About PGConf.dev

PGConf.dev is an annual developer event focused entirely on contributing to the PostgreSQL ecosystem, including core software development and community building. It serves as a primary hub for PostgreSQL hackers, maintainers, and ecosystem developers to meet, collaborate, and share knowledge.

This year’s conference was hosted once again in Vancouver, which also happens to be my hometown :D. On top of that, 2026 also marks PostgreSQL’s 30th anniversary, giving this year’s conference an even more meaningful atmosphere for long-time contributors and community members around the world.

This is one of my favorite PostgreSQL conferences because it gives people an opportunity to truly meet, learn from, and interact with the brilliant minds behind PostgreSQL. Many of the names we usually only see in hacker mailing list discussions, patch reviews and commit messages suddenly become real people standing right beside you, discussing ideas and working together to make PostgreSQL better.

It feels like the PostgreSQL hacker discussions have come to life in the real world. The conversations are real, the passion is real, and the community spirit is everywhere.

A huge thank you once again to all the organizers, volunteers, speakers, and sponsors who invested countless hours to make this conference possible. This blog is a personal summary of my own experience attending pgconf.dev 2026, and for those who could not attend, I hope it can give you at least a small glimpse of what it feels like to be part of this incredible PostgreSQL community.

<WARNING> Long Post Ahead!

Tuesday No Longer Feels Like a “Pre-Conference” Day

Unlike the “Tuesdays” from previous pgconf.dev conferences, this year’s Tuesday schedule is packed with an impressive variety of engaging sessions, including round-table discussions, interviews, keynote-style talks, brainstorming sessions, and community-led discussions focused on specific PostgreSQL topics and ecosystem challenges.

The program feels much more than a traditional pre-conference day; it is essentially an additional full day of high-value content and collaboration. This expanded format makes this year’s PGConf.dev significantly more diverse and interactive in both content and participation. It is no surprise that tickets sold out so quickly.

How immutability challenges extension packaging and distribution

(David E. Wheeler)

At the Extensions and Ecosystem Summit, a discussion session led primarily by David Wheeler explored how immutable infrastructure is beginning to reshape the PostgreSQL extension ecosystem. The conversation focused on the growing shift toward immutable PostgreSQL deployments, where extensions are baked directly into container images rather than dynamically installed onto running systems. Participants discussed how immutability can help reduce the risk of malicious or “poisoned” extensions being introduced into production environments, while also emphasizing that immutability and verification are not necessarily the same thing. The session also raised questions about whether extensions should be “replaced” as part of rebuilding images rather than traditionally installed and upgraded in-place, leading to different operational expectations in cloud-native environments such as CloudNativePG. Additional challenges discussed included preload extensions, SQL object versioning, shared library compatibility, and the growing operational complexity these issues create for organizations managing large fleets of PostgreSQL instances.

The Road to Enterprise-Grade Logical Replication: Feedback and Roadmap Brainstorming

(Amit Kapila, Dilip Kumar)

A brainstorming session led by Amit Kapila and Dilip Kumar focused on the future roadmap of PostgreSQL logical replication and the remaining gaps before it can be considered truly enterprise-grade. Due to limited session time, much of the discussion centered on rapidly listing missing features and pain points on the whiteboard rather than diving deeply into individual topics. DDL replication received the most discussion, with ideas ranging from schema diffing and dynamic ALTER generation to event-trigger and WAL-based approaches. There was also discussion around large object replication, though one attendee commented that there are currently no major plans to further enhance PostgreSQL large objects themselves, suggesting that improving large object replication may not be an immediate priority. Other topics briefly mentioned included major-version upgrades, partition replication behavior, table renaming during replication, failover, transaction dependency handling and centralized decoding (replace logical decoding plugin?).

