Scaling PostgreSQL at OpenAI: Lessons in Reliability, Efficiency, and Innovation
At POSETTE: An Event for Postgres 2025, Bohan Zhang of OpenAI delivered a compelling talk on how OpenAI has scaled Azure Database for PostgreSQL- Flexible Server to meet the demands of one of the world’s most advanced AI platforms running at planetary scale. The Postgres team at Microsoft has partnered deeply with OpenAI for years to enhance the service to meet their performance, scale, and availability requirements, and it is great to see how OpenAI is now deploying and depending on Flexible Server as a core component of ChatGPT. Hearing firsthand about their challenges and breakthroughs is a reminder of what’s possible when innovation meets real-world needs. This blog post captures the key insights from Bohan’s POSETTE talk, paired with how Azure’s cloud platform supports innovation at scale. PostgreSQL at the Heart of OpenAI As Bohan shared during his talk, PostgreSQL is the backbone of OpenAI’s most critical systems. Because PostgreSQL plays a critical role in powering services like ChatGPT, Open AI has prioritized making it more resilient and scalable to avoid any disruptions. That’s why OpenAI has invested deeply in optimizing PostgreSQL for reliability and scale. Why Azure Database for PostgreSQL? OpenAI has long operated PostgreSQL on Azure, initially using a single primary instance without sharding. This architecture worked well—until write scalability limits emerged. Azure’s managed PostgreSQL service provides the flexibility to scale read replicas, optimize performance, and maintain high availability to provide global low latency reads without the burden of managing infrastructure. This is why we designed Azure Database for PostgreSQL to support precisely these kinds of high-scale, mission-critical workloads, and OpenAI’s use case is a powerful validation of that vision. Tackling Write Bottlenecks PostgreSQL’s MVCC (Multi-Version Concurrency Control) design presents challenges for write-heavy workloads—such as index bloat, autovacuum tuning complexity, and version churn. OpenAI addressed this by: Reducing unnecessary writes at the application level Using lazy writes and controlled backfills to smooth spikes Migrating extreme write-heavy workloads with natural sharding keys to other systems. These strategies allowed OpenAI to preserve PostgreSQL’s strengths while mitigating its limitations. Optimizing Read-Heavy Workloads With writes offloaded, OpenAI focused on scaling read-heavy workloads. Key optimizations included: Offloading read queries to replicas Avoiding long-running queries and expensive multi-way join queries Using PgBouncer for connection pooling, reducing latency from 50ms to under 5ms Categorizing requests by priority and assigning dedicated read replicas to high-priority traffic As Bohan noted, “After all the optimization we did, we are super happy with Postgres right now for our read-heavy workloads.” Schema Governance and Resilience OpenAI also implemented strict schema governance to avoid full table rewrites and production disruptions. Only lightweight schema changes are allowed, and long-running queries are monitored to prevent them from blocking migrations. To ensure resilience, we categorized requests by priority and implemented multi-level rate limiting—at the application, connection, and query digest levels. This helped prevent resource exhaustion and service degradation. Takeaway OpenAI’s journey is a masterclass in how to operate PostgreSQL at hyper-scale. By offloading writes, scaling read replicas, and enforcing strict schema governance, OpenAI demonstrated PostgreSQL on Azure meets the demands of cutting-edge AI systems. It also reinforces the value of Azure’s managed database services in enabling teams to focus on innovation rather than infrastructure. We’re proud of the work we’ve done to co-innovate with OpenAI and excited to see how other organizations can apply these lessons to their own PostgreSQL deployments. Check out the on-demand talk “Scaling Postgres to the next level at OpenAI” and many more PostgreSQL community sessions from POSETTE.CFP talk proposal ideas for POSETTE: An Event for Postgres 2025
Some of you have been asking for advice about what to submit to the CFP for POSETTE: An Event for Postgres 2025. So this post aims to give you ideas that might help you submit a talk proposal (or 2, or 3) before the upcoming CFP deadline. If you’re not yet familiar with this conference, POSETTE: An Event for Postgres 2025 is a free & virtual developer event now in its 4th year, organized by the Postgres team at Microsoft. I love the virtual aspect of POSETTE because the conference talks are so accessible—for both speakers and attendees. If you’re a speaker, you don’t need travel budget $$—and you don’t have to leave home. Also, the talk you’ve poured all that energy into is not limited to the people in the room, and has the potential to reach so many more people. If you’re an attendee, well, all you need is an internet connection The CFP for POSETTE: An Event for Postgres will be open until Sunday Feb 9th at 11:59pm PST. So as of the publication date of this blog post, you still have time to submit a CFP proposal (or 2, or 3, or 4)—and to remind your Postgres teammates and friends of the speaking opportunity. If you have a Postgres experience, success story, failure, best practice, “how-to”, collection of tips, lesson about something that's new, or deep dive to share—not just about the core of Postgres, but about anything in the Postgres ecosystem, including extensions, and tooling, and monitoring—maybe you should consider submitting a talk proposal to the CFP for POSETTE. If you’re not sure about whether to give a conference talk, there are a boatload of reasons why you should. And there’s also a podcast episode with Álvaro Herrera, Boriss Mejías, and Pino de Candia that makes the case for why giving conference talks matters. For inspiration, you can also take a look at the playlist of POSETTE 2024 talks. And if you’re looking for even more CFP ideas, you’ve come to the right place! Read on… Ideas