Monitoring Sparse Metrics

문서 > 모니터링 > 모니터링 지침 > Monitoring Sparse Metrics

이 페이지는 아직 한국어로 제공되지 않으며 번역 작업 중입니다. 번역에 관한 질문이나 의견이 있으시면 언제든지 저희에게 연락해 주십시오.

Overview

Monitors that report data infrequently, can have unexpected results and queries may not evaluate as intended. There are tools and behaviors that you can use to ensure a monitor’s settings are appropriate for your data and expected evaluations.

This guide covers the following ways of troubleshooting and configuring monitors with sparse data:

Determine if you have sparse metrics
Consider the source of your monitor -> Metric-based monitor, Event-based monitor
Is the monitor running on a schedule?

How to determine whether you have a sparse metric

You can use a dashboard widget, a notebook, or even an existing monitor’s history graph and hover over the datapoints to see if the datapoints seem continuous, as opposed to straight lines filling the gaps between each point.

In a notebook, or widget, select the Bars display option to see the points of data and their frequency.

A metric displayed in a widget may look like this:

Metric graph with Line graph display going up and down in straight lines

But when the Bars style is applied, it looks like this:

Same data as the Metric Line graph above, except with bars for each datapoint, highlighting gaps in between bars of sparse metrics

With the bar graph display, you can visualize the gaps between datapoints more clearly.

If the graph editor does not have multiple options to change the graph style, you can apply the function default_zero() to the metric, which helps reveal the gaps in data. For more information on this function, see the Interpolation documentation.

Metric-based monitor

Is this a metric, change, anomaly, forecast, or outlier monitor? Adjust the following settings:

Under Advanced options, select Do not require a full window of data for evaluation.
Is the data often delayed? Consider adding time (in seconds) to the monitor evaluation delay. Under Advanced options add a value to the Delay monitor evaluation by X seconds field.
Adjust the evaluation (avg by, max by, min by, sum by) based on the expected frequency. The default evaluation is avg by, which may not be suited for sparse metrics.
If you are using the avg by aggregator, consider adding an interpolation function like default_zero() to ensure the gaps in the metric are evaluated as zero.
If you are using arithmetic in your query, take a look at Monitor Arithmetic and Sparse Metrics for some further guidance.

Event-based monitor

Is this a log, event, audit trail, or error tracking monitor? Look at the following:

Verify the “Missing data” setting corresponds to your expected monitor behavior: Evaluate as zero, Show NO DATA, Show NO DATA and notify, or Show OK
Adjust the evaluation period. If datapoints are expected to be available every 30 minutes, then the evaluation period should account for that.

Schedule-based monitoring

Are you monitoring an event that needs to happen at certain times of the day, week, month? A CRON task such as a backup job or export? Consider using Custom Schedules, which allow you to set RRULES to define when the monitor should evaluate and notify.