Accelerate Database Troubleshooting with Grafana Assistant's AI-Powered Insights

When your database slows down, pinpointing the root cause can be challenging even with detailed metrics. Grafana Cloud Database Observability provides RED metrics, execution samples, and wait event breakdowns, but interpreting that data requires expertise. The new Grafana Assistant integration brings AI-powered guidance directly into your investigation workflow, eliminating guesswork and saving time.

What is the Grafana Assistant for Database Observability?

The Grafana Assistant is an AI-driven feature integrated into Grafana Cloud's Database Observability tool. It helps you troubleshoot slow or degraded SQL queries by automatically analyzing your database's real-time metrics and logs. Instead of copying queries into a separate AI tool, the assistant runs analyses against your actual Prometheus and Loki data sources within the same time window you are investigating. It uses your real table schemas, indexes, and execution plans to provide context-aware recommendations. Built by database engineers, the assistant includes purpose-built analysis actions tailored to common performance problems rather than generic prompts. It can identify issues like inefficient joins, lock contention, or table scans that become problematic as data grows, and it explains complex wait events in plain language. The assistant is available directly within the query investigation interface, allowing you to click a button and receive a health assessment without manually assembling context.

Accelerate Database Troubleshooting with Grafana Assistant's AI-Powered Insights

How does the assistant differ from using a generic AI chatbot?

Generic AI tools require you to manually copy and paste SQL queries, schema details, and time ranges, which is time-consuming and error-prone. They lack direct access to your database environment. The Grafana Assistant overcomes these limitations by querying your actual Prometheus and Loki data sources automatically. It knows the exact time window you're viewing and has your table schemas, indexes, and execution plans already loaded. This means every analysis is based on real data from your database, not a static snapshot. The assistant's prompts are designed by database engineers for specific troubleshooting tasks, such as identifying why a query is slow or recommending schema changes. It synthesizes data from multiple sources into a single health assessment, correlating metrics like row examination ratios, latency spikes, and wait event overhead. This deep integration allows the assistant to provide specific, actionable advice that generic chatbots cannot match.

What pre-built prompts are available for common issues?

The Grafana Assistant comes with ready-to-use AI buttons that guide you through common database performance problems. These include prompts for analyzing slow or degraded queries, getting recommendations on schema or index changes, and understanding unexpected error spikes. For example, if you notice a query's duration jumping, you can click the 'Why is this query slow?' button. The assistant then automatically runs a multi-source analysis using both Prometheus and Loki to examine the selected time window. It checks whether the problem is due to excessive rows examined, lock contention, wait events, or CPU pressure. Another prompt helps with wait event diagnostics, translating cryptic names like wait/synch/mutex/innodb into understandable causes. You can still free-type your own questions, but these guided prompts ensure you get consistent, expert-level analysis without needing to formulate complex prompts yourself.

Can you give an example of how the assistant diagnoses a slow query?

Suppose you identify a query where the P99 latency has spiked and error rates are climbing. Clicking the pre-built prompt initiates an analysis. The assistant queries your Prometheus and Loki data sources for that specific time window and synthesizes the results into a health assessment. It might reveal that the number of rows examined is 50 times the number of rows returned, indicating most work is wasted on filtering. The P99 being 12× the median suggests an intermittent problem rather than a constant issue. It also checks CPU time—if that's healthy but wait events consume 40% of execution time, the bottleneck is likely I/O or locking. The assistant will then explain these findings in plain language and suggest next steps, such as reviewing the query plan, adding an index, or rewriting the join logic. This diagnosis happens in seconds, far faster than manual analysis.

How does the assistant explain complex wait events?

Database wait events often have names like wait/synch/mutex/innodb or io/table/sql/handler that are not self-explanatory. The Grafana Assistant understands these events because it has access to your real execution context, including the specific database engine version and configuration. It can interpret what each event means during the selected time window. For example, if wait/synch/mutex/innodb appears, the assistant may explain that this indicates contention for an internal InnoDB mutex, often caused by multiple threads trying to modify the same data page. It then correlates this with query patterns, row locks, and transaction isolation levels to provide targeted advice, such as reducing concurrent writes or adjusting innodb_thread_concurrency. By translating technical jargon into actionable insights, the assistant helps even less experienced users understand performance bottlenecks without needing deep database internals knowledge.

Is my data safe when using the assistant?

Yes, the Grafana Assistant is designed with privacy and security in mind. Your query text and schema metadata are used only for the current analysis and are not stored or used for model training. The assistant runs against your own Prometheus and Loki data sources within the same cloud environment, so your data never leaves your Grafana Cloud instance. Each analysis is ephemeral—once the result is delivered, the raw query details are discarded. This means sensitive SQL logic, table names, and business data remain under your control. The assistant's AI models are trained on anonymized public data and do not learn from your specific queries. You can confidently use the assistant to investigate production performance issues without worrying about data exposure or compliance violations.

How do I access the assistant during query investigation?

Accessing the Grafana Assistant is straightforward. When you are investigating a specific query in Database Observability, you will see an assistant icon or button within the query detail panel. Clicking it opens the assistant chat interface with the relevant context already loaded. The assistant automatically knows which query you are looking at, the time range you have selected, and the available schema and execution plan data. You can then either use one of the pre-built prompt buttons or type your own question. The assistant responds inline with its analysis, and you can continue the conversation to dive deeper. This seamless integration means you don't have to switch between tools or manually re-enter context. It works directly within your existing investigation workflow, making it faster to get answers and resolve performance issues.

Tags: