Metric cardinality

Metric cardinality is the number of unique time series produced by a metric, driven by the combination of every distinct value across its labels or tags; a request-count metric labeled by endpoint, status code, and user ID can generate far more unique series than the same metric labeled only by endpoint, since each additional label multiplies the possible combinations.

High cardinality differs from simply having "a lot of data" in that it specifically stresses the indexing layer of a time-series database, since each unique label combination must be tracked as its own series regardless of how few data points it has; a single high-cardinality label, like a raw user ID or request ID, can turn one metric into millions of series. This is distinct from log volume, which log aggregation systems handle differently, through text indexing rather than per-series storage.

Cardinality matters because uncontrolled growth in unique label combinations, often called a cardinality explosion, can slow queries or crash a monitoring backend entirely, and many observability platforms price on cardinality directly. The common pitfall is adding a high-uniqueness field, such as an unbounded user or session ID, as a metric label instead of only in logs or traces, where high-cardinality data belongs.