Overview of The future is Postgres (Interview)
Changelog Media talks with Craig Kerstiens about why Postgres feels like it’s entering a new golden age. The conversation covers the rise of Postgres as a default database choice, the power of extensions, the upcoming Postgres 19 release, and how new projects like PG Lake are expanding Postgres into data-lake and analytics territory. They also dig into the reality of maintaining a 30-year-old open source project, the small number of core committers behind it, and how AI-generated code and PRs are changing the ecosystem.
Key Themes and Takeaways
Postgres is becoming the default “serious” database
- Craig reflects on how Postgres went from the underdog to the obvious choice for many developers and companies.
- He points out that its success comes not just from hype, but from a long track record of being reliable, extensible, and well-designed.
- The interview frames Postgres as “boring in the best way”: dependable infrastructure that developers can bet on.
Extensions are Postgres’ superpower
- A major theme is that Postgres thrives because the core stays stable and predictable, while new capabilities are added through extensions.
- This lets Postgres act more like a platform than just a database.
- Examples discussed include:
- PostGIS for geospatial data
- text search extensions like BM25-style search
- vector storage
- time-series support
- PG Lake, which pushes into analytics and lakehouse workflows
Postgres is increasingly bridging OLTP and OLAP
- The episode explains the traditional divide:
- OLTP = transactional systems, row-based, fast inserts/updates
- OLAP = analytical systems, column-based, batch-oriented
- Craig argues that moving data between these systems has always been painful, expensive, and error-prone.
- The new wave of Postgres tools aims to reduce or eliminate that gap.
PG Lake and the Analytics Layer
What PG Lake is
- PG Lake is described as an open source extension ecosystem that brings:
- an Iceberg catalog
- DuckDB embedded inside Postgres
- support for Parquet files
- data in S3
- The goal is to let users query and manage analytics-style data directly from Postgres.
Why it matters
- PG Lake blurs the line between a transactional database and an analytics engine.
- It allows teams to:
- insert data normally into Postgres
- store it in open formats
- query it from Postgres or other engines that understand Iceberg
- Craig emphasizes that this can dramatically simplify data pipelines, reduce ops burden, and cut down on duplicated infrastructure.
DuckDB and Parquet, in plain English
- DuckDB is presented as “SQLite for analytics”: lightweight, embedded, and excellent for local/interactive analytical work.
- Parquet is explained as an open, columnar file format optimized for analytics.
- Iceberg is the metadata/catalog layer that makes those files behave more like a live, queryable dataset.
Postgres 19: A Big Release
Why Craig is excited
Craig says Postgres 19 feels like one of the most interesting releases in years. Some of the standout features discussed:
-
Repack concurrently
- Helps reduce table bloat without taking harsh locks
- Makes an advanced operational task much easier and safer
-
Query plan pinning
- Lets operators keep a known-good execution plan instead of being surprised by planner changes
- Especially useful in mature, high-stakes production environments
-
GROUP BY ALL
- A quality-of-life improvement for SQL writing
- Reduces repetitive column listing
-
Improved logical replication and autovacuum behavior
- More reliability and operational smoothness
-
Graph features in SQL
- Craig notes that graph capabilities are coming into Postgres through SQL standards work
The big picture
- Postgres 19 is framed as a release that doesn’t just add “nice-to-haves.”
- It improves the experience for people running serious production systems, especially those dealing with bloat, replication, or query stability.
Open Source, Community, and the Reality of Maintaining Postgres
A small group maintains a huge system
- Craig stresses that Postgres is powered by a surprisingly small group of maintainers and committers.
- The project still operates in a very old-school way in some respects:
- patches sent to mailing lists
- commitfests
- public review and discussion
- That process has worked remarkably well for decades.
Committers and contribution
- He highlights that people can contribute even if they don’t write C:
- testing beta releases
- reviewing patches
- helping on the users’ mailing list
- participating in commitfests
- He encourages developers to engage with the community rather than treating Postgres as a black box.
AI is creating new pressure
- One major concern is the flood of AI-generated PRs and contributions.
- With a relatively small number of committers, the community now has to filter more submissions, many of which may be low quality or incorrect.
- Craig sees this as a real challenge for open source maintainers, especially when correctness matters as much as it does in a database.
Security, Responsible Disclosure, and Trust
Open source doesn’t mean “everything public immediately”
- The interview spends time on how security issues are handled in a mature open source ecosystem.
- Craig explains that responsible disclosure often requires private coordination before public release.
- This is especially important for vulnerabilities that could expose data or enable remote compromise.
Why AI raises the stakes
- He argues that AI makes security more urgent:
- more people can find vulnerabilities
- bad actors may act faster
- maintainers need better tooling and coordination
- Despite the challenges, he says Postgres has proven itself as a trusted, battle-tested system.
Snowflake, Mirroring, and the Bigger Postgres Strategy
Snowflake’s connection to Postgres
- Craig’s work at Snowflake is part of a larger effort to integrate transactional and analytical data flows.
- One example discussed is database mirroring, where Postgres data can be synced into Snowflake in near real time.
Why this matters
- It reduces ETL friction.
- It gives teams one consistent data foundation for:
- operational workloads
- analytics
- AI/ML use cases
- The broader vision is a data platform where Postgres remains the operational core while other tools can build around it.
Notable Ideas and Quotes
- “Postgres is the boring but provocative topic.”
- “The best database is the one I don’t think about.”
- “Extensions are where Postgres thrives.”
- “PG Lake is like a Swiss Army knife for data workflows.”
- “The database should be boring; the things built on top of it should be exciting.”
Practical Recommendations / Action Items
For developers
- Try Postgres 19 when it becomes available.
- Explore extensions instead of introducing a separate database too early.
- Consider Postgres as a platform, not just a storage engine.
For curious learners
- Join or read:
- the pgsql-hackers mailing list
- the pgsql-users mailing list
- the Postgres commitfest process
- Test beta releases and help validate new features.
For teams
- Evaluate whether your ETL/analytics stack is overly complex.
- Look at whether Postgres plus extensions like PG Lake could simplify your workflow.
- If you run production Postgres, pay attention to bloat, replication, and query plan stability.
Final Thoughts
This episode is part technical deep dive, part love letter to Postgres. Craig Kerstiens makes the case that Postgres is not just surviving—it’s evolving into a broader data platform, one that can absorb new workloads without losing the qualities that made it successful in the first place. The core message: Postgres is still boring in the best possible way, but the ecosystem around it is getting far more powerful.