for talks you might propose in the POSETTE CFP On the CFP page there is a list of possible talk titles (screenshot below) you might submit—these are good ideas, although the list is by no means exhaustive, and we welcome talk proposals that are not on this list. Figure 1: POSETTE CFP talk topics taken from the CFP page on PosetteConf.com On Telegram the other day, when answering the question “Do you have any ideas of what I should submit?”, I found myself suggesting different TYPES of talks. Not specific ideas and talk titles, but rather I framed the different categories. So I decided to share these different “types” and “classes” of talks with all of you, in the hopes this might gives you a good talk proposal idea. First you need to pick your audience: Before you think about what type of talk to give, remember that the POSETTE team is focused on serving the needs of both the USER community—as well as the Postgres contributor & hacker communities. That means first you need to decide on your audience. Are you giving a talk for PostgreSQL users, or Azure Database for PostgreSQL customers, or the PostgreSQL contributor community? All are good choices. Then you need to decide: what do you want to accomplish with your talk? Do you want to skill up the Postgres hacker community?: If you want to help skill-up the developer/contributor community, maybe pick a part of Postgres that new contributors often ask a lot of questions about, get stuck on, need help with, etc—and give a “tour” of its mechanics, starting with the basics. Do you want to help grow the Postgres community?: If you want to help grow the Postgres community of contributors and developers, you could propose a talk that would motivate tomorrow's developers/contributors to get involved in the project. Imagine you were going to a university to give a talk about "why work on Postgres"… what would you say? And how would you entice people to work on Postgres? What pain points would you challenge them with? What benefits would you share from your own Postgres experience that might inspire these developers to think seriously about Postgres as a career path? You could also shine a light on the different ways people can (and do!) contribute to the Postgres community: from mentoring to translations to organizing conferences to podcasts to speaking at conferences to publishing PostgreSQL Person of the Week. Do you want to share your expertise with Postgres users?: If you want your talk to benefit users, maybe pick an area that you are already expert in (or want an excuse to dig into and learn about?) and create a Beginners Guide for it? Or Advanced Tips for it? Or Surprising Benefits of? Or Things People Might Not Know? Especially if there is a part of Postgres you feel like people sometimes mis-use, or don't take enough advantage of.... Do you want to share your customer experiences with Azure Database for PostgreSQL, or Postgres more generally?: Maybe you have a wild success story you think others will benefit from. Or you want to share a problem you had and how you used Postgres to solve it? People love customer stories. Do you want to shine a light on the broader Postgres ecosystem?: If you want to target users with your talk, don’t limit yourself the Postgres core. There is a rich ecosystem that surrounds Postgres and people need to understand the ecosystem, too. So maybe there are tools or Postgres extensions or forks or startups that you can give a useful talk about? Do you want to help experts in other database technologies learn about Postgres?: If you have expertise in other databases as well as Postgres, maybe you can help people who who are skilled in running workloads on other databases and are looking to skill up on Postgres—by helping them understand what’s similar, and what’s different. As if you’re giving them a dictionary to translate from their familiar database to Postgres, and vice versa. There are so many more possibilities: Often I look at the schedule from previous years to look for inspiration (and to make sure that my talk proposal is not a duplicate of a talk that’s already been given.) And I think about pain points, things people get confused about, or questions that come up a lot. Another thing to keep in mind: how can you help your story to "stick"? Can you make it entertaining? How do you share your story in a way that keeps people watching (versus looking at their phone instead?) Key things to know about POSETTE: An Event for Postgres 2025 CFP deadline: The CFP for POSETTE will close on Sunday, Feb 9th 2025 @ 11:59pm Pacific Time (PST) No travel required: free & virtual developer event Length of talks: 25 minutes/session Language: All talks will be in English Talks will be pre-recorded: All talks will be pre-recorded by the POSETTE team during the weeks of Apr 28th and May 5th (with accepted speakers presenting remotely) When is the event?: Jun 10-12, 2025 Format of the virtual event: All pre-recorded talks will be livestreamed in one of 4 unique livestreams on Jun 10-12, 2025—all with parallel live text chats on Discord. Two of the livestreams will be in Americas-friendly times of day (8:00am-2:00pm PDT) and two of the livestreams will be in EMEA-friendly times of day (8:00am-2:00pm CEST). All talks will be published online after the event is over. More info about the CFP: All the details, including key dates and how to submit on Sessionize, are spelled out on the CFP page for POSETTE 2025 Code-of-conduct: You can find the Code of Conduct for POSETTE online. Please help us to provide a respectful, friendly, and professional experience for everybody involved in this virtual conference. Figure 2: The CFP is open for POSETTE: An Event for Postgres 2025 until Sunday Feb 9th at 11:59pm PST. What Postgres story do you want to share?Ultimate Guide to POSETTE: An Event for Postgres, 2024 edition
This guide will help you navigate all 42 talks at the 3rd annual POSETTE: An Event for Postgres 2024, happening Jun 11-13. POSETTE is a free & virtual developer event organized by the Postgres team at Microsoft that is chock-full of Postgres database content.
