Interpolation

Fill

FunctionDescriptionExample
fill()Interpolate missing metric values for the metric.<METRIC_NAME>{*}.fill(<METHOD>, <LIMIT>)

The fill() function has two parameters:

  • METHOD: The function to use as an interpolation method; choose from:

    • linear: Gives you a linear interpolation between the beginning and the end of the gap.
    • last: Fills the gap with the last value of the gap.
    • zero: Fills the gap with a zero value.
    • null: Deactivates the interpolation.
  • LIMIT [optional, default=300, maximum=600]: The interpolation limit (in seconds) that represents the maximum size of a gap you want to interpolate.

Read Interpolation and the Fill Modifier for a detailed explanation of the .fill() function and its impact on interpolation.

Default zero

FunctionDescriptionExample
default_zero()Adds a default value to sparse metrics.default_zero(system.load.1{*})

The default_zero() function fills empty time intervals using the value 0 or, if interpolation is enabled, with interpolation. Note: Interpolation is enabled by default for GAUGE type metrics. Like most functions, default_zero() is applied after time and space aggregation.

Use cases

The default_zero() function is intended to address the following use cases (though it may also work for other use cases):

  • Aligning gauges as 0 when performing arithmetic on sparse metrics (note: COUNT or RATE type metrics queried as_count() or as_rate() are always aligned as 0, so using default_zero() does not change how they are aligned; it only affects GAUGE type metrics).
  • Resolving monitors before they enter a no-data condition. This works for both simple and multi alerts, but the value 0 must not cause the monitor to trigger. For example, this would not work for a monitor with the query avg(last_10m):avg:system.cpu.idle{*} < 10 because this monitor triggers (instead of resolving) when it evaluates to 0. Avoid using this function for error rate monitors with as_count() queries. See the as_count() in Monitor Evaluations guide for more details.
  • Filling in empty intervals in sparse (but nonempty) series for visual reasons or to affect the min/max/average of a timeseries in a monitor evaluation. If the evaluation window doesn’t contain any points of data, default_zero() has no effect.
  • Showing the value 0 on the timeseries widget when there is no data.

Example

To demonstrate how the default_zero() function works, consider this single point created for a custom metric using DogStatsD:

$ echo -n "custom_metric:1|g" | nc -4u -w0 127.0.0.1 8125

When this metric is queried over the past 30 minutes, there is a single timestamp, because only one of the query’s rollup intervals has a point:

avg:custom_metric{*}

+---------------------+---------------+
| Timestamp           | custom_metric |
+---------------------+---------------+
| ---------           | ---------     |
| 2019-04-17 17:45:00 | 1             |
+---------------------+---------------+

The default_zero() function interpolates this point five minutes forward in time (the default interpolation limit for gauges), then fills the remaining empty intervals with zeros:

default_zero(avg:custom_metric{*})

+---------------------+-----------------------------+
| Timestamp           | default_zero(custom_metric) |
+---------------------+-----------------------------+
| ---------           | ---------                   |
| 2019-04-17 17:30:00 | 0                           |
| 2019-04-17 17:31:00 | 0                           |
...
| 2019-04-17 17:44:00 | 0                           |
| 2019-04-17 17:45:00 | 1                           |
| 2019-04-17 17:46:00 | 1                           |
| 2019-04-17 17:47:00 | 1                           |
| 2019-04-17 17:48:00 | 1                           |
| 2019-04-17 17:49:00 | 1                           |
| 2019-04-17 17:50:00 | 1                           |
| 2019-04-17 17:51:00 | 0                           |
| 2019-04-17 17:52:00 | 0                           |
...
+---------------------+-----------------------------+

Further reading

PREVIEWING: may/unit-testing