Skip to main content
The Prometheus Exporter sink exposes metrics on an HTTP endpoint that Prometheus can scrape. It aggregates distribution metrics into histograms or summaries and maintains metric state between scrapes.

​
Configuration

[sinks.prometheus_exporter]
type = "prometheus_exporter"
inputs = ["my_metrics"]

# HTTP server address
address = "0.0.0.0:9598"

# Optional namespace prefix
default_namespace = "service"

# Flush expired metrics interval
flush_period_secs = 60

​
Core Parameters

​
address
string
default:"0.0.0.0:9598"
The address to expose for scraping. Metrics are exposed at /metrics.
address = "0.0.0.0:9598"      # Listen on all interfaces
address = "127.0.0.1:9598"    # Localhost only
address = "192.160.0.10:9598" # Specific interface
​
default_namespace
string
Default namespace prefix for metrics that don’t have one.The namespace is prepended to metric names with an underscore separator.
default_namespace = "service"  # metrics become: service_metric_name
default_namespace = "app"      # metrics become: app_metric_name
​
flush_period_secs
integer
default:"60"
Interval in seconds to flush expired metrics.Metrics not seen since the last flush are considered expired and removed. Set this higher than your Prometheus scrape interval.
flush_period_secs = 60   # Flush every minute
flush_period_secs = 300  # Flush every 5 minutes
​
suppress_timestamp
boolean
default:"false"
Suppress timestamps in Prometheus output.Useful when aggregating metrics over long periods or replaying old metrics.
suppress_timestamp = true

​
Distribution Metrics

Vector’s distribution metrics must be aggregated into Prometheus-compatible formats:

​
Histograms (Default)

​
distributions_as_summaries
boolean
default:"false"
When false (default), distributions are aggregated into histograms.
​
buckets
array
Bucket boundaries for aggregating distributions into histograms.
# Default buckets (suitable for response times in seconds)
buckets = [0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1.0, 2.5, 5.0, 10.0]

# Custom buckets for milliseconds
buckets = [1, 5, 10, 25, 50, 100, 250, 500, 1000, 2500, 5000]

# Memory usage in MB
buckets = [10, 50, 100, 250, 500, 1000, 2500, 5000]

​
Summaries

​
distributions_as_summaries
boolean
default:"false"
When true, distributions are aggregated into summaries.
​
quantiles
array
default:"[0.5, 0.75, 0.9, 0.95, 0.99]"
Quantiles to calculate for aggregating distributions into summaries.
distributions_as_summaries = true
quantiles = [0.5, 0.75, 0.9, 0.95, 0.99, 0.999]

# Use histograms (default, recommended)
[sinks.prom]
type = "prometheus_exporter"
inputs = ["metrics"]
buckets = [0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1.0, 2.5, 5.0, 10.0]

# Or use summaries
[sinks.prom_summary]
type = "prometheus_exporter"
inputs = ["metrics"]
distributions_as_summaries = true
quantiles = [0.5, 0.9, 0.95, 0.99]

​
Authentication

​
auth.username
string
HTTP Basic authentication username.
​
auth.password
string
HTTP Basic authentication password.
[sinks.prometheus.auth]
username = "prometheus"
password = "${PROM_PASSWORD}"

​
TLS Configuration

​
tls.enabled
boolean
default:"false"
Enable TLS/SSL for the HTTP endpoint.
​
tls.crt_file
string
Path to TLS certificate file.
​
tls.key_file
string
Path to TLS private key file.
[sinks.prometheus.tls]
enabled = true
crt_file = "/etc/vector/certs/server.crt"
key_file = "/etc/vector/certs/server.key"

​
Metric Types

The Prometheus exporter handles Vector’s metric types:
Vector TypePrometheus TypeDescription
CounterCounterMonotonically increasing value
GaugeGaugeValue that can increase or decrease
DistributionHistogram or SummaryStatistical distribution of values
SetGaugeCount of unique values

​
Complete Examples

​
Basic Configuration

[sinks.prometheus]
type = "prometheus_exporter"
inputs = ["host_metrics"]

address = "0.0.0.0:9598"
default_namespace = "vector"
Then configure Prometheus to scrape: 
# prometheus.yml
scrape_configs:
  - job_name: 'vector'
    static_configs:
      - targets: ['localhost:9598']

​
With Authentication

[sinks.prometheus_secure]
type = "prometheus_exporter"
inputs = ["metrics"]

address = "0.0.0.0:9598"

[sinks.prometheus_secure.auth]
username = "prometheus"
password = "${PROMETHEUS_PASSWORD}"
Prometheus configuration: 
scrape_configs:
  - job_name: 'vector'
    static_configs:
      - targets: ['vector:9598']
    basic_auth:
      username: prometheus
      password: ${PROMETHEUS_PASSWORD}

​
With TLS

[sinks.prometheus_tls]
type = "prometheus_exporter"
inputs = ["metrics"]

address = "0.0.0.0:9598"

[sinks.prometheus_tls.tls]
enabled = true
crt_file = "/etc/vector/tls/server.crt"
key_file = "/etc/vector/tls/server.key"
Prometheus configuration: 
scrape_configs:
  - job_name: 'vector'
    scheme: https
    static_configs:
      - targets: ['vector:9598']
    tls_config:
      ca_file: /etc/prometheus/ca.crt
      insecure_skip_verify: false

​
Custom Histogram Buckets

[sinks.prometheus_custom]
type = "prometheus_exporter"
inputs = ["response_times"]

address = "0.0.0.0:9598"
default_namespace = "http"

