Network extensions let you set the polling interval to any value from 1 to 60 minutes. Intervals longer than 1 minute change how some metrics behave, so custom dashboards, notebooks, and alerts that were built for 1-minute polling can show wrong values or stop triggering.
This page applies to Latest Dynatrace. In Dynatrace Managed, you can't correct the values at query time, so keep the polling interval at 1 minute.
Count metrics (metric keys ending in .count, such as com.dynatrace.extension.network_device.if.bytes_in.count) report how much a counter increased since the previous poll. The longer the polling interval, the larger each value.
The metric stays accurate and totals over any period are unchanged, but anything built around the values of the old polling interval is affected:
Metrics from network extensions carry a polling.interval dimension with the interval in minutes, for example "5". Use it to normalize each count to "per minute", so your conversions and thresholds keep working whatever the polling interval is.
The following example converts interface traffic to bits per second.
Before
timeseries bytes_in = sum(com.dynatrace.extension.network_device.if.bytes_in.count),by: {dt.smartscape.ext_network_interface}| fieldsAdd bits_per_second = bytes_in[] * 8 / 60
After
timeseries bytes_in = sum(com.dynatrace.extension.network_device.if.bytes_in.count, rollup: avg),by: {dt.smartscape.ext_network_interface, polling.interval},union: true// Normalize each value to "per minute"| fieldsAdd polling_minutes = coalesce(toDouble(polling.interval), 1)| fieldsAdd bytes_in_has_value = iCollectArray(if(isNotNull(bytes_in[]), 1, else: 0))| fieldsAdd bytes_in = bytes_in[] / polling_minutes// Merge back into one series per interface, keeping gaps as gaps| summarize {bytes_in_summed = sum(bytes_in[]),bytes_in_samples = sum(bytes_in_has_value[]),interval = takeFirst(interval),timeframe = takeFirst(timeframe)}, by: {dt.smartscape.ext_network_interface}| fieldsAdd bytes_in = iCollectArray(if(bytes_in_samples[] > 0, bytes_in_summed[], else: null))| fieldsRemove bytes_in_summed, bytes_in_samples| fieldsAdd bits_per_second = bytes_in[] * 8 / 60
After normalization, all count metrics are counts per minute, whatever the polling interval. Any conversion you apply afterward must start from counts per minute. For example, divide by 60 to get a rate per second, as in the last line of the example.
sum(<metric>.count, rollup: avg) for each count metric. After normalization, the values are rates, so you must average the samples in a time bucket, not add them: five samples of 10 per minute are still 10 per minute. The sum adds up the series that your by: doesn't split apart, for example, total traffic of a device across its interfaces. If you already split down to one series per entity, such as per interface, avg(<metric>.count) returns the same result.polling.interval to by:, or add by: {polling.interval} if your query has none. Add union: true so that data without the dimension is kept.coalesce(toDouble(polling.interval), 1). The 1 covers data without the dimension, such as data recorded before it existed or sent by extension versions without a configurable interval. Those extensions polled every minute.summarize by your original by: fields, keeping interval and timeframe with takeFirst. Restore null where no series had a sample, because summing empty buckets would show 0. Then remove the helper fields.default: 0 from the count aggregation. Zero-filling before the merge hides the gaps.Delete the alert and create it again from the updated alert template. The updated templates already normalize the values to the polling interval. What you do next depends on where the template comes from:
Infrastructure & Operations set the sliding window and violating samples automatically. Check that they suit your needs. If you change a polling interval later, create the alert again.Replace the alert query with the updated version from Update your queries.
Alert sliding windows assume one value per minute. With a longer polling interval, fewer values fall into the same window, so alerts might no longer trigger as you configured them. Longer polling intervals affect every alert on the metrics of these extensions, including alerts created from templates, not only alerts on count metrics.
For example, an alert with the default of 3 violating samples out of 5 never triggers with 5-minute polling, because a 5-minute window contains only one value.
In the alert's Advanced tab in
Anomaly Detection (see Configure an advanced custom alert), adjust the sliding window and the number of violating samples to the new polling interval:
It is recommended to keep both the polling interval and the sliding window reasonably small for accurate alert detection. Larger windows detect problems later and can miss short ones.
If Alert on missing data is turned on, review it as well. With a longer polling interval, most 1-minute buckets are empty, and the alert can fire only because of those gaps. For details, see Anomaly detection configuration.
Extensions