Amazon Web Services announced on November 17, 2026 that Amazon Aurora DSQL can now record statistics for every SQL statement a cluster runs. The numbers are published as CloudWatch OpenTelemetry metrics and appear in the Top SQL view of Amazon CloudWatch Database Insights.
This fills a clear gap. Until now, Database Insights for Aurora DSQL relied on sampled session activity, which shows where load comes from but not exact totals per query. Teams running the serverless, PostgreSQL-compatible database can now see call counts, timing, rows and DPU cost for each statement, which makes tuning and cost work far more precise.
What AWS Announced
The new feature is called statement statistics. When it is switched on, Aurora DSQL measures every successful run of a statement instead of taking samples, then rolls the results up for each distinct statement. You can then rank statements by total time, frequency or cost.
For each statement, the service reports the following figures:
- How many times it ran, counted as execution calls.
- Planning plus execution time, with a split between time spent in the query processor and time spent on storage reads.
- Rows returned or changed, plus an estimate of the rows that storage scans visited.
- Compute, read and write DPUs, and a separate write DPU figure for the peered Region of a multi-Region cluster.
The feature is offered in every AWS Region where Aurora DSQL runs. AWS adds no Aurora DSQL charge for it, but normal CloudWatch metrics and CloudWatch Logs pricing applies.
How Statement Statistics Work
According to the Aurora DSQL user guide, the counters cover SELECT, INSERT, UPDATE and DELETE. Utility commands such as BEGIN, COMMIT and SET are left out, and so are DDL statements. Only top-level statements that your application sends directly are counted.
Statements are grouped by normalized text. Aurora DSQL strips comments, collapses extra spaces and swaps each literal for a numbered placeholder, so two queries that differ only in their values share one query ID. Up to 32 KiB of that normalized text is written to a shared CloudWatch Logs group, where you can look it up by query ID.
Two counting rules are worth knowing. Statements inside a transaction that later fails to commit are still counted. If your code retries that transaction, the statements count again, so retry storms show up clearly in the call numbers.
Data reaches CloudWatch in one-minute windows. Every point carries just the delta since the last one, and the service tracks a distinct series per combination of cluster, query ID, connecting IAM principal and app name. Database Insights adds these up for you.
Where to See the Data
The main entry point is Database Insights. You pick the Aurora DSQL cluster in the fleet view, then choose a statement under Top SQL. The detail page shows the normalized SQL, calls and rows per second, average duration, DPU per call, and scanned versus returned rows per call.
For automation, the same metrics can be queried with PromQL in CloudWatch Query Studio. The guide explains that the values are deltas, so queries should sum data points over a window rather than use rate-style functions. A single query returns at most 500 series and covers up to 24 hours.
What It Means for Developers
The most useful signal for many teams will be scanned rows compared with returned rows. A statement that visits thousands of rows to return a handful usually needs a better index or a tighter filter. With per-statement data, that pattern is now easy to spot without guessing from samples.
DPU per call matters just as much. Aurora DSQL bills by distributed processing units, so knowing which statements burn the most read or write DPUs lets you tie cost directly to code. A small fix to one hot query can lower the monthly bill noticeably.
Set the application name in your connection settings. When it is present, the metrics carry it as a label, so you can tell which service sent a costly query. Without it, statements from different apps that share an IAM role are harder to separate.
Keep the CloudWatch cost in mind. Every distinct statement creates its own series, so workloads that build SQL with many unique shapes can produce a lot of metrics. Using bind variables keeps normalized statements grouped and the series count under control.
How to Turn It On
Statement statistics are off by default and are set per cluster and per Region. In the console, open the cluster, go to the Monitoring tab and enable statement statistics under Advanced monitoring, or tick the option in Advanced configuration when you create a cluster.
From code, add the STATEMENT_STATISTICS metric group through the metricGroups parameter of CreateCluster or UpdateCluster, or the MetricGroups property in CloudFormation. Multi-Region clusters need the setting in each Region. The change takes effect within a few minutes.
Conclusion
Aurora DSQL now gives exact, per-statement numbers for time, rows and DPU use. It is free on the DSQL side and simple to switch on.
Teams running Aurora DSQL in production should enable it on key clusters and watch the CloudWatch bill for the first few days. For another look at serverless PostgreSQL options, see our report on AlloyDB Express with scale to zero.
Sources
- AWS What's New: Amazon Aurora DSQL now offers SQL statement statistics in Database Insights
- Aurora DSQL User Guide: Monitoring SQL statements with Aurora DSQL statement statistics
See my articles more often when you search Google.