Stop Debating, Start Implementing: Unifying TDE Efforts for Postgres

(Alistair Turner, Zsolt Parragi)

A session led by Alistair Turner and Zsolt Parragi revisited the long-running discussion around Transparent Data Encryption (TDE) in PostgreSQL. Much of the technical discussion involved Peter Eisentraut from EDB, who shared practical concerns and lessons learned from EDB’s TDE implementation efforts. Topics included authenticated encryption challenges, WAL-level encryption design, key management complexity, compliance and certification requirements, extension-vs-core implementation debates, and concerns around using LSNs as IVs for encryption. Multi-tenant encryption and encryption of internal components such as SLRU were also briefly discussed. Despite the session title encouraging the community to “stop debating,” the discussion once again highlighted how difficult it remains to reach consensus on a unified PostgreSQL TDE design after many years of ongoing debate.

Why is PostgreSQL Terrible?

(Christophe Pettus)

Christophe Pettus delivered a keynote-style talk titled “Why is PostgreSQL Terrible?”, which humorously but realistically highlighted many long-standing operational pain points in PostgreSQL. Topics included autovacuum tuning difficulties, replication lag management, fragile logical replication behavior, planner tuning complexity, lack of proper query hints and debugging tools, upgrade limitations caused by on-disk format compatibility, pooling challenges, and PostgreSQL’s fragmented HA story. Christophe also pointed out that PostgreSQL observability still relies heavily on polling statistics views and analyzing text logs, both of which can become expensive or painful in production environments. Despite the provocative title, the talk was ultimately less about criticizing PostgreSQL itself and more about openly acknowledging the operational complexity that experienced PostgreSQL users continue to face at scale.

PostgreSQL at 30: Community Moments That Matter

(Valeria Kaplan)

Valeria Kaplan delivered a warm and reflective talk celebrating 30 years of the PostgreSQL community through the lens of conferences, meetups, and community events. Rather than focusing purely on technical milestones, the session highlighted how PostgreSQL events have helped shape the ecosystem itself — bringing together contributors, companies, organizers, volunteers, and new users from around the world. Valeria shared stories about the growth of PostgreSQL conferences, the importance of community participation, the role of the hackers mailing list, and how events often become the place where long-term collaborations, friendships, and even careers begin. The talk ultimately served as both a celebration of PostgreSQL’s community culture and a reminder that the project’s long-term success is deeply tied to the people and relationships behind it.

Unexpected successes & epic failures by PostgreSQL committers: A Roundtable

(Álvaro Herrera, Claire Giordano, Daniel Gustafsson, Greg Burd, Peter Eisentraut, Thomas Munro)

The final session of Tuesday was a lively and surprisingly personal roundtable featuring Álvaro Herrera, Daniel Gustafsson, Peter Eisentraut, and Thomas Munro, moderated by Claire Giordano and Greg Burd. Most speakers chose to begin with failures — reverted patches, stalled encryption efforts, painful OpenSSL work, or difficult upgrade-related bugs — reinforcing the idea that setbacks are a normal part of PostgreSQL development. One particularly memorable story came from Peter, who shared how PostgreSQL’s TAP test framework unexpectedly became one of his most successful contributions. Originally inspired by Debian packaging test infrastructure, the framework ended up being widely adopted and deeply appreciated across the PostgreSQL community. The session overall highlighted that in long-term open source development, even contributions that begin as small experiments or accidental ideas can eventually become foundational parts of the ecosystem.

Meet + Eat (Tuesday) + Social

As one of the Meet & Eat leaders — and also a Vancouver local with Chinese roots — it felt only natural for me to take the group to Kirin Restaurant, one of the most well-known high-end Chinese restaurants in downtown Vancouver.

I handled the ordering entirely in Mandarin, which was probably the moment when the group realized that something serious — and hopefully delicious — was coming. Fortunately, the food absolutely delivered. Dish after dish arrived at the table, and the group seemed genuinely impressed by both the quality and variety of authentic Chinese cuisine.

