Skip to main content

Metrics

The manager registers these series through Micrometer. They are emitted to whichever sink is active (QOD_METRICS_SINK); under the default Prometheus sink they appear at GET :20900/metrics. For how to scrape, push to a cloud monitor, or import the Grafana dashboard, see Observability.

Application metrics

MetricTypeLabelsMeaning
statements_totalcountertenant, pool, statusStatements executed, partitioned by outcome status. Drives QPS and error-rate panels.
statement_duration_secondshistogramtenant, poolStatement execution latency; the source of the p50/p95/p99 percentiles.
flightsql_sessions_activegauge(none)Currently open FlightSQL sessions.
pool_nodesgaugetenant, pool, roleNode count per pool, broken down by node role.
node_healthygaugetenant, pool, node_id, role1 when the node is healthy, 0 otherwise.
node_draininggaugetenant, pool, node_id, role1 when the node is draining in-flight work before shutdown.
node_in_flightgaugetenant, pool, node_id, roleStatements currently executing on the node.
node_ewma_latency_secondsgaugetenant, pool, node_id, roleEWMA of completed-statement latency, the signal the router uses to pick the least-loaded node.

JVM and process metrics

Registered by the Micrometer JVM and process binders:

MetricTypeMeaning
jvm_memory_used_bytesgaugeHeap and non-heap memory in use.
jvm_gc_pause_seconds_sumcounterCumulative GC pause time.
jvm_threads_live_threadsgaugeLive thread count.
process_uptime_secondsgaugeProcess uptime since manager start.

Common labels

Every series can carry static deployment labels when these are set, useful for separating environments in a shared Grafana:

VariableLabelExample
QOD_METRICS_DEPLOYMENTdeploymentprod-eu
QOD_METRICS_REGIONregioneu-west-1