Conversation about Autonomous PostgreSQL: Magnus Hagander

Last week, Magnus Hagander joined me for a conversation about autonomous databases, AI, and PostgreSQL. Magnus is a key player in PostgreSQL; as President of PostgreSQL Europe, member of the Core Team, and one of the people who ported PostgreSQL to Microsoft Windows, he has left his footprint in the software and in the open-source community.

Watch the full interview above — here are a few of my notes and key takeaways.

We started with our key topic: 'Autonomous Database', as a label for AI that either augments or replaces the DBA in many tasks, such as index tuning, vacuum configuration, or even schema design. Magnus sees the biggest opportunities for AI in database tuning, maybe less so in schema design, even though it could also play a role in query building.

Then we dove into how one should draw the line between recommendations to the DBA, for example, identifying a column that would benefit from a B-tree index and explaining why, and the AI agent automatically acting on behalf of the DBA, for example, by adjusting random_page_cost.

Magnus sees this as a debated point and that the answer may depend on the maturity and skill set of the organization. Traditional DBAs with a lot of PostgreSQL experience often say that the AI should not change anything. Developers, who often have less experience with databases, SQL, or PostgreSQL, may just want things to work magically. For organizations with fewer in-house DBA skills, AI-based tuning may allow them to push back the point at which they have to hire a DBA.

According to Magnus, AI will influence the nature of what DBAs focus on, similar to what happened in the last 30 years when databases evolved from Indexed Sequential Access Methods (ISAM) to Structured Query Language (SQL). In ISAM, DBAs needed to know how to access the data; in SQL, the database takes care of that, and the DBA only describes what data they want to access. Magnus expects that database management systems will continue to become more intelligent as AI and ML evolve, and take on more of those tasks where the traditional DBA needed to have 'How To' knowledge.

We also talked about the fact that PostgreSQL itself is making a lot of progress; for example, the planner is becoming smarter and smarter. One topic was how that will impact the role of the database tuner, whether they be an AI agent or a human DBA.

Magnus' perspective on this is that in an ideal world, tuning would not be needed — the database should just do it optimally. But that's a very long-term perspective. In reality, this kind of intelligent tuning will probably start outside of the database, and then, in true PostgreSQL fashion, some of it will be absorbed either as full-fledged functionality or as hooks that will make it easier for external AI/ML tools to work with PostgreSQL.

We ended our discussion with a look forward to and a forecast of what we should expect in the near future in the AI and PostgreSQL space. Magnus expects progress in both areas (AI for the analyst and AI for DBA). LLMs may be a better match for the analysts' tasks as LLMs are really good at helping analysts write complex queries, such as integrating multiple window functions into a single query.

"Optimizing the DBA workload may need some sort of AI or ML technology, not just LLMs"

Optimizing the DBA workload may need some sort of AI or ML technology, not just LLMs. One thing AI could do really well is help DBAs continually monitor, observe patterns, identify changes, and adjust the database in a continuous feedback loop.

This was the third in a series of interviews about Autonomous PostgreSQL. The first interview was with Luigi Nardi, CEO and founder of DBtune, and the second was with Bruce Momjian, member of the PostgreSQL Core Team.

If you have any feedback, questions, or suggestions for future interviews, please contact us at autonomouspostgresql@dbtune.com.