Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Postgres FM

Postgres FM artwork

Postgres year in review 2025

Jan 2, 202647 minAudio
ShareXLinkedInFacebookWhatsAppEmail

Nik and Michael discuss the events and trends they thought were most important in the Postgres ecosystem in 2025. Here are some links to things they mentioned: Postgres 18 release notes https://www.postgresql.org/docs/18/release-18.html Our episode on Postgres 18 https://postgres.fm/episodes/postgres-18 LWLock:LockManager benchmarks for Postgres 18 (blog post by Nik) https://postgres.ai/blog/20251009-postgres-marathon-2-005 PostgreSQL bug tied to zero-day attack on US Treasury https://www.theregister.com/2025/02/14/postgresql_bug_treasury PgDog episode https://postgres.fm/episodes/pgdog Multigres episode https://postgres.fm/episodes/multigres Neki announcement https://planetscale.com/blog/announcing-neki Our 100TB episode from 2024 https://postgres.fm/episodes/to-100tb-and-beyond PlanetScale for Postgres https://planetscale.com/blog/planetscale-for-postgres Oracle's MySQL job cuts https://www.theregister.com/2025/09/11/oracle_slammed_for_mysql_job Amazon Aurora DSQL is now generally available https://aws.amazon.com/about-aws/whats-new/2025/05/amazon-aurora-dsql-generally-available Announcing Azure HorizonDB https://techcommunity.microsoft.com/blog/adforpostgresql/announcing-azure-horizondb/4469710 Lessons from Replit and Tiger Data on Storage for Agentic Experimentation https://www.tigerdata.com/blog/lessons-replit-tiger-data-storage-agentic-experimentation Instant database clones with PostgreSQL 18 https://boringsql.com/posts/instant-database-clones turbopuffer episode https://postgres.fm/episodes/turbopuffer Crunchy joins Snowflake https://www.crunchydata.com/blog/crunchy-data-joins-snowflake Neon joins Databricks https://neon.com/blog/neon-and-databricks ~~~ What did you like or not like? What should we discuss next time? Let us know via a YouTube comment, on social media, or by commenting on our Google doc! ~~~ Postgres FM is produced by: Michael Christofides, founder of pgMustard Nikolay Samokhvalov, founder of Postgres.ai With credit to: Jessie Draws for the elephant artwork

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Michael: Ho ho ho hello and welcome to Postgres.FM. My name is Michael I'm a founder of pgMustard and I'm joined as always by Nik from PostgresAI Merry Christmas Nik and happy new year. Nikolay: Yeah, you too and to all our audience which yeah roughly as we just counted across all the platforms, roughly 10,000 people. So it's impressive. Michael: Yeah, it's kind of wild, but yes, thank you everyone for another good year. Nikolay: Yeah.

So obviously we wanted to do some overview of this year 2023 for Postgres. I think it's a great year for Postgres because a lot of stuff, a lot happened and a lot is still happening. And vector of development of product and communities around it and companies and everything ecosystem it's it's astonishing right Michael: yeah absolutely Feels like it's going from strength

to strength and yeah, as you say another good year Nikolay: Yeah, so where to start let's maybe start with some technical topics first. Maybe Postgres 16 Obviously 1 of them and we had a whole episode about that and what do you remember? What are achievements of Postgres 16 which shine? Michael: Yeah good question I guess 1 thing we didn't have when Postgres 16 came out and we did the episode was a little bit

of hindsight. We didn't know how it was going to go. With a major version, you're always a tiny bit worried there's going to be a big issue in it. And once you get through the release candidates and the beta without that, that either means there isn't a big issue or it hasn't been found in testing yet. So now with a bit of benefit of hindsight I think it's gone pretty well hasn't it? Like I mean touch wood it feels like not only has it gone down well but people have been reporting some performance improvements, people have actually been upgrading.