More importantly, it became another great example of what makes PostgreSQL conferences special. Beyond the technical talks, these shared meals create opportunities for people from different countries and backgrounds to connect in a much more personal and relaxed setting. By the end of the dinner, everyone was laughing, sharing stories, discussing PostgreSQL, and simply enjoying the experience together.

Special thanks also to Paul Ramsey for organizing and coordinating all of the Meet & Eat groups and for assigning me to such a fun and enjoyable group.

Official Opening on Wednesday

Here marks the official opening of PGConf.dev 2026. As always, the conference is filled with an incredible lineup of technical talks, community discussions, and social activities that bring PostgreSQL enthusiasts from around the world together. Returning after their successful debut at previous pgconf.dev events, both the poster sessions and the popular “community meet & eat” gatherings are back again this year, creating even more opportunities for collaboration and conversation across the PostgreSQL ecosystem.

This year’s conference is especially meaningful as it also celebrates PostgreSQL’s 30th anniversary. To mark the occasion, attendees received a special commemorative 30-year anniversary poster, with three signed editions autographed by PostgreSQL committers included as raffle prizes. Beyond the technical sessions, the event also features birthday cake cutting celebrations, a PostgreSQL community booth where attendees can casually discuss anything PostgreSQL-related, dedicated group photos, a themed 30-year anniversary photo booth with props, and even PostgreSQL karaoke events. Altogether, pgconf.dev 2026 feels more than just a technical conference; it feels like a global reunion and celebration of the PostgreSQL community itself.

Update on index prefetching

(Peter V Geoghegan, Tomas Vondra)

Peter Geoghegan and Tomas Vondra shared an update on the ongoing PostgreSQL index prefetching work — an effort aimed at reducing random I/O costs during index scans, similar to how sequential scans already benefit from prefetching today. The talk walked through several architectural attempts over the past few years, including approaches implemented purely in the index access method layer, then entirely in the executor, before eventually moving much of the logic into the heap table access method as a more practical design. A major challenge discussed was balancing aggressive prefetching with correctness and visibility checks, especially around visibility maps, xmin/xmax handling, buffer hit/miss behavior, and lock management across index pages. The speakers also briefly touched on global unique indexes and why they remain difficult long-term due to cross-partition checks, locking complexity, and maintenance costs such as REINDEX requirements after operations like VACUUM FULL. Overall, the session showed how deceptively difficult it is to optimize index scans inside PostgreSQL without breaking existing architectural assumptions.

Experimenting with a Global Index in PostgreSQL: Design, Implementation, and Challenges

(Dilip Kumar)

Dilip Kumar presented ongoing experimental work on bringing Global Index support to PostgreSQL partitioned tables — a feature frequently requested by users migrating from databases such as Oracle. The talk focused on how global indexes could enforce uniqueness across partitions, solving one of PostgreSQL’s current limitations where unique constraints must include partition keys. Dilip explained the proposed storage design, where global index tuples would contain both TID and partition identifiers to uniquely locate rows across partitions.

The session also explored the practical challenges that make global indexes extremely difficult to implement correctly inside PostgreSQL. While a global index may reduce the need to perform separate index scans on every partition, PostgreSQL would still often need to visit multiple partitions during query execution and maintenance operations. Several major unresolved problems were highlighted, including VACUUM scalability and dead tuple cleanup across partitions, DDL rewrite operations such as VACUUM FULL and ATTACH PARTITION, and complicated locking behavior caused by cross-partition uniqueness checks. Overall, the talk showed that although global indexes could significantly improve PostgreSQL’s partitioning story for enterprise users, the feature still faces major architectural and operational challenges before it could realistically become production-ready.

