TESTIMONIALS

“Received the latest edition of Professional Security Magazine, once again a very enjoyable magazine to read, interesting content keeps me reading from front to back. Keep up the good work on such an informative magazine.”

Graham Penn
ALL TESTIMONIALS
FIND A BUSINESS

Would you like your business to be added to this list?

ADD LISTING
FEATURED COMPANY
Announcement

Is Your Infrastructure Ready?

by Roy

PostgreSQL’s Momentum Continues to Grow with the Proliferation of AI: Is Your Infrastructure Ready?

By Don Boxley, CEO and Co-Founder, DH2i (www.dh2i.com)

I’ve been watching… PostgreSQL is gaining ground in the enterprise, and I can’t say I’m surprised. Companies for quite some time have been paying a lot of money for database technology, sometimes because they needed certain capabilities, but sometimes because that’s simply what they’ve always done. It’s familiar. It’s supported. Nobody wants to be the person who suggests changing something that’s been running the business for twenty years.

But the economics are changing. And so are the workloads.

PostgreSQL gives organizations a pretty compelling alternative. It’s open source. It’s mature. It handles serious workloads. And, you’re not paying database licensing fees every time you need more processing power. You can run it on-prem, in the cloud, on virtual machines (VMs), or in Kubernetes. You’re not married to one infrastructure provider.

That’s real freedom. And with everything happening around AI, it’s becoming even more valuable. I think however, there is something that we’re overlooking in all the excitement about PostgreSQL. We’re spending a lot of time talking about what it can do… Not nearly enough time talking about what happens when it can’t.

That’s a conversation worth having before PostgreSQL becomes the database your business absolutely cannot afford to lose, right?

AI Changes the Conversation

Here’s what I find interesting about PostgreSQL and AI. Everyone wants to talk about the models. How intelligent they are. How quickly they respond. How much computing power they need. And yes, those things matter. But an AI model that’s supposed to help run your business needs access to your business information. Otherwise, what exactly is it working with?

Take a customer service application. Somebody asks about an order, a warranty, or a technical issue – then the AI needs to find information that’s relevant to that specific question. And it needs to find it quickly. That’s where PostgreSQL has become particularly useful.

With extensions like pgvector, PostgreSQL can store and search vector embeddings. Essentially, that’s numerical representations that help apps find information based on meaning instead of exact word matches.

This makes it useful for retrieval-augmented generation (RAG). That’s where an application searches for relevant information and provides that context to a language model before it generates a response. What’s really nice is that companies don’t necessarily need to build an entirely separate database infrastructure just to support AI. PostgreSQL can manage traditional business records while also supporting the vector searches that many AI applications depend on.

And there’s a financial advantage here too. The need for computing resources grows as AI workloads grow. You can add CPU cores and memory without watching your database licensing bill grow right along with them, with PostgreSQL.

That’s a big deal. But there’s a catch. All that impressive AI technology is still depending on a database being available when it needs it. But if that database goes down, you’ve got a problem that no amount of AI intelligence is going to fix.

The Part Nobody Likes to Talk About

PostgreSQL is an excellent database. But excellent doesn’t mean complete.

PostgreSQL gives you replication. You can maintain standby databases. You have backup and recovery capabilities. There’s a lot there to work with. What it doesn’t give you out of the box is a complete, automated high availability solution. And there’s a pretty big difference.

Say your primary database goes down in the middle of the day. You’ve got a standby, so you should be fine, right? Well, maybe

Does something recognize that the primary has actually failed? Can it tell the difference between a database failure and a temporary network problem? Who decides which standby takes over? How does the application know where to reconnect?

And what happens if the original primary comes back online and still thinks it’s in charge?

Now you’ve potentially got two databases accepting writes, which is a much bigger mess than the original outage.

These are solvable problems. But somebody has to solve them. That’s what concerns me about organizations moving business-critical workloads to PostgreSQL without fully appreciating the infrastructure work that comes with that decision.

You’ve eliminated one expense, the database license. Great. But if you now need to assemble, integrate, monitor, and maintain multiple technologies to get the availability your business requires, you haven’t necessarily made your life easier.

You may have just moved the complexity somewhere else.

And for organizations running across on-premises infrastructure, multiple clouds, and Kubernetes, that complexity can add up pretty quickly.

Kubernetes Doesn’t Magically Fix It Either

There’s a tendency in our industry to believe the latest infrastructure technology will somehow take care of problems we’ve been dealing with for decades. Kubernetes is a good example.

It’s a terrific platform for orchestrating containerized applications. But restarting a failed container and recovering a database safely are two very different things.

A PostgreSQL database can be running without being healthy. It might have stopped processing transactions. Replication might be falling behind. A network issue might have isolated it from the rest of the environment.

From the infrastructure’s perspective, everything could look fine. From the application’s perspective, nothing works. And that distinction matters. Especially when you’re talking about high availability. You need to:

  • Know what’s happening inside the database
  • Take the right failover decision, not just the fastest one
  • Keep applications connected and make sure you’re not sacrificing data integrity in the process

Then, there’s the question of how much data you’re willing to lose. Asynchronous replication can leave a gap between the primary and standby. Synchronous replication offers stronger protection for acknowledged transactions. But it brings its own performance and availability tradeoffs. There are good reasons to choose either approach; but those choices need to be deliberate.

You can’t just check the replication box and then declare the database protected and secure.

So Where Does That Leave Us?

I wouldn’t let any of this stop me from adopting PostgreSQL. Quite the opposite. But let’s go into it with our eyes wide open. If we’re putting PostgreSQL underneath an app that matters to the business, we want to know exactly what happens when something breaks. Not what the architecture diagram says should happen. What actually happens.

How quickly can we detect a failure? Can we automatically and safely move operations to another database? How do we prevent split-brain situations? Will applications reconnect without somebody scrambling to change configurations? And can we manage all of this without building a complicated collection of scripts and processes that only a handful of people understand?

I would be asking those questions.

Database-level monitoring, reliable failover orchestration, sound quorum and fencing protections, and tested recovery procedures – I would certainly want all of that. And I’d want an approach that works consistently across the environments where my applications actually run.

Because the whole attraction of PostgreSQL is flexibility. It would be a shame to gain that flexibility at the database level and then lose it because the availability infrastructure is too complicated or too tightly tied to one environment.

This becomes even more important with AI. We’ve already seen how quickly companies can go from experimenting with AI to depending on it. The underlying database is no longer just another piece of infrastructure, once employees and customers start relying on those applications.

It’s part of the business. And businesses don’t care why something went down. They care that it went down.

The Next PostgreSQL Challenge

I believe PostgreSQL is going to keep gaining ground. The technology is there. The economics make sense. And AI is giving organizations even more reasons to take it seriously. But I also think we’re going to see some companies learn the hard way that adopting a great database and running a highly available database are not the same thing.

These problems aren’t new, that’s the good news. Addressing database failures, managing failover, protecting/securing data, and maintaining application connectivity – we know how to do that. The challenge is making all of that work together without piling on unnecessary complexity.

Because if PostgreSQL is going to take on a bigger role in the enterprise, the infrastructure supporting it has to grow  right along with it. Otherwise, we’re going to have a lot of companies celebrating what they saved on database licenses right up until the day their applications go down.

That’s a pretty expensive way to learn about high availability.

###

 

Related News