I'm actually seeing people, in our case, submitting version 16 query plans to the product. So they're using it on production and already looking into performance things with it. So they've not only upgraded, but have had it in production for a little bit. So yeah, it seems like people are upgrading which isn't always the case when a new major version first comes out and Seeing some wins, which is cool. Nikolay: Yeah, I also feel the pace of upgrades is improving from year to year I think managed Postgres providers say polish

the procedure and make new Postgres available sooner than in previous years. So it's obviously being noticed, right? So, but in terms of Postgres 18, I remember things like in any overview somehow UUID version 7, native support of it made it to the short list of always mentioned things. And it's great because again, it started during Postgres TV hacking

sessions online. I'm super proud of this. Like, you know, like finally it made it after a couple of years of waiting because of, like, waiting for RFC standard to be finalized. Postgres was very conservative compared to others. It's great that this thing made it. It's not a huge thing, but it's obviously very useful, helpful for people. Michael: Not a reason to upgrade though, right? Nikolay: No, no. Among reasons, I wanted to mention a specific one. I know we have customers who are working on upgrades sooner and

already made a decision to upgrade much sooner than usually to Postgres 18 just because LockManager, lightweight lock contention is fully solved. And I have a series of articles exploring this in detail, showing benchmarks that you just raise, if you upgrade, you just raise max_locks_per_transaction parameter, which requires a restart, unfortunately, but it's easier than suffering, you know? Suffering from lightweight LockManager contention. And that's it, it's solved.

So this work also took a few years to make it into final release and finally we have it and this is definitely one of the reasons to upgrade. Now if there are other reasons, I also noticed as we discussed AIO. That's a Michael: huge one for people. I actually haven't seen many, I think a few people have blogged kind of theoretically about the performance improvements that could be possible there, but I haven't seen anybody publish anything that they upgraded and saw a big improvement due to AIO, but

I would say it has driven a lot of interest in the major version and it could it could be that it is proving really good for performance but but more than that it's like a good marketing feature like it has driven a lot of interest in the upgrade and people actually considering it. Nikolay: Yeah I saw it not in upgrade, but I saw it in... Well, it's actually my blog post... Maxim's and mine blog post about pgBackRest. And I suddenly noticed that before single-threaded pg_basebackup

only could do 300-400 megabytes per second, but in Postgres 18 it's gigabyte per second on fast disks so it was big surprise to me and it's also like 1 of the reasons I think should should be considered to upgrade sooner Oh, yeah, we can dive into technical details for another episode, but yeah, let's move on. What's next? What's what is big in 2025 on your opinion in terms of technical stuff Michael: on just the technical side that's an interesting question

I've got a lot actually in and around the community, but on the technical side, I think 1 of the bigger things we saw was more efforts, actually maybe that's community still, do you consider kind of the efforts to empower and grow more hackers, like more actual Postgres developers, people working on the Postgres core base, do you consider that technical or more community? Nikolay: Well, it's less technical. Technical, if you mentioned the word hackers, let's use this

word in a different meaning. Remember US treasury hack? Michael: Oh yeah. Nikolay: This is quite technical. There was a case when it was not so good, right? And Postgres obviously used everywhere, so vulnerabilities happen from time to time and it's obviously technical wise, Postgres has great process to fix, to issue new minor, set of minor releases which close vulnerability, but still due to very wide use of

Postgres. I think this year Postgres made it to more technical media because of that. And it was mentioned in the context of, okay, There was like hackers from China hacked the US infrastructure and Postgres was involved. So this I think it's also kind of trend and it's slightly less technical but there is some technical aspect in it and obviously it means that you need to upgrade in terms of minor upgrades always, like patches should be applied promptly Because things

happen, bad things happen, and it's better to keep Postgres up to date in terms of minor version Right, So this is what I just remember this this thing. Michael: Nice. I have 1 more technical 1. Yeah Feels like it was a big year for Announcements at least in the sharding space. We got at least 3 3 big new projects kicked off to try and shard Postgres specifically for OLTP

Nikolay: Yes, exactly. So we we had episodes with PgDog, Lev Kokotov, PgDog. We have had episode with Sugu Multigres, which is developed inside Supabase. And yeah, these are 2 things which are purely open source, and there is a non-open source product, PlanetScale is developing, right? So, yeah. Michael: I think they've said they're gonna open source it, but it's all private at the moment. Nikolay: Yeah. Yeah. We have 2 open source things we can also, actually not we can,

my team is already, my team has already tested in terms of latency overhead, almost all of them. Michael: Is Multigres testable already? Nikolay: I've heard kind of yes, I cannot say for sure. There is something there already. So I will let them announce all the things, but it's definitely being developed quite rapidly. You can go and check pull requests. What I like about open source is it's very transparent, right?

