Skip to main content
In a consumption model, configure budgets to take action and protect against overspending. For each budget, define thresholds and priorities to define the actions to take when a threshold is exceeded, and control the data that gets dropped. You can then attach budgets to a partition. The consumption model includes the following concepts, which are volume-based licensing resources for telemetry data that you can spend credits on.

​
Logging license consumption

Spend credits flexibly across the following log resources in the consumption model:
  • Persisted bytes: Log bytes stored in the database.
  • Processed bytes: Log bytes matched for transformation and reshaping.

​
Tracing license consumption

Spend credits flexibly across the following trace resources in the consumption model.
  • Persisted bytes: Trace bytes stored in the database.
  • Processed bytes: Trace bytes matched for transformation and reshaping.

​
Metrics license consumption

Spend credits flexibly across the following metric resources in the consumption model.

​
Metrics persisted datapoint

A metrics persisted datapoint is an individual, timestamped data point that Observability Platform persists to storage. Each persisted data point counts as exactly one unit, regardless of metric type. Persisted datapoint pricing varies by effective datapoint resolution (EDR). EDR is the average time between consecutive persisted data points of a time series, computed across all persisted series and data points in your tenant.

​
Metrics persisted series

A metrics persisted series is a unique time series, defined by a distinct combination of metric name and labels, that Chronosphere Observability Platform persists to storage. Unlike capacity licensing, which counts all active time series within a 2.5-hour rolling window, the consumption model counts one time for each unique persisted series when first observed.

​
How persisted series counting works

In the consumption model, a persisted series is counted only the first time a unique time series (a specific combination of metric name and labels) is saved to storage. The following animated image shows five instances of a service deploying at a four-minute interval. Each instance produces one unique series, for a total of five persisted series. Because each series is counted only once at first observation, the total persisted series count is five, regardless of how many data points those series produce over time.
Animated image showing how persisted series are counted as five service instances deploy.
Observability Platform uses a 48-hour window to determine whether a series is new. If a time series continuously receives data points, or reappears within 48 hours of its last write, it isn’t counted again. Only if a series reappears after a gap of more than 48 hours is it treated as new and incurs a new persisted series charge. The following example shows an incoming series, followed by gap that exceeds the 48-hour window. Any series after that period is considered a new series, which incurs a new persisted series charge.
Animated image showing how data series are calculated across a 48-hour window.
In practice, the vast majority of time series receive data points continuously for their lifetime and are counted only once. The 48-hour window prevents double-counting for series with brief gaps, rather than defining what’s active at any moment.

​
How the consumption model differs from capacity

In the capacity model, the 2.5-hour window is inclusive: all series seen within the window count toward the cardinality limit, regardless of whether they’re still actively emitting. Churn inflates the count because both old and new series overlap in the window. In the consumption model, the 48-hour window is exclusive: it prevents recently seen series from being double-counted. A longer window reduces duplicate charges, and 48 hours is long enough that most series are never counted more than once. This separation means cardinality in the consumption model reflects the actual number of distinct series persisted, independent of timing effects.

​
Combining persisted series and datapoints

In the consumption model, persisted datapoints and persisted series form the two distinct halves of your total persistence cost, acting together to separate cardinality from volume:
  • Persisted series: captures how many unique time series exist. Counted once per unique series at first observation.
  • Persisted datapoints: captures how much data those series produce over time. Counted for every data point written to storage.
The previous animated image contains five instances. With an effective datapoint resolution (EDR) of 60 seconds, if each series emits one data point per minute over a four-minute interval, the persisted data points total 20:
The persisted series count remains five because each series is counted only once, while the persisted data points count grows with each data point written.

​
Metrics persisted histogram datapoint bucket

A metrics persisted histogram datapoint bucket is a bucket within a histogram data point that Observability Platform persists to storage. One persisted histogram datapoint bucket is incurred for each non-zero bucket within a persisted histogram datapoint. Only populated bucket positions in the histogram’s sparse encoding are counted.

​
Metrics matched datapoint

A metrics matched datapoint is a data point matched for aggregation. One matched data point is incurred for each aggregation rule matched for each datapoint. If a data point matches one rule, that’s one matched datapoint. If a data point matches two rules, that’s two matched data points. Recording rules don’t count toward matched data points.

​
Metrics matched histogram datapoint bucket

A metrics matched histogram datapoint bucket is a bucket within a histogram data point matched for aggregation. One matched histogram datapoint bucket is incurred for each non-zero bucket within a matched histogram datapoint, multiplied by the number of aggregation rules matched.