Table repacking, done right

(Álvaro Herrera)

Álvaro Herrera presented the upcoming REPACK CONCURRENTLY feature planned for PostgreSQL 19, aiming to finally provide a built-in solution for online table repacking and bloat removal. Traditionally, PostgreSQL users relied on VACUUM FULL to reclaim table space, but that operation rewrites the table while holding an exclusive lock, potentially blocking access for a long time on large production systems. Over the years, external tools such as pg_repack and pg_squeeze emerged to work around this limitation by copying live data into a new table while tracking concurrent changes through triggers or WAL-based mechanisms.

Álvaro’s talk explained how PostgreSQL is now moving toward integrating these ideas directly into core PostgreSQL with a native REPACK command. The feature essentially combines the goals of VACUUM FULL and CLUSTER, while allowing normal table activity to continue during most of the operation. The presentation also compared different approaches used by existing extensions — including trigger-based synchronization in pg_repack and WAL-driven tracking in pg_squeeze — and discussed why bringing such functionality into PostgreSQL core is important for long-term maintainability and operational simplicity. Overall, the session highlighted a major quality-of-life improvement for PostgreSQL DBAs, especially for large production systems struggling with table bloat and maintenance downtime.

Optimizing code in the hot path; with examples from tuple deformation

(David Rowley)

David Rowley gave a deep technical talk on optimizing PostgreSQL’s hot execution paths through improvements to “tuple deformation” — the internal process of extracting column values from tuples during query execution.

Using profiling tools, benchmarks, and CPU-level analysis, David showed how functions involved in tuple attribute extraction and NULL checks were consuming significant CPU time in scan-heavy workloads. The session walked through several low-level optimizations, including reducing unnecessary work, improving NULL bitmap checks, and helping the compiler generate more efficient machine code.

Benchmark results demonstrated noticeable performance gains, with some test cases running much faster than PostgreSQL 18. The talk highlighted how modern PostgreSQL optimization increasingly depends on detailed profiling, compiler behavior, cache efficiency, and micro-optimizations in critical code paths.

pg_plan_advice: Plan Stability and User Planner Control for PostgreSQL?

(Robert Haas)

Robert Haas introduced pg_plan_advice, a proposed PostgreSQL 19 contrib module designed to give users more control over query planning and improve plan stability.

The idea behind pg_plan_advice is to let PostgreSQL both generate and accept a small “advice language” that describes important planner decisions such as join order, hash joins, sequential scans, and gather merge operations. Users could capture a known-good execution plan and later ask PostgreSQL to reproduce similar planner decisions, or manually modify the advice string to influence future plans.

Robert explained that this aims to solve a long-standing PostgreSQL problem where planner decisions can unexpectedly change due to statistics variations or cost estimation errors, sometimes causing severe performance regressions. The system works by disabling planner paths that conflict with the supplied advice while allowing compatible paths to remain.

The talk also covered future possibilities, including cardinality hints, aggregation strategy control, planner experimentation, and even self-learning plan generation systems built on top of the new framework.

PostgreSQL Commitfest Metrics: A Quantitative Analysis

(Andreas Scherbaum, Jimmy Angelakos)

Andreas Scherbaum and Jimmy Angelakos presented a quantitative analysis of the PostgreSQL CommitFest process using data collected from the CommitFest application itself.

The talk showed how PostgreSQL development has scaled significantly over the years. In 2015, CommitFest handled 418 patches from 125 authors, while in 2025 it handled 885 patches from 272 authors — roughly doubling both patch volume and contributor count. However, reviewer and committer capacity has not scaled equally well. The speakers showed that 72.1% of “Needs Review” patches currently have no reviewer assigned, and 196 out of 331 active patches in the current CommitFest have zero reviewers.

The session also highlighted how contributor retention has declined over time. Around 73% of first-time patch authors in 2015 returned with another patch, but that number dropped to around 36–37% in recent years. Patch processing times are also long: median time to commit is 80 days, the 90th percentile is 256 days, and some patches remained open for over 1900 days.

Another important point was concentration of project work among a small group of committers. The top 5 committers are responsible for 55% of all commits, while the top 10 account for 74%. The presenters emphasized that PostgreSQL’s growth is stressing the current review and commit infrastructure, making reviewer participation and contributor onboarding increasingly important.