So you know what's already happening in vector In which it's being developed in both cases. This is great power of open source Yeah, yeah Michael: true and it's yeah, it's Interesting like how they're choosing slightly different trade-offs and I think it will be fascinating in the coming years, kind of 3, 4 years time I'd be really interested to see who's using which systems, what are the trade-offs, like which kind of systems benefit most from

1 approach versus the other, Does the license matter? I think PgDog and Multigres have gone different directions on the licensing front. Language. Yeah, exactly. So it will be interesting, but this felt like the big year. I think it was, I've written down, it was May that PgDog was first announced. June that Multigres was first announced and August that Neki was first announced. So quite short, you know, if you consider how long it's been since we had a new potential sharding solution and then 3 come along in a few months, it feels like the timing was right for that to happen now.

Nikolay: So yeah, there are companies which do need it. And there is big demand, including companies who don't need it. This is like, yeah, with sharding, more people think they need it and actually need it. But it always was a weak side of Postgres ecosystem, lack of reliable and mature sharding solution. We had Citus, we had PgCat, we had SPQR. And none of them was publicly known to be used in big OLTP projects

like huge huge scale like you know like for Vitess. Vitess we have GitHub, Slack and so on, Pinterest. Here we didn't have such stories. We still don't have such stories yet but I'm 100% sure we will have it soon, just because of this year's development. Michael: Yeah, the big projects that sharded Postgres, almost all of them seemed, or at least the ones that have talked about it publicly, did so application side. So they have sharded, but it's all on their own heads, basically,

or, you know, using a framework, not a Nikolay: Oh, yeah, yeah, we had episode 100 terabyte of Postgres. It was great. And 2 or 3 guests, I think all the guests use basically application side sharding. Yeah. But at least 1 of them have good articles about this. I think Notion, right? Notion. Michael: And Figma did too. Nikolay: Okay, great. So, you can do sharding yourself, of course, right? But it's good to have some system.

And obviously we are going to have multiple systems which will be used in big companies soon. Michael: I think the Vitess thing is really interesting as well though, because I think there's, you mentioned more people want sharding than need sharding. I think there is like a really attractive thing to early stage startups that, you know, having to tell all their investors they're going to be the biggest thing ever, probably have to believe that as well. So they want, and some of them, it is true, right? Like some of them are going to need, we talked about 1 last, last week where it was, they should have partitioned a year ago,

but they only launched a year ago. So there are these startups that need a small database at the beginning, and they want, they're going to need sharding within a couple of years because they're going to scale that fast. So for them the story of, and it used to be let's say PlanetScale with Vitess or just Vitess in general, was you can start small with us and then we, you know, you can have a sharded setup when you're ready. It was the same story with Mongo many years ago. It was this kind of like, I'm going to grow for web scale. Do you remember that phrase? So it is really, yeah, it will be nice.

And I think that kind of transitions onto 2 things I had on my list. 1 was, I think MySQL's loss is Postgres' gain this year. I think there's been quite a few, like firstly PlanetScale coming into the Postgres ecosystem, but also I think there were some stories more recently around a lot of MySQL's team getting laid off. So I think momentum on that's been kind of shifting and slowing for a while since the Oracle acquisition, maybe longer. But I think Postgres is ready to take up the slack and the work

on sharding is part of that story because I don't think we did have a good answer to Vitess until, well, and maybe arguably still don't, but it feels promising that 1, hopefully 1 of these 3, if not more than 1 of these 3, proved to be good. Nikolay: Yeah, I agree with you. And that's why I actually mentioned the stories matter. So we have stories published about successful sharding of Postgres to great scale in DIY approach.

But we need more like case studies. And I think I'm looking forward to next couple of years to hear about successful use of new sharding solutions. So it's going to be great. This is a great foundation this year, basically answer that Postgres won't be considered as very limited OLTP database system where you will be scared to hit some walls. All right, so this is a great year for this topic.

