AI DBA?
September 12
The adaptive optimizer is a hot topic in every database development circle. But has anyone truly surveyed whether clients prefer stability over miraculous, last-minute recoveries? Oracle has offered optimizer_adaptive_plans for over a decade. Yet, in my experience managing large enterprise environments, we almost always kept it disabled. Why? Because in high-load OLTP systems, it often introduced more performance issues than it solved.
Dozens of threads on AskTom and other Oracle forums echo the same sentiment: users asking how to disable a feature meant to improve performance (even on Oracle 19c). In OLAP systems, the impact is less severe. Lower concurrency and more relaxed response time expectations make the unpredictability less damaging. But even OLAP workloads are shifting toward purpose-built architectures with columnar storage. Optimizing for them inside a traditional row-store RDBMS like Oracle or Postgres makes little sense.
So when I hear a database boasts runtime auto-tuning of query plans, it doesn’t inspire confidence. For mission-critical workloads, predictability and control matter far more than runtime guesswork.