The Missing Link: Connecting Tens of Thousands of Chinese Users to the PostgreSQL Core

(Grant Zhou)

Grant Zhou presented “The Missing Link,” a talk focused on bridging the gap between China’s massive PostgreSQL user base and the global PostgreSQL core community.

The session explained that China has tens of thousands of PostgreSQL users running large-scale production systems in finance, government, manufacturing, IoT, and database migration projects, but much of this operational experience rarely reaches the global PostgreSQL development process. Grant discussed how projects such as IvorySQL and SynchDB were created to solve real-world enterprise problems around Oracle compatibility and heterogeneous database synchronization.

A major focus of the talk was PGNexus.AI, presented as a “missing link” platform designed to lower the barrier for developers, DBAs, and enterprise users to engage with the PostgreSQL ecosystem. The platform aggregates PostgreSQL blogs, news, social posts, and hacker mailing list discussions into a single hub, while also providing AI-generated summaries, patch analysis, contributor profiles, and developer sandboxes for testing patches and reproducing issues quickly.

Grant emphasized that the goal is not only to make patches more visible, but also to make people more visible — helping contributors build reputation, discover experts, share experiments, and connect local PostgreSQL adoption back into upstream PostgreSQL development. The overall message was that large-scale downstream usage can provide valuable feedback to PostgreSQL core if the community builds better tooling and communication channels to surface it.

30 Years of PostgreSQL Retrospective

(Bruce Momjian, Jan Wieck, Jolly Chen, Jonathan Katz, Melanie Plageman, Thomas Lockhart, Tom Lane, Vadim Mikheev)

The “30 Years of PostgreSQL Retrospective” session at PGConf.dev 2026 was a special roundtable celebrating PostgreSQL’s 30th anniversary, bringing together several early contributors and core figures who helped shape the project in its formative years.

Moderated by Jonathan Katz and Melanie Plageman, the discussion featured stories from PostgreSQL’s early days as an academic Berkeley project evolving into a globally adopted open source database. Panelists reflected on early development challenges, mailing list culture, distributed collaboration before modern open source tooling existed, and important architectural decisions that still influence PostgreSQL today.

One memorable highlight for me was Jolly Chen sharing how he helped add SQL support to Postgres as a graduate student at University of California, Berkeley in 1994 — a contribution that became a foundational step toward PostgreSQL as we know it today.

Social Event + Hallway Meets

One of the most memorable parts of the conference was not only the technical talks, but the social interactions happening everywhere around them — at restaurants, hallways, lounges, parks, coffee breaks, and designated social events.

This is one of the things I love most about the open source community. People can be extremely friendly and approachable socially, while at the same time being technically rigorous, opinionated, and sometimes stubborn during engineering discussions — all driven by a shared goal: making PostgreSQL better for everyone.

Many of the most valuable conversations happen outside the presentation rooms. Sometimes a hallway conversation, a dinner discussion, or a quick whiteboard sketch can lead to new ideas, collaborations, patches, or a much deeper understanding of how PostgreSQL evolves as a global community project.

Thursday – The Energy of PGConf.dev Continues

How to Hack on Logical Replication: Insights from contributors

In this session, Hayato Kuroda and Zhijie Hou shared practical insights into how developers can contribute to PostgreSQL logical replication and help make it more enterprise-ready.

The talk first walked through the architecture of logical replication, including components such as walsender, walreceiver, apply workers, replication slots, logical decoding, output plugins, and table synchronization workers. The speakers explained that each of these areas represents a potential entry point for contributors interested in replication internals.

Using real examples from recent patches, they discussed challenges around publication DDL semantics, output plugin behavior, catalog lookups, compatibility between different PostgreSQL versions during upgrades/downgrades, and replication slot safety. One interesting example showed why logical replication row filters only allow immutable built-in functions, since extension functions may disappear or behave differently when using historic snapshots.

