Class StatefulMetric<D extends DataPoint, T extends D>

java.lang.Object
io.prometheus.metrics.core.metrics.Metric
io.prometheus.metrics.core.metrics.MetricWithFixedMetadata
io.prometheus.metrics.core.metrics.StatefulMetric<D,T>
All Implemented Interfaces:
Collector
Direct Known Subclasses:
Counter, Gauge, Histogram, StateSet, Summary
public abstract class StatefulMetric<D extends DataPoint, T extends D> extends MetricWithFixedMetadata
There are two kinds of metrics:
  • A StatefulMetric actively maintains its current values, e.g. a stateful counter actively stores its current count.
  • A CallbackMetric gets its values on demand when it is collected, e.g. a callback gauge representing the current heap size.
The OpenTelemetry terminology for stateful is synchronous and the OpenTelemetry terminology for callback is asynchronous. We are using our own terminology here because in Java synchronous and asynchronous usually refers to multi-threading, but this has nothing to do with multi-threading.

  • Field Details

    • data

      private final ConcurrentHashMap<List<String>, T extends D> data
      Map label values to data points.

    • noLabels

      private volatile T extends D noLabels
      Shortcut for data.get(Collections.emptyList())

  • Constructor Details

  • Method Details

    • collect

      protected abstract MetricSnapshot collect(List<Labels> labels, List<T> metricData)
      labels and metricData have the same size. labels.get(i) are the labels for metricData.get(i).

    • collect

      public MetricSnapshot collect()
      Description copied from interface: Collector
      Called when the Prometheus server scrapes metrics.
      Specified by:
      collect in interface Collector
      Specified by:
      collect in class Metric

    • initLabelValues

      public void initLabelValues(String... labelValues)
      Initialize label values.

      Example: Imagine you have a counter for payments as follows 

      payment_transactions_total{payment_type="credit card"} 7.0
payment_transactions_total{payment_type="paypal"} 3.0
      Now, the data points for the payment_type label values get initialized when they are first used, i.e. the first time you call 
      counter.labelValues("paypal").inc();
      the data point with label payment_type="paypal" will go from non-existent to having value 1.0.

      In some cases this is confusing, and you want to have data points initialized on application start with an initial value of 0.0: 

      payment_transactions_total{payment_type="credit card"} 0.0
payment_transactions_total{payment_type="paypal"} 0.0
      initLabelValues(...) can be used to initialize label value, so that the data points show up in the exposition format with an initial value of zero.

    • labelValues

      public D labelValues(String... labelValues)

    • remove

      public void remove(String... labelValues)
      Remove the data point with the given label values. See https://prometheus.io/docs/instrumenting/writing_clientlibs/#labels.

    • removeIf

      public void removeIf(Function<List<String>, Boolean> f)
      Remove the data points when the given function.

    • clear

      public void clear()
      Reset the metric (remove all data points).

    • newDataPoint

      protected abstract T newDataPoint()

    • getNoLabels

      protected T getNoLabels()

    • getMetricProperties

      protected MetricsProperties[] getMetricProperties(StatefulMetric.Builder<?,?> builder, PrometheusProperties prometheusProperties)
      Metric properties in effect by order of precedence with the highest precedence first. If a MetricProperties is configured for the metric name it has higher precedence than the builder configuration. A special case is the setting StatefulMetric.Builder.withoutExemplars() via the builder, which cannot be overridden by any configuration.

    • getConfigProperty

      protected <P> P getConfigProperty(MetricsProperties[] properties, Function<MetricsProperties, P> getter)