The built-in monitoring in RDS and Aurora was deprecated on 30 June 2026. Most teams have not replaced it properly, and the gap is quieter than an outage.
Start a conversationAmazon deprecated the built-in Performance Insights monitoring in RDS and Aurora on 30 June 2026, extended from an original November 2025 date.
The deadline passed quietly, which is the problem. Nothing broke. No alarm fired. Databases carried on serving traffic exactly as before. The only change is that the tool many teams relied on to understand why a database was slow is no longer there.
Performance Insights was not another CPU graph. It sampled active sessions continuously and attributed database load to the thing causing it: a specific SQL statement, a specific wait event, a specific host or user.
That distinction matters. CloudWatch tells you the database is busy. Performance Insights told you which query, waiting on what, and since when.
For most teams it was the only wait-event visibility they had, and it was free enough at the default retention that nobody had to justify it.
Three questions, answerable in about ten minutes:
A surprising number of estates answer no to all three and have not noticed, because the absence only shows up under pressure.
There is no single drop-in. The realistic options fall into three groups.
PostgreSQL exposes most of what you need through pg_stat_statements, pg_stat_activity, and the wait event columns. Sample active sessions on a short interval, store the samples, and you have the same data model Performance Insights used.
Cheapest, and it works everywhere including on-premises and other clouds. It is also the option most teams underestimate, because the collection is easy and the retention, aggregation and presentation are not.
Several vendors do database performance monitoring properly, with query-level attribution, plan history and regression detection. Better than what you lost, and priced accordingly.
If there is already a Prometheus, Datadog or New Relic footprint, database exporters will get you part of the way. Be honest about how far. Generic infrastructure monitoring rarely gives real wait-event attribution, and a dashboard that looks complete while missing the diagnostic layer is worse than an obvious gap.
Whatever you choose, capture a baseline now.
The most useful property of a monitoring system is not the live view. It is the ability to answer whether this is normal. That requires history, and history only exists if collection started before the incident.
Every week without collection is a week of comparison you cannot get back.
Managed services remove operational burden and, occasionally, they remove a capability. When that happens, the need does not disappear with the feature. It becomes invisible, and it lands on whoever is on call the next time something is slow at two in the morning.
Worth checking what else in your estate depends on a managed feature you did not choose and cannot control.
If any of the above matches a problem in front of you, the fastest way in is a fixed-scope assessment. Two to three weeks, written findings, yours to keep either way.