I like it a lot. Okay, let's move on. Maybe I mentioned less technical stuff. I think there is a big movement in terms of startups and Postgres companies, pure Postgres companies and less pure, but also very Postgres related companies. For example, let's start from maybe like, there are bad news and good news. Bad news, I think, what do you think about Yugabyte? I think Yugabyte is struggling. It's not pure Postgres, but it showed some way to, also solid

solution, right? But I think it's struggling. I see people like, left, many people, and I also think, the topic we just discussed, that's a native sharding solution will have multiple OLTP to achieve great scale and automation. They are going basically to say Yugabit is not needed anymore. Michael: Yeah, I honestly don't know. And I think when we said lots of people, more people want sharded

than want, than need it. I think the same is true for distributed. I categorize them slightly differently in my mind, like sharded and distributed, and I think maybe the difference is the idea of writing to multiple primaries in different availability zones. It feels like it's this kind of like, you pay the you pay the tax on latency and for some it's 100% worth it. But for a lot it's not that trade off isn't worth it. And like I think a lot more people think they want multi master

multi or whatever the I don't know what the appropriate phrase is for that I'm a bit multi primary Nikolay: not so much so is fine these days again I'm joking interesting Michael: yeah but yes so my main point is I think for the kind of financial institutions or big banks that need it, it's hugely valuable, but I think most companies don't need that and the trade-offs aren't worth it. But I have no idea how they're doing. And also it's kind of Postgres compatible, right?

Do you consider it Postgres? Nikolay: Well, yeah, yeah, you're right. You're right. It's more Postgres compatible than CockroachDB, but it's definitely not pure Postgres. Anyway, I think solutions we just discussed are going to meet needs of people who start with just small 1 Postgres like more naturally, you know, like then migrating to other systems. Although at the same time, we must admit that, for example, AlloyDB feels good. I have customers who migrated to AlloyDB and actually I can mention

Gadget for example, we had an episode with Gadget CTO, right, Harry, and we just found out that our monitoring system works well with AlloyDB and like AlloyDB also has bloat and so on things to take care of it's interesting so it feels like Postgres although there is a lot of rewritten also this year I think It's important to mention Amazon DSQL although it's not open source at all, but it also plays in the same area for enterprises.

They are not limited, distributed, everything, multi-region, everything. And they are quite active in terms of development, I think, although it's not my area at all. Like I'm pure Postgres guy and I consider all these friends of Postgres so to speak, some of them open source, some are not. It's just like for me it's kind of it's great to have them around, but I think pure Postgres should be majority of in terms of choices. Michael: Yeah and well like each of the each of the hyperscalers

has their version of this product right this kind of distributed Postgres compatible for whatever definition of Postgres compatible that they're going with this year But there has still been significant contributions from these hyperscalers to pure Postgres. And I think there's a kind of alternative timeline where we wouldn't be seeing the likes of AWS and Microsoft and even a little bit Google Cloud, but like lots and lots of companies, but including those hyperscalers actually contributing back to pure Postgres,

not just making their own proprietary forks. So it's been quite, I really hope it continues, but this year it certainly feels like a lot of the contributions we got a lot of the improvements in Postgres 18, a lot of the community efforts came from these big companies. And I would include EDB in that as well in terms of big companies continuing to like invest in Nikolay: pure Postgres. Microsoft HorizonDB, this is fresh news. I already, I lost track for this enterprise attempt, like enterprise

Postgres topic, right? Like so many options. I think Pure Postgres is a player in enterprise and it's going to stay as a player in enterprise. Although we have Aurora, AlloyDB, HorizonDB, DSQL, who else? EDB, everything. Michael: Yeah, but all of them also offer a separate product that is much, much closer to pure Postgres. Right. So AWS have RDS, Google have Cloud SQL, you know, if every actually

always forget what Microsoft call theirs, but every single 1 of them have 1 that looks just like Postgres. That acts just like Postgres. Yeah, it's something like for PostgreSQL or something. Nikolay: We have customers, but I don't even remember. With Microsoft, I touch Microsoft much less than others. Michael: Microsoft is good at a lot of things, but naming products is not 1 of them. Nikolay: Well, Google is a champion.

Michael: Yeah, fair. Nikolay: So anyway, we have so many flavors of Postgres, right? And obviously there is an interesting competition in the area of Sharding and there is all this competition, which not started this year, it started much earlier, but now it's becoming Definitely like kind of red ocean in terms of enterprise policies topic, right? This is absolutely a red ocean every big company is playing there. Every big cloud company is playing there Michael: Yeah, that's an interesting point because if you're

talking about it being highly, highly competitive, normally what you'd expect to see if that was the case would be prices starting to come down. And I don't think we've seen that. Like I still think they're charging quite a premium for it. So but if it was truly a red ocean, I don't know, maybe I'm not seeing the kind of enterprise contract negotiations. Maybe that's where the prices are coming down. — Nikolay: Well, yeah, Maybe you're right. I think there are segments there and in the segment of like big Postgres, enterprise Postgres.