Another key theme was that hacking on logical replication is often less about difficult coding and more about carefully designing semantics, backward compatibility, and crash safety. The session provided a good behind-the-scenes look at the complexity involved in evolving PostgreSQL replication features for large-scale enterprise use cases.

Is There a Future for Genetic and Learning-Based Methods in Query Optimizers?

In this session, Alena Rybakina explored alternative approaches to PostgreSQL query optimization, focusing on complex join-heavy queries such as A JOIN B JOIN C JOIN D ... where the planner search space grows exponentially as the number of joins increases.

The talk compared several optimization strategies, including PostgreSQL’s built-in cost-based planner, GEQO (Genetic Query Optimizer), simulated annealing (SAIO), Monte Carlo Tree Search (MCTS), reinforcement-learning-inspired approaches, and newer machine-learning-assisted systems such as Bao, GenJoin, and SkinnerDB.

A key theme was how iterative search algorithms work: starting from an initial plan, modifying candidate plans, evaluating cost, and repeating until the optimization budget is exhausted. While techniques like GEQO, SAIO, and MCTS can sometimes discover better plans for difficult join queries, many of them struggle to scale beyond roughly n > 12 joins because the search space grows faster than the available optimization budget.

The presentation also showed how newer learning-based approaches attempt to guide PostgreSQL’s planner instead of replacing it entirely. Systems such as Bao use machine learning to evaluate candidate plans and feed runtime latency back into the model, while GenJoin learns which join algorithms work best for different table pairs and lets PostgreSQL still perform the final planning.

Benchmark results showed that some experimental systems can significantly outperform PostgreSQL on very large and difficult analytical joins, though they may also introduce overhead for smaller queries. Overall, the session provided a realistic view of both the promise and the limitations of AI- and search-based query optimization techniques in future PostgreSQL research.

“Developer U”: Lessons Learned from a Global Training Program for Postgres Developers

(Andrew Dunstan)

In this session, Andrew Dunstan introduced EDB’s internal “Developer U” program — a multi-week training initiative designed to grow new PostgreSQL contributors.

The program teaches more than just C coding and PostgreSQL internals. It also focuses on how the PostgreSQL community works: reading pgsql-hackers discussions, following CommitFest workflows, writing tests and documentation, reviewing patches, and understanding the patience and communication style required for upstream open source development.

Andrew also shared the practical challenges of running a global training program: participants are spread across time zones, have different technical backgrounds, and still need to balance training with their regular day jobs. The initiative has already been held in cities such as Barcelona and Montreal, with future cohorts planned to dive even deeper into PostgreSQL internals and real upstream contribution work.

Overall, the talk highlighted EDB’s effort to systematically train and grow the next generation of PostgreSQL developers.

Community Booth Hour is Back

I signed up for a time slot at the community booth to share the ongoing PGNEXUS.AI initiative our company has been building, so unfortunately I had to miss several sessions happening during that period.

At the booth, I introduced both the PGNEXUS.AI and IvorySQL communities to attendees from around the world and received many interesting suggestions, ideas, and feedback.

In short, PGNEXUS.AI is a collaborative platform designed to help connect PostgreSQL users, developers, DBAs, researchers, and contributors from different countries to the global PostgreSQL ecosystem. I described it as “LinkedIn for PostgreSQL” but with much deeper technical and community-oriented capabilities.

Powered by numerous background AI Agents, the platform aggregates, curates, scrapes PostgreSQL-related blogs, news, social posts, hacker mailing list discussions, patches, conference content, and ecosystem updates into a single searchable and multilingual knowledge hub. It uses AI to summarize discussions, analyze patches, identify potential issues or risks, and make technical information easier to discover and understand across language barriers.

