working-with-seq-metrics

First search metric definitions via the @Definitions object, then list dimensions with seqcli metrics dimensions, and finally run aggregate queries against the series data source. Always include where Has(<metric name>) to avoid row count skew.

Search the @Definitions object on metric samples, for example Keys(@Definitions)[?] like '%cpu%' ci, or filter by service with @Resource.service.name. Never query the series data source directly for metric definitions.

Use the dimension values MCP tool or the seqcli metrics dimensionvalues CLI command. Do not use select distinct on the series data source, as the dedicated tools are optimized for this purpose.

Use the phist function with lateral unnest over the metric's buckets column, grouping by time intervals. For fixed histograms compute bucket midpoints with coalesce; exponential histograms expose bucket.midpoint directly.

Counters appearing as gauges indicate misconfigured aggregation temporality in the OpenTelemetry exporter. This can produce wildly inaccurate aggregation results and must be fixed at the instrumentation source.

Group keys are automatically included in result rowsets and must not appear in the select list. To order by a group key, alias it with group by <expr> as <alias> and use order by <alias>.