Latest / The CTO Podcast with Fexingo: Technical Leadership, Architecture, and Engineering Org / How Postgres Powers 40 Percent of New Cloud Databases
Transcript
- Lucas: There's a quiet takeover happening in the database world, and it's not coming from a startup with a billion-dollar valuation. It's coming from a project that first shipped in 1996. Luna: You're talking about PostgreSQL. And 'quiet takeover' might be underselling it — last I checked, something like 40 percent of new cloud database instances are running Postgres. Lucas: Yeah, 40 percent on AWS alone. And that number keeps climbing. What's interesting is that Postgres wasn't designed for the cloud. It was designed as a research project at Berkeley. But its architecture turned out to be incredibly well-suited for modern workloads. Luna: So what specifically about Postgres's design allowed it to become the default? I mean, MySQL was the open-source darling for years. Lucas: Right. And MySQL still has a huge install base. But Postgres pulled ahead on a few key architectural decisions. The first is its extension system. You can dynamically load modules that add entirely new capabilities — think geospatial with PostGIS, full-text search, key-value store, even machine learning. That flexibility meant Postgres could adapt to new use cases without a major rewrite. Luna: So AWS and Google Cloud basically said, instead of building their own proprietary database for every new workload, let's just add an extension to Postgres. Lucas: Exactly. Aurora Postgres, Cloud SQL, Azure Database for Postgres — they're all managed versions of the same core. And because the extension API is stable, the cloud providers could compete on operational features — auto-scaling, backups, replication — while keeping the SQL dialect and tooling identical. For developers, that lowers switching costs. Luna: But let's talk about the concurrency model. Postgres uses MVCC — multi-version concurrency control. That's not unique, but how Postgres implements it is. Lucas: Yeah. Postgres stores multiple versions of each row so readers never block writers and writers never block readers. In a cloud environment where you have mixed OLTP and analytics hitting the same database, that's huge. MySQL historically used table-level locking with MyISAM, and even InnoDB's row-level locking had different isolation semantics. Postgres's snapshot isolation — repeatable read — gave developers a consistent view of data without sacrificing write throughput. Luna: And that's why a company like Instacart uses Postgres for real-time inventory. They have thousands of shoppers querying available items while behind the scenes, inventory is being updated constantly. If readers blocked writers, the app would freeze. Lucas: Right. And Instacart runs Postgres on Amazon RDS for exactly that reason. They also use extensions like pg_stat_statements to monitor query performance in production. It's a pattern you see across a lot of modern tech companies. But the architectural story doesn't stop at the engine itself — there's also the sharding layer. Luna: You mean Citus, which open-sourced horizontal scaling for Postgres. Microsoft acquired them and turned it into Azure's Hyperscale offering. Lucas: Exactly. Citus distributes tables across multiple Postgres nodes and presents a single SQL interface. That solved the biggest historical criticism of Postgres: that it couldn't scale horizontally. With Citus, you can shard by tenant for multi-tenant SaaS apps, or by time for time-series data. And because it's an extension, you don't have to rewrite your application. Luna: Which is the opposite of what NoSQL databases required you to do — redesign your data model around denormalization. Lucas: Right. That's a big reason why we're seeing migrations from MongoDB and Cassandra back to Postgres. The extension ecosystem gives you document storage via JSONB, which supports indexing and querying similar to MongoDB. So developers get the flexibility of NoSQL with the consistency and joins of SQL. Luna: I want to talk about the community and governance model, because I think that's also architectural in a sense. Postgres has a very structured development process — the PostgreSQL Global Development Group releases major versions every year, with a clear commit fest process. Lucas: It's a very boring, reliable process. And that's actually its strength. No single company controls the roadmap. So when a cloud provider like AWS contributes a feature, it goes into the core and everyone gets it. Compare that to MongoDB, where the company drives the direction, or even MySQL, where Oracle has complete control. That neutrality is part of the architectural trust. Luna: It also means that if you build on Postgres, you're not locked into a single vendor's API. You can run it on-prem, on AWS, on GCP, on a Raspberry Pi — the same SQL works everywhere. Lucas: That portability is increasingly important as companies adopt multi-cloud or edge computing. And it's one reason why the database industry is consolidating around Postgres. I think we're going to see more proprietary databases either adopt Postgres as a storage layer or build on top of it. Luna: Speaking of which, we've been talking about Postgres as a technology, but behind the scenes, this show itself is a kind of infrastructure. We deliberately keep it ad-free, and that's a choice that relies on the support of listeners who find value in these conversations. Lucas: It's true. We spend a lot of time digging into the architecture decisions that actually matter for people building products. If you find these episodes useful, one way to keep them going without any ads is to toss a couple bucks at the show on Buy Me a Coffee. Luna: Yeah, it's buy me a coffee dot com slash fexingo. No pressure, just an option if you want to support the ad-free model. Lucas: And that support goes right back into research and production. So thanks to everyone who's already done that. Now, back to Postgres — let's talk about the migration path for teams that are still on Oracle or SQL Server. Luna: That's been a huge trend in the last five years. Large enterprises moving off expensive proprietary licenses to open-source Postgres. Lucas: Right, and the migration isn't trivial, but tools like pgloader and the Oracle foreign data wrapper make it much smoother. The bigger challenge is usually around stored procedures and PL/SQL syntax differences. But Postgres's PL/pgSQL is close enough that most logic can be ported with minor changes. Luna: And the cost savings can be enormous. A large bank I talked to saved over a million dollars a year in licensing by moving from Oracle to Postgres on Amazon RDS. Lucas: And they didn't lose any functionality. In fact, they gained features like native JSONB support and better replication. That's the story we're seeing across industries. Postgres isn't just an alternative anymore — it's becoming the default choice for new projects. Luna: So what's next for Postgres? There's talk about native vector support for AI workloads. Lucas: Yeah, pgvector is already an extension, but there's a push to include vector indexing in the core. That would let Postgres serve as the database for semantic search and embedding-based retrieval without needing a separate vector store. It's another example of the extension architecture absorbing new paradigms. Luna: And the community is working on better parallel query execution and improved partitioning. The pace of innovation hasn't slowed down despite the project being nearly thirty years old. Lucas: That's the beauty of a well-designed core with a strong extension mechanism. The architecture enables evolution without fragmentation. And that's the lesson for any platform builder — design for extensibility first, and the ecosystem will take care of the rest. Luna: I think that's a great place to wrap up. Next episode we're going to look at how GitHub Actions scaled to millions of daily builds. Lucas: Yeah, that's going to be a good one. See you then.