We are also exploring developer-focused sandbox and experimentation environments with incremental build support, allowing contributors to quickly apply, test, benchmark, and revert PostgreSQL patches on live sandbox instances. The goal is to lower the barrier to entry for PostgreSQL development and provide a practical collaboration platform for both existing contributors and newcomers to the community.

The Cake Cutting

PostgreSQL also celebrated its 30th anniversary during PGConf.dev 2026, with Bruce Momjian representing the community in the official cake-cutting ceremony. A large crowd gathered around the celebration area to join the moment, making it feel less like a technical conference and more like a global birthday party for the PostgreSQL community.

I was lucky enough to be one of the few attendees who personally received a slice of cake from Bruce himself — and it was actually a very delicious cake. There was also a raffle draw for several special signed PostgreSQL 30th anniversary posters, autographed by PostgreSQL committers. Unfortunately, as usual, I did not win the raffle this time either.

Still, the overall experience was fantastic. The entire celebration took place against Vancouver’s beautiful waterfront backdrop, with the mountains, harbor, and sunset atmosphere making the event feel even more memorable. Moments like these are a reminder that PostgreSQL is not only a database project, but also a long-running global community built by people who genuinely enjoy gathering, collaborating, and celebrating together.

Lightning Talks

Lightning Talks are one of the long-standing traditions at PGConf.dev 2026, where attendees are given less than five minutes on stage to present anything related to PostgreSQL. The signup process itself is also quite fun; participants write their names and topics on paper slips and drop them into a raffle box, using different colored papers to indicate whether they are experienced speakers or first-time presenters.

To be honest, I am still not entirely sure what practical difference color paper make, since everyone who gets selected still ends up on stage speaking anyway. But perhaps that is part of the charm of the session; it is informal, relaxed, and intentionally lightweight compared to the highly technical talks during the day.

The topics can range from serious technical ideas and patch discussions to funny PostgreSQL stories, community experiences, experiments, or random thoughts people simply want to share with the community. After an entire day of deep technical sessions, Lightning Talks feel more like an open community gathering where everyone gets a chance to participate, speak freely, and enjoy the lighter side of the PostgreSQL ecosystem.

Meet & Eat (Thursday) + Social

I was also one of the Meet & Eat leaders for the Thursday evening gathering, and this time I brought the group to Cactus Club Cafe, one of Canada’s most popular restaurant chains.

The atmosphere was very relaxed — steaks, burgers, beers, and plenty of conversations about PostgreSQL, open source, work, and life. After a long day of technical sessions, it was nice to simply sit down and enjoy good food with PostgreSQL people from around the world.

Overall, it was another fun and memorable evening, and everyone seemed to have a great time.

The Unconference + Closing on Friday

Another long-standing tradition of PGConf.dev 2026 is the Unconference — a community-driven discussion event where attendees propose topics and vote on which discussions they want to join.

Unlike regular talks, the sessions are highly interactive and often lead to open brainstorming around PostgreSQL internals, future ideas, tooling, and community topics. This year’s Unconference was once again hosted by Nathan Bossart and Andres Freund, continuing one of pgconf.dev’s most unique and community-focused traditions.

Unconference – Global Index

One of the Unconference discussions was a follow-up session on Global Indexes, again led by Dilip Kumar. The discussion focused on the three major unresolved challenges from the earlier Global Index talk: VACUUM scalability, DDL/rewrite complexity, and locking behavior.

Participants explored difficult edge cases involving ATTACH/DETACH PARTITION, REINDEX, tuple cleanup, partition truncation, and how to safely remove global index entries belonging to dropped partitions. Locking was another major concern, since cross-partition uniqueness checks could potentially require broader index locking and introduce new contention or deadlock scenarios.

