Skip to main content
This page documents the chart’s Prometheus metrics surface.

​
Metrics endpoints

ComponentPortPort namePathWhat it exposes
Engine pods9090metrics/metricsFirebolt engine gauges, including firebolt_running_queries and firebolt_suspended_queries.
Gateway pods (Envoy)9090 (configurable)metrics/stats/prometheusEnvoy connection, request, and cluster stats, proxied from the admin interface by a read-only stats listener.
Metadata Servicen/an/an/aThe Metadata Service does not currently expose a Prometheus endpoint.
The gateway metrics port defaults to 9090 and is configurable through gateway.metricsPort in values.yaml.

​
Scrape with the Prometheus Operator

The chart ships two optional PodMonitor resources gated on values.yaml. Both are disabled by default. The Prometheus Operator CRDs must already be installed in the cluster, otherwise helm install fails with an unknown-kind error. Enable either or both: 
# my-values.yaml
podMonitor:
  engines:
    enabled: true
    interval: 15s
  gateway:
    enabled: true
    interval: 15s
Apply with helm upgrade (or helm install for a new release): 
# Roll the change out to an existing release.
helm upgrade firebolt ./helm -n firebolt -f my-values.yaml
After Prometheus picks up the PodMonitors, the engine pods appear as targets at :9090/metrics and the gateway pods at :9090/stats/prometheus.

​
Selectors and labels

Each PodMonitor selects pods by the chart’s firebolt/component label.
The Firebolt Kubernetes Operator stamps its workloads with a different label namespace (firebolt.io/...). Selectors written for that convention do not match chart-deployed pods.
PodMonitorSelectorPod target labels propagated to series
<release>-enginesfirebolt/component: engineapp.kubernetes.io/name, firebolt/engine
<release>-gatewayfirebolt/component: gatewayapp.kubernetes.io/name
The firebolt/engine label carries the engine name from engines[].name, so engine series are partitionable per engine in PromQL: 
sum by (firebolt_engine) (firebolt_running_queries)

​
Per-engine PodMonitors

The chart-level engine PodMonitor applies the same scrape configuration to every engine in the release. If you need per-engine control (different intervals, selective enablement, custom relabelings), disable the chart-level one and deploy your own keyed on the engine name: 
podMonitor:
  engines:
    enabled: false  # disable the chart's blanket PodMonitor
Then create a per-engine PodMonitor: 
apiVersion: monitoring.coreos.com/v1
kind: PodMonitor
metadata:
  name: firebolt-engine-analytics
  namespace: firebolt
spec:
  selector:
    matchLabels:
      firebolt/component: engine
      firebolt/engine: analytics
  podMetricsEndpoints:
    - port: metrics
      path: /metrics
      interval: 30s

​
Gateway stats listener

The Envoy admin interface is bound to 127.0.0.1:9901 inside the gateway pod and is not reachable from the cluster. Binding it elsewhere would expose mutation endpoints (POST /healthcheck/fail, POST /quitquitquit). Instead, the chart renders a dedicated read-only stats_listener on gateway.metricsPort (default 9090) that proxies only /stats/prometheus from the admin interface through an internal static cluster. The PodMonitor for the gateway scrapes that listener. To pick a different metrics port (for example to avoid a conflict with another sidecar), set: 
gateway:
  metricsPort: 9095
The container port name stays metrics, so the PodMonitor does not need to change.

​
Cross-namespace scraping

The chart’s PodMonitor resources are namespace-scoped and select only pods in the release namespace. To scrape engines and gateways in other namespaces, deploy a hand-rolled PodMonitor with a broader namespaceSelector: 
spec:
  namespaceSelector:
    any: true
  selector:
    matchLabels:
      firebolt/component: engine

​
What this chart does not expose

  • Controller and CR-status metrics. The chart has no controller loop, so there are no per-CR status gauges.
  • Metadata Service metrics. The Metadata Service runs as a gRPC server only. It exposes no /metrics endpoint.
  • PostgreSQL metrics. The bundled PostgreSQL StatefulSet does not include postgres_exporter. Run your own exporter sidecar or scrape an external PostgreSQL through its existing observability stack.
The Firebolt Kubernetes Operator exposes continuous-reconciliation and per-CR status metrics (engine phase, replica counts, drain check errors, last-reconciled timestamps) that this chart cannot provide. If that observability surface matters to you, see the operator upgrade path.