At a glance
Statement Latency p50 (ms) is the median service latency for SQL statements across the cluster: half of all statements completed faster than this number, half slower. It is the “typical statement” gauge. Unlike the tail percentiles, p50 only moves when something affects the broad population of statements, so a rising p50 signals a cluster-wide problem (capacity, an undersized node set, a slow shared resource) rather than a contention hotspot. Read it as the baseline against which the p95 and p99 tails are judged.
What it tracks
This card reports the Statement Latency p50 (ms) for the selected period, computed from CockroachDB’ssql.service.latency metric at the 50th-percentile quantile over a rolling 5-minute window (RT/5m). Service latency covers the time CockroachDB spent planning and executing a statement inside the gateway node, from request receipt to result ready, and excludes the client network round-trip.
Vortex IQ reads the cluster-wide merged quantile rather than averaging per-node figures, and converts the underlying nanosecond histogram to milliseconds. There is no alert threshold on this card (alert: -): p50 is a context metric. Its diagnostic value comes from comparison. A flat p50 alongside a rising Statement Latency p95 (ms) or Statement Latency p99 (ms) points at a tail-only contention problem; a p50 that climbs in step with the tail points at a genuine cluster-wide slowdown. Because it is a percentile it is not additive, so you cannot average p50 across windows or nodes.
Reconcile against DB Console → Metrics → SQL dashboard → “Service Latency: SQL, 50th percentile” (or the equivalent under Monitoring → SQL in CockroachDB Cloud), matching the chart’s time range to the Vortex IQ window. For raw figures, query crdb_internal.node_statement_statistics or scrape sql_service_latency from a node’s _status/vars endpoint.