Depending on the computing resources being assigned to a Prometheus instance (that is, CPU, memory, disk space, and IOPS), it will gracefully handle a number of time series. This number can be thought of as the primary indicator of capacity for that instance, and it will inform your scraping decisions: will you have thousands of targets with relatively few metrics, fewer targets with a thousand metrics each, or something in between? In the end, Prometheus will only be able to handle that amount of time series without performance degradation.

It is in this context that the concept of cardinality appears. This term is often used to mean the number of unique time series that are produced by a combination of metric names and their ...