An interesting part of the discussion was whether PostgreSQL actually needs a “true” Global Index at all if the real requirement is simply cross-partition uniqueness. Alternative approaches were discussed, including an earlier “Global Unique Index” patch approach from 2023 — a patch I also co-authored. That design avoided some of the VACUUM, rewrite, and locking problems by keeping PostgreSQL’s existing index structure unchanged, but introduced other tradeoffs such as slower INSERT/UPDATE performance and possible deadlocks during cross-partition uniqueness checks.

Overall, the session showed that while many users want global uniqueness support for partitioned tables, every proposed implementation still comes with very hard architectural tradeoffs, and there is still no clear consensus on the best long-term solution.

Unconference – Code, AI and You

Another interesting Unconference session was “Code, AI and You,” hosted by Joe Conway. The discussion explored how AI is rapidly changing PostgreSQL development, review workflows, and even community culture.

Topics ranged from whether AI-assisted code should require disclosure or attribution, to how LLMs could help with bug finding, feasibility studies, patch review, testing, and reviewer assistance. Several attendees discussed the possibility of training PostgreSQL-specific AI models or creating “reviewer skills” based on historical mailing list and patch review data.

At the same time, the discussion also raised concerns around ethics, copyright, stability, and over-reliance on AI-generated code. Many participants agreed that PostgreSQL ultimately remains a database system where predictability, safety, and human understanding are more important than simply generating code faster. One particularly interesting point was that AI may not necessarily accelerate coding itself, but could instead help reduce PostgreSQL’s long-standing bottlenecks around patch review, feedback, and committer bandwidth.

Overall, the session reflected a growing realization that AI-assisted development is already happening across the industry, and that the PostgreSQL community will eventually need to define how these tools should (or should not) fit into upstream development.

Unconference – Logical Replication Warts and Missing Pieces

Another Unconference discussion, hosted by Hannu Krosing, focused on the remaining “warts and missing pieces” in PostgreSQL logical replication.

The discussion covered many long-standing limitations and operational edge cases, including DDL replication, replication of tables without primary keys, object remapping, failover visibility of logical replication slots, and synchronization problems during failover/failback scenarios in complex replication topologies. Participants also discussed whether PostgreSQL should become more “cluster aware” when managing logical replication metadata and slots.

One particularly interesting topic was whether CTID could be exposed and used as a lightweight identifier for logical replication when users do not want to define primary keys, though concerns were raised around HOT updates and tuple movement semantics. Compatibility between different PostgreSQL protocol versions and the long-term maintainability of logical replication wire protocols were also debated.

Overall, the session highlighted that while PostgreSQL logical replication has matured significantly over the years, many difficult edge cases still remain unsolved, especially for enterprise-grade HA, failover, and heterogeneous replication environments.

Closing

This is always the saddest part of the conference — the time to say goodbye.

After several days filled with meaningful technical talks, deep discussions, spontaneous hallway conversations, late-night dinners, social gatherings, and interactions with so many brilliant minds from around the world, the conference finally comes to an end. It is amazing how quickly a group of strangers can turn into friends, collaborators, and familiar faces within just a few days of sharing the same passion for PostgreSQL and open source.

PGConf.dev is never only about technology. It is also about the people behind PostgreSQL; the contributors, reviewers, organizers, sponsors, volunteers, speakers, DBAs, developers, and community members who continue to push the ecosystem forward together.

A huge thank you once again to all the organizers, volunteers, sponsors, speakers, and attendees who made this conference such a memorable and successful experience. Hopefully the PostgreSQL community will continue to grow stronger, more welcoming, and even more inspiring in future conferences to come.

Conclusion

PGConf.dev 2026 was once again an incredible experience. Beyond the deep technical discussions and PostgreSQL internals talks, what truly makes this conference special is the people behind the project and the strong community spirit surrounding it.

This year felt even more meaningful as PostgreSQL celebrated its 30th anniversary in my hometown, Vancouver. I left the conference with new ideas, new friendships, and even greater respect for the amount of collaboration and passion behind PostgreSQL.

And with PGConf.dev 2027 already announced for Montreal next year — will I see you there?