Skip to main content

Metrics Formats

Sumo supports the Graphite, Carbon 2.0, and Prometheus metric formats. 


Graphite metrics are formatted like this:

metric_path metric_value metric_timestamp


  • metric_path is a dot-separated string that identifies the thing being measured.
  • metric_value is any numeric value.
  • metric_timestamp is a UNIX timestamp.

Here’s an example of a graphite metric:  

cluster-1.node-1.cpu-1.cpu-idle 97.29 1460061337

In the metric above: 

  • The thing being measured is cluster-1.node-1.cpu-1.cpu-idle, which we could colloquially refer to as the cpu-idle value for cpu-1, in node-1, in cluster-1.  
  • The value measured is 97.29.
  • The timestamp for the instant that the metric was measured is 1460061337.

Graphite does not support meta tags. However, you can use Sumo's metric rules editor to tag metrics with key-value pairs derived from a Graphite metric’s metric_path. Then, you can use those key-value pairs in metric queries. For more information, see About Metric Rules.

Inferred metric name

The Graphite format doesn't have a notion of named tags. Since Sumo Logic requires each metric to have a valid metric name, for Graphite metrics, the metric name must be inferred. To accomplish that, for every Graphite metric sent to our backend, Sumo Logic adds an additional metric tag, whose value is equal to the last segment of a dot-separated metric_path.

For example, for below graphite metric sent to Sumo Logic:

cluster-1.node-1.cpu-1.cpu-idle  73.12 1112470620

the inferred metric name will be metric=cpu-idle and such a key-value pair will be added

Carbon 2.0

Carbon 2.0 metrics conform to the Metrics 2.0 specification.

Carbon 2.0 metrics look like this:

intrinsic_tags  meta_tags value timestamp

There are two spaces between intrinsic_tags and meta_tags. If a tag is listed before the double space, then it is an intrinsic tag. If a tag is listed after the double space, then it is a meta tag.


  • intrinsic_tags is one or more space-separated key-value pairs that uniquely identify what is being measured and are metric identifiers. Intrinsic tags are also referred to as dimensions. If you have two data points sent with same set of dimension values then they will be values in the same metric time series.


    intrinsic_tags must be followed by two spaces.

  • meta_tags is zero or more space-separated key-value pairs that provide additional, but not identifying information about the thing being measured. A meta tag is a piece of metadata that might be useful in querying your metrics. It doesn’t identify the thing being measured, but provides additional information of interest. Meta tags are meant to be used in addition to intrinsic tags so that you can more conveniently select the metrics. For example, it might be interesting to know the collection agent that obtained the measurement. 

  • value is any numeric value.

  • timestamp is a UNIX timestamp.

In the Graphite-formatted metric described above, the bit that identifies the thing being measured—the metric_path—is:


In the Carbon 2.0 format, that metric_path translates to a set of intrinsic tags:

cluster=cluster-1 node=node-1 cpu=cpu-1 metric=cpu_idle

Carbon 2.0 is a richer metric format than Graphite, because it supports meta tags. So, a Carbon 2.0-formatted version of our example metric could include additional key-value pairs that will make our metrics easy to query. For example, below we include a meta tag that identifies the agent that collected the metric.

cluster=cluster-1 node=node-1 cpu=cpu-1 metric=cpu_idle  agent=biggie <value> <timestamp>

The following is an example of intrinsic_tags with an agent meta_tags, a value, and a timestamp:

cluster=cluster-1 node=node-1 cpu=cpu-1 metric=cpu_idle  agent=biggie 97.29 1460061337

And an example of empty set of meta tags and two spaces between the intrinsic tags and value with timestamp :

cluster=cluster-1 node=node-1 cpu=cpu-1 metric=cpu_idle  97.29 146006133

This blog post provides additional examples and discussion.

Mandatory metric name

Unlike Prometheus, Carbon 2.0 format doesn't enforce the presence of a metric name. It also cannot be reliably inferred automatically. Therefore, Sumo Logic require a metric key to be present among intrinsic_tags. All metrics without a metric key specified will not be ingested to Sumo and a MetricsMetricNameMissing Health Event for the associated Metric Source will be triggered (for more information on Halth Events, see About Health Events).

For example, the following metric will be correctly ingested to Sumo Logic:

cluster=cluster-1 node=node-1 cpu=cpu-1 metric=cpu_idle  73.12 1112470620

while the one below will be dropped:

cluster=cluster-2 node=node-2 cpu=cpu-2  73.12 1112470620


In the Prometheus format, a time series is uniquely identified by its metric name and a set of labels, which are key-value pairs. It's formed like this:

# HELP metric_name metric_description
# TYPE metric_name metric_type
metric_name labels value timestamp
metric_name labels value timestamp

Here is an example of a Prometheus metric exposition for two time series. The process of making metrics available to Prometheus is called exposition.

# HELP http_requests_total The total number of HTTP requests.
# TYPE http_requests_total counter
http_requests_total{method="post",code="200"} 1027 1395066363000
http_requests_total{method="post",code="400"} 3 1395066363000

See the table below for descriptions of the components of a Prometheus metric exposition.

metric_nameSpecifies the general feature of a system that is measured. For example: http_requests_total
metric_descriptionAn arbitrary description or category for the metric. For example: requests
metric_typethe type of the metric, one of counter, gauge, histogram, summary, or untyped.
labelsZero or more space-separated key-value pairs that identify a particular dimensional instantiation of the metric, for example: http_requests_total{method="post",code="200"} 1027 1395066363000
http_requests_total{method="post",code="400"} 3 1395066363000
valueValue of the metric.
timestampThe time the metric was collected, in int64 format.

The Prometheus format does not support metadata in the format itself. You can attach metadata to Prometheus metrics by specifying it the HTTP header when you upload the metrics to Sumo. For more information, see Upload Metrics to an HTTP Source.

Sumo Logic YouTubeSumo Logic Twitter
Privacy Statement
Terms of Use

Copyright © 2022 by Sumo Logic, Inc.