I think there are different methods to cut prices. They all like normal cloud ways, you know, like, like savings plans, saving plans and so on. Yeah. But they don't rush into going down in terms of prices for Postgres for 1 machine. Obviously, there is big premium. In terms of technical things trends this year, not touching AI yet, but maybe moving closer to it. I think branching is becoming, like, slowly, very slowly, it

makes its way to being commodity. It's not yet there at all, like I think. And branching is still very rare. It's not, like, not everyone has it. — I feel Michael: like there's 2 trends, right? Like I feel like there's branching and there's also cloning and these are very kind of like similar topics and they're kind of being converged at and some, some platforms and some systems offer, offer cloning. Nikolay: Well, some Michael: platforms, some platforms offer thin cloning some,

but mostly it's thick cloning, but that's still, that's still valuable for some amounts of experimentation and test runs. Nikolay: Yes, had always, since the very beginning. Michael: And Heroku even, Heroku even offered it. Nikolay: Right, But if it's thick, well, tooling is improving there for sure. But, I'm big fan of, you know, like I'm, I'm big fan of thin cloning and we have our own product, DBLab for database branching

and thin cloning. And, I think is if before this year, Neon was there, DBLab was there and that's it. Now we see Timescale, Tigris data playing there. Michael: That happened this year. Nikolay: Understanding its role for AI and experiments is great. Like This is exactly how I see it for many years already. And their CTO Mike just posted yesterday an article how Replit is doing this similar thing. So experimentation at scale, like in isolated environments and

to make it fast and cheap, you need copy on write. So this is the way you like basically you can have a lot of ideas how to improve your database. And not everything can be verified on thin cloning, but a lot of ideas can be verified and in the isolated Postgres instances and database branching is great. For example, definitely everything related to query optimization at macro level, we discussed it many times, working with plans, verifying ideas, this is absolutely needed.

Think otherwise you become too slow, too expensive, and you are limited. And in the topic of self-driving Postgres, we actually, yesterday I noticed big trend, Like people started talking about autonomous Postgres and self-driving Postgres. We had, I have a post, someone from EDB mentioned something and I also posted. So this is part of autonomous Postgres for sure. Automated experimentation and branching is essential part of this vision. And we have pieces already when companies start implementing

this. So this works at GitLab, DBLab for many years already so I'm happy this topic becomes more and more popular and future of self-driving Postgres also like it's still foggy But we have components which already becoming quite clear and we see how to achieve levels of automation very high levels of automation So Michael: did you see? In this on this topic area of like branching, thin cloning, and

copy on write, did you see there was a recent post by Radim from Boring SQL talking about a feature in Postgres 18 that I'd missed that is like you can create database within Postgres and and specify a strategy and that can use if your file system supports Nikolay: yeah yeah yeah I actually I also missed I remember this discussion on hackers mailing list, but I somehow overlooked it. So it's already Postgres 18, are you sure?

Michael: Yeah. Wow. I haven't verified myself, but this blog post says so, yeah. Nikolay: It has a limited scope of use. So I think if you control everything, all experimentation environments very well, you can definitely use it. So basically you have 1 single Postgres instance and each create database. If it's based on copy and write, it's fast and cheap. It brings you isolated logical database for experimentation, but it's not for everything. For example, in the case of DBLab, you can perform major upgrade of clone.

Sure. And test. Here, it will be not possible because it's a single instance. Although, working with like ideas how to verify indexes, to tone tuning, this will work for sure. This is great. Michael: Yeah. Yeah. And baked into Postgres already, that's super cool. Nikolay: Yeah, well, that's cool. That's cool. I need to revisit this somehow, I overlooked as well. Great, great. I think what we like, what we see in the future, more tools which will automate experimentation for optimization, performance optimization,

and instead of like throwing something to LLM and sitting with ideas which you don't know which of them are correct which are completely wrong. You will throw to some tool and LLM will generate ideas. Well, this is Actually already happening. We just don't have it connected to all the pieces, right? But we ideas are verified on close on thin clones, right? Yeah, and the user receives already ideas which were considered

during brainstorm phase and then ideas which didn't work and ideas which worked best proposed for to apply in production. Michael: Yeah, I personally think this is something, this is a topic that has got a lot of legs for the coming years But I don't think we've seen yet that much like I don't think I think we've only seen the beginnings of this topic Yeah, Nikolay: and the pieces Some basically like building blocks we have. Yeah, the final like the whole solution is yet to be built. I agree

