How does OpenMetrics differ from the Prometheus text format?

OpenMetrics is a standardized superset based on the Prometheus text format. Key additions are the mandatory # EOF terminator, a # UNIT metadata line, the _total suffix requirement for counters, and native support for exemplars and the Info and StateSet metric types.

What is the # EOF line for?

OpenMetrics requires every exposition to end with a line containing exactly # EOF. This lets a parser know it received the complete payload and was not truncated mid-stream, which the older Prometheus format could not guarantee.

What metadata lines can a metric have?

Each metric family can declare # HELP for a description, # TYPE for the metric type such as counter or gauge, and # UNIT for the base unit. The UNIT must match the suffix on the metric name, for example a unit of seconds with a metric ending in _seconds.

What is an exemplar in OpenMetrics?

An exemplar links a sample to a trace by appending a hash followed by labels and an optional value and timestamp, for example after a histogram bucket value. It lets observability tools jump from a metric data point straight to a representative trace.

What suffixes does a counter use in OpenMetrics?

An OpenMetrics counter named x must expose its value as x_total. It may also expose an optional x_created line giving the Unix timestamp when the counter was first initialized, which helps detect resets.

OpenMetrics Exposition Format

OpenMetrics exposition format reference

OpenMetrics is the IETF-track standard for exposing metrics as text, evolved from the Prometheus exposition format. A scrape endpoint returns a plain-text body of metadata lines (# HELP, # TYPE, # UNIT), sample lines, and a mandatory # EOF terminator. This reference lists every keyword and syntax element so you can read or generate a valid exposition.

Filter the elements below by keyword to find the exact line syntax and rules.

How it works

An OpenMetrics document is a sequence of metric families. Each family begins with optional metadata lines: # HELP name text (a description), # TYPE name type (one of counter, gauge, histogram, summary, info, stateset, gaugehistogram, unknown), and # UNIT name unit (the base unit, which must match the metric-name suffix).

Sample lines follow the form metric_name{label="value",...} value [timestamp] [# exemplar]. Counters expose their value with a required _total suffix and may add _created. Histograms expose _bucket{le="..."}, _sum, and _count. Strings in label values are escaped for \, ", and newline. The whole payload ends with a line containing exactly # EOF; a parser treats its absence as truncation. The HTTP Content-Type is application/openmetrics-text; version=1.0.0; charset=utf-8.

Tips and examples

A minimal valid exposition:

# HELP http_requests Total HTTP requests.
# TYPE http_requests counter
# UNIT http_requests requests
http_requests_total{method="GET",code="200"} 1027 1700000000.0
http_requests_created{method="GET",code="200"} 1699990000.0
# EOF

A histogram bucket with an exemplar pointing to a trace:

http_latency_seconds_bucket{le="0.1"} 8 # {trace_id="abc123"} 0.067 1700000000.0

Always terminate with # EOF, keep the _total suffix on counters, and ensure the declared # UNIT matches the name suffix or strict parsers will reject the payload.

OpenMetrics Exposition Format

Email me this result

OpenMetrics exposition format reference

How it works

Tips and examples