Metric types

Each ingested and aggregated metric has a metric type, a critical piece of metadata that indicates what kind of data a metric represents. Metric types affect how the data aggregates for storage, long-term downsampling behaves, and how that data displays within the Chronosphere Observability Platform.

If a metric is ingested or aggregated with the wrong type, you can get unexpected results when aggregating or querying the data, or data loss when persisting data in long term storage.

Observability Platform types

Observability Platform supports ingesting many different metric types.

Cumulative counter

A counter that stores a strictly increasing count of distinct events or a reset-to-zero on restart. Counters are a fundamental metric construct that keeps track of the number of times a certain event has occurred.

• Each datapoint stores the running total at the given timestamp.
• Aggregated using metric_type=CUMULATIVE_COUNTER rollup rules
• Queryable as a Prometheus counter (opens in a new tab).
• Subject to long-term downsampling.
• View incoming cumulative counters with a __metric_type__:cumulative_counter filter in the Live Telemetry Analyzer.

Examples of cumulative counters include:

• Number of requests served
• Number of user logins
• Number of heap allocations

Delta counter

A counter where each data point stores a strictly positive delta increment at the given timestamp. With this method, the client only keeps track of how many events since the last emission or flush of the metric value and only emit that value. It doesn't keep a running sum.

• Stores the same kinds of data as cumulative counters, but with a different format.
• Aggregated with metric_type=DELTA_COUNTER rollup rules. Raw delta counter data is automatically aggregated unless explicitly dropped by a rollup rule.
• Queryable as a Prometheus counter (opens in a new tab).
• View incoming delta counters with a __metric_type__:delta_counter filter in the Live Telemetry Analyzer.

An example of a delta counter is:

• A client sends a 1 every time a request is served to count the number of requests served.

Gauge

A gauge is a basic metric type that tracks a value which changes over time.

• Each datapoint stores the value at the given timestamp. It's impossible to have two values at the same point in time.
• Aggregated with metric_type=GAUGE rollup rules.
• Queryable as a Prometheus gauge (opens in a new tab).
• Subject to long-term downsampling.
• View incoming gauges with a __metric_type__:gauge filter in the Live Telemetry Analyzer.

Some examples of gauges are:

• Current number of in-flight requests.
• Current number of signed-in users.
• Current amount of memory in use.

Histogram

The frequency of values in a set of measurements that fall within certain ranges called buckets. A histogram uses several buckets defined in a bucket layout. Observability Platform processes and persists each histogram as a single structured value and time series.

Observability Platform supports OpenTelemetry exponential histograms and Prometheus native histograms, and also supports both cumulative and delta temporality. The histogram's bucket layout defines the histogram metric type in Observability Platform.

You can query histograms using PromQL histogram functions (opens in a new tab). They're subject to the same long-term downsampling as other metric types.

Some examples of histograms include:

• Distributions of request latencies
• Distributions of payload sizes

Histograms are limited in these ways:

• The maximum bucket limit is 160 buckets.
• The maximum exponential histogram scale is 8, and the minimum scale is -4.

Cumulative exponential histograms

The cumulative exponential histogram type is compatible with both OpenTelemetry cumulative exponential histograms and legacy Prometheus native histograms with exponential bucket layouts.

• Observability Platform automatically reduces the exponential histogram scale when the histogram exceeds the 160-bucket limit. When downscaling, it doubles the exponential histogram bucket width to summarize the observations with fewer but wider buckets, effectively reducing the exponential histogram's resolution.

  When aggregating or querying exponential histograms with different bucket scales, Observability Platform downscales them to the lowest common histogram scale. Zero-threshold handling similarly widens scales until values fit.

• Observability Platform aggregates them with metric_type=cumulative_exponential_histogram rollup rules.

• You can view incoming measurements by using a __metric_type__:cumulative_exponential_histogram filter in the Live Telemetry Analyzer.

Delta exponential histograms

The delta exponential histogram type is compatible only with OpenTelemetry delta exponential histograms.

• Observability Platform aggregates them with metric_type=delta_exponential_histogram rollup rules.
• You can view incoming measurements by using a __metric_type__:delta_exponential_histogram filter in the Live Telemetry Analyzer.

Measurement

A raw observation with some discrete value. Measurements differ from counters as each measurement event has a discrete, arbitrary value. Measurements also differ from a gauge, as there can be multiple values at any point in time.

• You must use a metric_type=MEASUREMENT rollup rule with drop_raw=true to aggregate measurements into a persistable metric type. Raw measurement data can't be persisted.
• The MEASUREMENT metric type is best used when all values should be aggregated at once across many matching time series.
• View incoming measurements with a __metric_type__:measurement filter in the Live Telemetry Analyzer.

Some examples of measurements are:

• The latency of an individual request
• The payload size of an individual request

Supported formats

Observability Platform can ingest metrics from many different external metric clients. Each external metric client has their own concept of "metric type" which is mapped into a Observability Platform metric type.

Prometheus

OpenTelemetry

Observability Platform doesn't support delta non-monotonic sums (also known as delta up-down counters).

Datadog

SignalFX

Wavefront

StatsD

GCP Metrics