Intelligent Insights which applies to Azure SQL and SQL Managed Instances, uses artificial intelligence based on data gained from monitoring many thousands of Azure SQL databases, to proactively warn you of performance issues within your database, what the root cause of the problem is, and recommend a solution.

While the AI of Intelligent Insights can leverage off all Azure SQL databases to some extent, for any one database it compares current performance and workload over the last hour, with that of the past week, or seven days. This is a way for Intelligent Insights to perform a kind of baseline comparison analysis, with a moving average baseline.

While a baseline comparison gives you an idea of relative performance, there are absolute metrics like errors and corrupted data that are also monitored and reported. Not only does Intelligent Insights use AI to monitor performance, but artificial intelligence is also utilized to diagnose and establish the root cause of any detected issues. This all sounds great in theory, but at the same time a little vague.

So how does Intelligent Insights actually detect performance issues? Well, it uses a combination of query duration, timeout requests, excessive wait times, and errors. Query duration is reasonably self-explanatory, as in a query taking a long time to complete, in comparison with its baseline performance, that is average completion time over the last seven days.

In a similar mode, it monitors the number of requests that are currently timing out versus the baseline. Excessive wait times can give a more granular view of what is happening with query execution, and what might be the root cause of slow or failing queries.

Errors are never a good thing and artificial intelligence helps to determine whether the error, is due to a database or degrading performance issue. An insight is an instance of a database issue or performance degradation. Each insight is recorded in the SQL Insights Log, with the following information.

There is the database where the issue has occurred, along with when it occurred and for how long it continued to be an issue. The impact or issue is defined in terms of those metrics we have just talked about, query duration, excessive wait time, timed-out requests, and errored-out requests, with their respective values.

Any queries that have been impacted are identified via a query hash or error code. Intelligent Insights will try to categorize the problem, as one of, what Azure calls 15 detection patterns. These detection patterns are things like, running out of resources due to an increase in workload, or reaching the limit of the current pricing tier, or due to a pricing tier downgrade.

There are several contention patterns, like page latch or TempDB contention, or excessive locking. There are configuration patterns, as in a database configuration change has been detected, or the MAXDOP setting has been changed, affecting the degree of parallelism. There are also query-based patterns, like a missing index, a new query is impacting performance, or a query is suffering from plan regression.

For a full description of the detection patterns, visit docs.microsoft.com/azure/azure-sql/database/intelligent-insights-troubleshoot-performance. The insight also includes a human-readable root cause analysis of the issue, with a performance improvement recommendation if applicable.