Simplify Database Diagnostics: Grafana Assistant Integrates AI to Speed Up Query Analysis

So your database is running sluggishly. Where do you even begin? Grafana Cloud Database Observability already offers deep visibility into SQL queries with RED metrics, execution samples, wait event breakdowns, table schemas, and visual explain plans. But visibility alone isn't enough. You might see a query's P99 latency spike, but what should you do about it? Or you encounter wait events like wait/synch/mutex/innodb firing, but what does that actually mean? The new Grafana Assistant integration for Database Observability bridges that gap, turning raw data into clear, actionable guidance.

From Raw Data to Actionable Answers

The Grafana Assistant is not a generic AI tool. It's built directly into the Database Observability interface, so you don't have to copy SQL or manually describe your schema. Instead, the assistant automatically connects to your actual Prometheus and Loki data sources, pulling in the exact time window you're investigating, along with real table schemas, indexes, and execution plans. This means every analysis is grounded in your database's live state.

Purpose-Built Analysis Actions

Rather than relying on vague prompts, each tab in the assistant comes with targeted analysis actions designed by database engineers. These actions are tailored to common performance scenarios, such as slow queries, lock contention, or inefficient joins. Your query text and schema metadata are used only for the current analysis and are never stored or used for model training, ensuring your data remains private.

Tackling Common Database Issues with Pre-Built Prompts

You can still ask free-form questions in the assistant's chat box, but the real power lies in the out-of-the-box AI buttons. These buttons provide a guided, step-by-step experience for diagnosing and resolving performance problems. Let's walk through a typical example to see how the assistant accelerates troubleshooting.

Why Is This Query Slow?

Imagine you've spotted a query in the overview: its duration is spiking and the error rate climbing. You dive into its detail page and see time-series performance data—but the cause isn't obvious. Is it a bad join? Lock contention? A table scan that was fine until data grew? With one click on the assistant's pre-defined prompt, the issue becomes clear.

The assistant immediately queries both Loki and Prometheus across your chosen time window, synthesizing the data into a single health assessment. It might reveal that the duration is spiking because the number of rows examined is 50 times the number of rows returned—meaning most work is wasted on filtering. It also notes that the P99 latency is 12 times the median, indicating an intermittent issue, not a constant one. CPU time looks healthy, but wait events are consuming 40% of execution time.

Decoding Cryptic Wait Events

Wait event names like wait/synch/mutex/innodb or io/table/sql/handler are not self-explanatory. The assistant, however, understands these names and translates them into plain language. It tells you: "During this wait, the database is experiencing contention on internal structures, likely due to concurrent updates on the same rows." This guidance lets you take specific actions—such as optimizing indexing or reducing lock contention—without needing to memorize dozens of event codes.

Real-Time, Real-Data Analysis

What sets the Grafana Assistant apart is its direct integration with your observability stack. Every recommendation is based on your actual data, not a snapshot or a generic best practice. The assistant can cross-reference execution plans, wait event frequency, and historical patterns to give advice that fits your unique workload. And because it uses your existing Prometheus and Loki setup, there's no extra infrastructure to deploy.

To get started, simply open the assistant from any query detail page and click the pre-built prompt that matches your symptom. You'll receive a concise, data-driven diagnosis with actionable next steps. For more advanced investigations, you can still type custom questions into the chat.

Related: Learn more about how the assistant gathers context or explore common troubleshooting scenarios.