Refreshing warehouse native experiment results

This topic explains how often LaunchDarkly refreshes results for experiments that use warehouse native metrics, how to change the refresh frequency, and how to refresh results on demand.

​

How result refreshes work

For experiments that use warehouse native metrics, LaunchDarkly computes results inside your warehouse. On a schedule, LaunchDarkly runs the computations for each metric of each running experiment iteration linked to the integration, and stores the aggregated results. Your raw event data never leaves your warehouse. Each refresh recomputes the metric over the full duration of the iteration, not only the time since the last refresh. This means:

• A refresh reflects any late-arriving or corrected data in your warehouse, as long as the data is within the iteration’s time range.
• The warehouse compute used by each refresh grows with the age of the iteration, the number of metrics in the experiment, and the volume of events in the metrics’ data sources.
• Refreshing more often multiplies that compute. Choose a refresh frequency that matches how quickly you need new data.

The experiment’s Results tab shows when the results were last updated and when the next scheduled update is due. To learn more, read Warehouse.

​

Refresh frequency

Each warehouse integration has a Refresh frequency. It applies to every experiment that uses metrics from the integration. You can choose one of the following:

• 15 minutes
• 30 minutes
• 1 hour
• 3 hours
• 6 hours
• 12 hours
• 24 hours, the default

Shorter intervals give you fresher results and use more warehouse compute. Longer intervals do the opposite. If a metric’s computation is still running when the next refresh is due, LaunchDarkly skips that metric for that refresh. The metric is recomputed on the next scheduled refresh after its earlier computation has finished.

​

New iterations refresh every 15 minutes

During the first 24 hours after an iteration starts, LaunchDarkly refreshes its results every 15 minutes regardless of the integration’s refresh frequency. This lets you confirm early that exposures and metric events are arriving as expected, when a configuration problem is easiest to fix. After 24 hours, the iteration follows the integration’s refresh frequency. While an iteration is in its first 24 hours, the Warehouse section in the Results tab shows “15 minutes” as the update frequency. Click the info icon next to it to display when the iteration switches to the integration’s refresh frequency.

​

Change the refresh frequency

To change the refresh frequency for a warehouse integration:

1. In the left sidebar, click the gear icon to open Organization settings.
2. Under “Connections”, click Integrations. Find your warehouse native Experimentation integration.
3. Click View all. A list of existing integrations appears.
4. Click the pencil icon next to the integration you want to change. The integration details panel appears.
5. Under “Refresh frequency,” click the pencil icon and choose a new frequency. An “Update refresh frequency” dialog appears.
6. Click Update.

The new frequency applies to every experiment that uses the integration, starting from the next scheduled refresh. Changing the frequency does not start a refresh.

​

Warehouse

The Results tab of an experiment that uses warehouse native metrics includes a Warehouse section, which displays the following information:

• Connection status: “Connected” when LaunchDarkly can reach your warehouse. “Disconnected” when it cannot, along with when LaunchDarkly lost the connection and why. LaunchDarkly cannot refresh results while the integration is disconnected. To learn how to troubleshoot, read Warehouse health checks.
• Results updated: When LaunchDarkly last computed the results from your warehouse.
• Next update: When the next scheduled refresh is due.
• Update frequency: The refresh frequency currently in effect for the iteration.

While a refresh is running, whether scheduled or requested on demand, the section displays “Refresh in progress.” When the Results tab shows “Refresh in progress,” it reloads the results after it detects completion. Reload the page if newer results are not shown. If the most recent refresh failed, the section also displays the time of the attempt and the reason, such as a query that timed out or a table LaunchDarkly no longer has permission to read. The previous results remain available while you resolve the problem. Scheduled refreshes continue at the configured frequency. After you fix the underlying issue, the next refresh runs as scheduled.

​

Refresh results on demand

You can refresh an experiment’s results without waiting for the next scheduled refresh. This is useful after you backfill or correct data in your warehouse, or when you need to make a decision before the next scheduled update. To refresh results on demand:

1. Navigate to the experiment’s Results tab.
2. Click Refresh.

LaunchDarkly starts a refresh of every metric in the running iteration. While it runs, the button displays Refreshing and its tooltip shows how many of the metric computations have finished. When the Results tab shows “Refresh in progress,” it reloads the results after it detects completion. Reload the page if newer results are not shown. If the refresh fails, the Warehouse section displays the reason. An on-demand refresh recomputes every metric of the iteration over its full duration, the same work a scheduled refresh does for that iteration. It refreshes experiment results only. Data source activity, such as the event counts and last-seen times on the data source page, updates on the integration’s scheduled refresh. An on-demand refresh does not change the schedule. The next scheduled refresh still runs at its original time. Each on-demand refresh is recorded in the audit log.

​

When you can refresh

The Refresh button is available when all of the following are true:

• The experiment has a running iteration. You cannot refresh the results of a stopped iteration.
• The warehouse integration is connected.
• The environment has exactly one enabled warehouse native Experimentation integration for the experiment’s warehouse.
• You have permission to update the experiment. To learn more, read Experiment actions.
• No other refresh of the iteration is in progress, whether it was scheduled or requested on demand.

A refresh takes as long as your warehouse takes to run the metric queries, typically a few minutes. If a refresh has not reported completion after one hour, LaunchDarkly lets you request another one.