High Cardinality in Observability: Causes, Costs & Fixes
Blog post from OpenObserve
High cardinality occurs when telemetry fields such as user IDs, request IDs, trace IDs, or pod names have many distinct values; in metrics, each unique label combination creates a separate time series, potentially increasing memory use, slowing queries, causing outages, and raising costs when vendors charge per series. These fields remain valuable for identifying the customer, request, or infrastructure behind an incident, so the article recommends keeping unbounded identifiers off metric labels and retaining them in logs and traces instead. It describes measuring cardinality with Prometheus queries or SQL, and controlling growth by using bounded metric labels, templating raw URLs, filtering attributes through the OpenTelemetry Collector, aggregating or downsampling metrics, and monitoring series growth. It distinguishes cardinality—the number of unique values in a field—from dimensionality—the number of fields in an event—and presents OpenObserve’s approach of storing telemetry in columnar Parquet on object storage, using bloom filters for high-cardinality lookups, and charging by ingestion volume rather than series count. The article also cites performance and storage-cost comparisons for OpenObserve, while noting that results can vary by workload.
| Trend | Post Mentions | Total Month Mentions | Posts | Companies | MoM |
|---|---|---|---|---|---|
| Observability | 20 | 2,390 | 523 | 156 | -23% |
| OpenTelemetry | 12 | 822 | 142 | 56 | -4% |
| Kubernetes | 5 | 4,377 | 419 | 114 | -15% |
| LLM | 1 | 3,674 | 823 | 199 | -44% |
| MCP | 1 | 9,757 | 959 | 203 | -18% |
| Real-time | 1 | 3,194 | 698 | 199 | -24% |
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.