# Buckets optimized for response times in milliseconds
buckets = [1, 5, 10, 25, 50, 100, 250, 500, 1000, 2500, 5000, 10000]

​
Using Summaries

[sinks.prometheus_summary]
type = "prometheus_exporter"
inputs = ["request_durations"]

address = "0.0.0.0:9598"

# Use summaries instead of histograms
distributions_as_summaries = true
quantiles = [0.5, 0.9, 0.95, 0.99, 0.999]

​
Multiple Namespaces

# API metrics
[sinks.prom_api]
type = "prometheus_exporter"
inputs = ["api_metrics"]
address = "0.0.0.0:9598"
default_namespace = "api"

# Database metrics  
[sinks.prom_db]
type = "prometheus_exporter"
inputs = ["db_metrics"]
address = "0.0.0.0:9599"
default_namespace = "database"

​
High-Cardinality Handling

[sinks.prometheus_limited]
type = "prometheus_exporter"
inputs = ["filtered_metrics"]

address = "0.0.0.0:9598"

# Flush more frequently to limit memory usage
flush_period_secs = 30

# Use histograms to bound cardinality
buckets = [0.1, 0.5, 1.0, 5.0, 10.0]

​
Kubernetes Deployment

​
Service

apiVersion: v1
kind: Service
metadata:
  name: vector-metrics
  labels:
    app: vector
spec:
  type: ClusterIP
  ports:
    - port: 9598
      targetPort: 9598
      name: metrics
  selector:
    app: vector

​
ServiceMonitor (Prometheus Operator)

apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  name: vector
  labels:
    app: vector
spec:
  selector:
    matchLabels:
      app: vector
  endpoints:
    - port: metrics
      interval: 30s
      path: /metrics

​
PodMonitor

apiVersion: monitoring.coreos.com/v1
kind: PodMonitor
metadata:
  name: vector
spec:
  selector:
    matchLabels:
      app: vector
  podMetricsEndpoints:
    - port: metrics
      interval: 30s
      path: /metrics

​
Prometheus Queries

Example PromQL queries for Vector metrics: 
# Rate of events processed
rate(vector_events_processed_total[5m])

# P95 processing latency (histogram)
histogram_quantile(0.95, rate(vector_processing_duration_seconds_bucket[5m]))

# Memory usage by component
sum by (component) (vector_memory_bytes)

# Error rate
rate(vector_errors_total[5m])

# Events throughput by sink
sum by (sink) (rate(vector_sink_events_total[5m]))

​
Troubleshooting

​
Endpoint Not Accessible

If you can’t reach the /metrics endpoint:
  1. Verify the address binding (use 0.0.0.0 to listen on all interfaces)
  2. Check firewall rules and network policies
  3. Ensure the port is not already in use
  4. Review Vector logs for binding errors
  5. Test with curl http://localhost:9598/metrics

​
No Metrics Appearing

If /metrics is empty:
  1. Verify metrics are flowing to the sink (check inputs)
  2. Check metric types are compatible
  3. Ensure metrics have been generated since Vector started
  4. Review Vector logs for errors
  5. Wait for the first scrape interval

​
High Memory Usage

For memory issues:
  1. Reduce flush_period_secs to expire metrics faster
  2. Limit metric cardinality (reduce label combinations)
  3. Use histograms instead of summaries
  4. Filter high-cardinality labels before the exporter
  5. Reduce buckets or quantiles array size

​
Metrics Disappearing

If metrics vanish between scrapes:
  1. Increase flush_period_secs (must be > scrape interval)
  2. Ensure Prometheus scrape interval aligns with flush period
  3. Check if metrics are being continuously generated
  4. Review Vector logs for errors

​
Best Practices

  1. Set flush period > scrape interval to prevent metric expiration
  2. Use histograms over summaries for better aggregation
  3. Choose appropriate buckets based on your metric ranges
  4. Add a default namespace to avoid name collisions
  5. Enable authentication in production environments
  6. Use TLS for sensitive metrics
  7. Limit cardinality to prevent memory issues
  8. Monitor the exporter with Prometheus queries
  9. Use ServiceMonitor in Kubernetes with Prometheus Operator
  10. Document custom metrics for your team

​
Metric Cardinality

High cardinality can cause memory issues. Tips to manage:
  1. Limit label values: Avoid unbounded labels (user IDs, UUIDs)
  2. Aggregate before export: Use transforms to group metrics
  3. Drop high-cardinality labels: Remove unnecessary labels
  4. Use histograms: Bound distribution metrics
  5. Increase flush frequency: Expire metrics sooner
  6. Sample if needed: Reduce metric volume
# Drop high-cardinality label
[transforms.reduce_cardinality]
type = "remap"
inputs = ["metrics"]
source = '''
  del(.tags.user_id)  # Remove unbounded label
  .tags.endpoint = replace(.tags.endpoint, r'/\d+/', "/id/")  # Normalize IDs
'''

[sinks.prometheus]
type = "prometheus_exporter"
inputs = ["reduce_cardinality"]
address = "0.0.0.0:9598"

​
Performance Considerations

  1. Scrape interval: Balance freshness with overhead (15-60s typical)
  2. Flush period: Set to 2-3x scrape interval
  3. Histogram buckets: More buckets = more memory, choose wisely
  4. Summaries: More expensive than histograms, use sparingly
  5. Authentication: Minimal overhead, enable freely
  6. TLS: Small overhead, worth it for security

​
See Also

Build docs developers (and LLMs) love

Get started for freeTalk to us