.

Prometheus cumulative counters

Cloud Observability accepts a Counter, but for technical and philosophical reasons, we do not provide the underlying cumulative count to our graphs and instead plot them as deltas. The resolution of these deltas varies based on the age of the data - for example, 45 second buckets for data recorded in the past hour, 2 minute buckets for data recorded between one and four hours ago, and so forth.

For an example of how this might manifest, consider the promhttp_metric_handler_requests_total counter that Prometheus exposes for itself (at the /metrics endpoint). Assuming a regular scape cadence of 1 scrape per 10 seconds, graphing this metric in Prometheus would result in a plot of values increasing steadily, up and to the right. In Cloud Observability, the same metric represented as a count would result in a straight line over time assuming a consistent rate of scrape requests.

Default grouping for rate calculations

When taking the rate of a counter in Prometheus, the default plot will implicitly group the plot by attributes (tabs/labels). In Cloud Observability, you must explicitly specify the attribute to group by.

See also

Create and manage dashboards

Create and manage panels

Create alerts

Updated May 25, 2021