Skip to content
BoilStream

Monitor a deployment

operations
TL;DR

Scrape the Prometheus metrics endpoint from a trusted monitoring network and watch errors, backpressure, memory and request latency. Pair metrics with application result checks: a byte or request counter does not prove durability or visibility. Counters reset after a process restart, so use rates for trends.

Message sequenceMetrics expose observations; operators decide the response
Metrics expose observations; operators decide the responsePrometheus scrapes the BoilStream metrics endpoint, stores samples and evaluates alert rules. An operator investigates an alert using health, logs and application checks.BoilStreamPrometheusOperatorScrape metrics endpointCounters, gauges, histogramsEvaluate rates and alert rulesAlert with observed signalInvestigate health and workloadTime moves downward · dashed arrows are responsesMetrics expose observations; operators decide the responsePrometheus scrapes the BoilStream metrics endpoint, stores samples and evaluates alert rules. An operator investigates an alert using health, logs and application checks.BoilStreamPrometheusOperatorScrape metrics endpointCounters, gauges,histogramsEvaluate rates and alertrulesAlert with observedsignalInvestigate health andworkloadTime moves downward · dashed arrows are responses
A metric is evidence about one subsystem. Correlate it with the affected client path before taking recovery action.

BoilStream exposes service metrics in Prometheus text format:

HTTP
GET http://HOST:8081/metrics

The port is configurable. Keep the metrics endpoint on a trusted monitoring network; it is separate from the authenticated dashboard.

Configure collection

Section titled “Configure collection”

Merge this excerpt into the generated server configuration:

YAML
metrics:
  port: 8081
  flush_interval_ms: 1000

The corresponding environment variables are METRICS_PORT and METRICS_FLUSH_INTERVAL_MS. To add deployment labels such as environment or region, configure them in your monitoring system.

Metric names and types

Section titled “Metric names and types”

Use the metric reference for exact names. Check the deployed endpoint’s HELP and TYPE output before writing alerts.

Example queries

Section titled “Example queries”

The following PromQL expressions use the metrics above:

PromQL
# Kafka requests per second
rate(kafka_requests_total[5m])


# Storage errors per second
rate(storage_errors_total[5m])


# 95th percentile Flight request duration, in seconds
histogram_quantile(0.95, sum by (le) (rate(flight_request_duration_seconds_bucket[5m])))


# Backpressure above 80 percent
window_backpressure > 80

Choose alert thresholds and scrape intervals using your workload. Check sustained errors, backpressure and memory growth alongside application-side latency and result checks.

Cluster health

Section titled “Cluster health”

Operators can also inspect cluster status through the administration CLI:

Shell
boilstream-admin cluster status
boilstream-admin cp-metadata status

Investigate unhealthy metadata status or prolonged loss of a leader. See Cluster mode for quorum and recovery requirements.