Metrics

The OSO Kafka Backup Operator exposes controller and completed-operation metrics on port 8080.

Endpoint	Description
/metrics	Prometheus metrics
/healthz or /health	Liveness response
/readyz or /ready	Readiness response

Operator and CLI metrics are different

The operator runs kafka-backup-core in its own process. Its /metrics endpoint exposes the kafka_backup_operator_* families listed below; it does not expose the CLI's live kafka_backup_lag_records, throughput, or snapshot progress registry. Use the Strimzi operator when you need separately scrapeable backup/restore Job pods.

Enable scraping

Metrics are enabled by default. With the Prometheus Operator installed, enable the chart's ServiceMonitor:

values.yaml

metrics:
  enabled: true
  serviceMonitor:
    enabled: true
    interval: 30s
    scrapeTimeout: 10s
    labels:
      release: prometheus

helm upgrade --install kafka-backup-operator oso/kafka-backup-operator \
  --namespace kafka-backup \
  --create-namespace \
  --version 1.1.1 \
  --values values.yaml

The chart also adds Prometheus scrape annotations to the operator pod by default. To inspect the endpoint directly:

kubectl port-forward -n kafka-backup \
  svc/kafka-backup-operator-metrics 8080:8080
curl --fail http://127.0.0.1:8080/metrics

Metrics reference

Metric	Type	Labels	Description
kafka_backup_operator_reconciliations_total	Counter	kind	Reconciliation attempts
kafka_backup_operator_reconciliation_errors_total	Counter	kind	Reconciliation errors
kafka_backup_operator_reconcile_duration_seconds	Histogram	kind	Reconciliation duration
kafka_backup_operator_backups_total	Counter	outcome, namespace, name	Completed backups by outcome
kafka_backup_operator_backup_size_bytes	Gauge	namespace, name	Last backup size
kafka_backup_operator_backup_records_total	Gauge	namespace, name	Records processed in the last backup
kafka_backup_operator_restores_total	Counter	outcome, namespace, name	Completed restores by outcome
kafka_backup_operator_offset_resets_total	Counter	outcome, namespace	Offset resets by outcome
kafka_backup_operator_offset_reset_duration_seconds	Histogram	namespace	Offset reset duration
kafka_backup_operator_validations_total	Counter	outcome, namespace, name	Validations by outcome
kafka_backup_operator_cleanups_total	Counter	kind	Finalizer cleanup operations
kafka_backup_operator_health	Gauge	none	Operator health (1 is healthy)

kafka_backup_operator_backup_records_total is a gauge despite its historical name: it represents the most recently completed backup, not a monotonically increasing counter.

Example queries

Reconciliation error rate:

sum by (kind) (
  rate(kafka_backup_operator_reconciliation_errors_total[5m])
)

Failed backups during the last hour:

sum by (namespace, name) (
  increase(kafka_backup_operator_backups_total{outcome="failure"}[1h])
)

99th-percentile reconciliation duration:

histogram_quantile(
  0.99,
  sum by (le, kind) (
    rate(kafka_backup_operator_reconcile_duration_seconds_bucket[5m])
  )
)

Operator unavailable:

absent(kafka_backup_operator_health) or kafka_backup_operator_health != 1

Example alerts

operator-alerts.yaml

groups:
  - name: kafka-backup-operator
    rules:
      - alert: KafkaBackupOperatorDown
        expr: absent(kafka_backup_operator_health) or kafka_backup_operator_health != 1
        for: 5m
        labels:
          severity: critical
        annotations:
          summary: Kafka Backup Operator is unavailable

      - alert: KafkaBackupOperatorReconcileErrors
        expr: increase(kafka_backup_operator_reconciliation_errors_total[10m]) > 0
        for: 5m
        labels:
          severity: warning
        annotations:
          summary: Kafka Backup Operator reconciliation is failing

      - alert: KafkaBackupFailed
        expr: increase(kafka_backup_operator_backups_total{outcome="failure"}[1h]) > 0
        labels:
          severity: critical
        annotations:
          summary: Kafka backup failed

Resource-level spec.metrics

KafkaBackup.spec.metrics is retained in the CRD and passed into the embedded core configuration. It does not create a second HTTP listener or add live core progress families to the operator endpoint. Monitor completion through CR status and the operator metric families above.