Clues

Almost all signs of impending or current problems with a database’s performance are visible on the AWS console, either in Cloudwatch Performance Insights or RDS Performance Insights. The metrics for these symptoms should have alarms set up so that we are notified when the symptom becomes visible and can respond quickly.

Signs that there is a problem

• High CPU utilisation
• High DB load (CPU vs non-CPU)
• High connection count
• High wait load
• Worst-case scenario: performance impact for users

These clues require knowing the usual values for a specific project, so it’s difficult to prescribe concrete numbers representing a problem. Each project must analyse its standard workload, including predictable peak periods, and then decide on reasonable alarm limits for these metrics.

Each of the signs can also be caused by any number of issues, so if one of these metrics goes into alarm it is important to work through the potential causes to diagnose the problem. High connection counts often go hand-in-hand with high CPU utilisation, for example, so figuring out whether the connection count is an isolated issue or caused by something else is imperative.