Michael: Did you see much progress on the vector search side of things this year? Nikolay: Oh yeah, I remember I started here looking at turbopuffer. We found that we also had an episode. So it confirms it's a great successful year for Postgres, but also for PostgresFM podcast, because we had great episodes covering areas which are important for users, right? So, turbopuffer were great, and this approach with vectors in S3,

and you know S3 also now has a vector type. I revisited recently and it's outside of Postgres, but I started looking at it because I noticed clients who come to us using Postgres, they use them. And also I noticed it's used in Cursor, so that's how I know there's turbopuffer. So a combination of Postgres plus turbopuffer, it's kind of like I saw trends similar to you know, like elastic, Postgres plus

elastic. Similar, like something happening. And I thought like it's great like in terms of price and like technology is great. And S3 also has vectors now. But I think it's still to be understood use cases for all of these Approaches and pgvector is going to stay. I wish it was in core. I don't see how it's possible now But Michael: I've had conversations about it there I haven't I haven't been looking closely, but I might have missed them.

Have there been conversations about doing something in core, like, formally? Nikolay: All beginnings of conversations were just interrupted. Like, this index type is not normal because it's answering different results, right? Because it's due to its approximate nature. Basically, it's not deterministic, right? And this brings new challenges. So to be considered as a core thing, but in my opinion, it must

be in core to be developed properly and be considered as something like super standard. Although, pgvector is super standard thanks to all managed platforms supporting it as well. So all people start there. I wanted to say this, I don't see anyone who solved 1 billion vectors problem yet. All those who claim they solved it, I think they are lying. Michael: It's like the spacing, right? What do they call it? Nikolay: They say, you know it's bad, but you need to basically

cluster, not cluster, partition it, right? So it's not 1 index, it's multiple indexes. Like key spaces, namespaces, you call it, use any word. But what it means is that you don't have a single index which supports good OLTP scale, OLTP latencies, meaning it's definitely much faster than 1 second for a single query and has covering 1 billion vectors. Nobody does it.

S3 including, right? S3 also like has maximum, I don't remember, 50 million or how many vectors in 1 index supports. So anyway, this is unsolved problem and it's unsolved, I think, everywhere, not inside Postgres ecosystem. So it's a big problem still. How to have a single index? Why do we need a single index? Maybe we don't need it. Maybe we can always split it. PgDog also showed how to combine sharding and pgvector, right?

So you can have multiple Postgres instances and so on. Questions about price and so on also can pop up, but turbopuffer is not open source. Again, like, we have great open source tooling already and and if you combine with sharding maybe you are fine but I haven't seen a single 1 billion scale index which would work. Michael: — Why does it matter? Like, if I have a partition table, that's not, that doesn't have a single index on it.

— Right. — So why does it matter? — Nikolay: Complexity. If you talk about regular indexes, having a few billion records in a table is not a problem at all. And latency will be amazing if you have index-only scan, or just index scan. If you don't have these rows filtered out, right? But when it comes to vectors, if you have 1 billion vectors and you try to cover it with single index, build time will be terrible for HNSW, latency will be terrible, so it's not OLTP.

It cannot be, it doesn't meet OLTP standards, which we know like 100, 200 milliseconds maximum because of human perception. Michael: Yeah, but not for search. Nikolay: For search including. Search is Michael: 1 of those use cases where people are willing to, okay fine. Nikolay: No, no, no. Are you okay if you Google something and you need to wait 10 seconds? You're not okay. Michael: Yeah it's a good point actually. I think Google is probably the proof that search is important to be fast. Nikolay: Of course. Because That's what they've Michael: always focused Nikolay: on. People need it everywhere.

Like you have mobile app, you type something, you want to move it very fast. If it's below 100 milliseconds, it doesn't feel slow. That's great. So we have human perception, which defines what multi-piece. Michael: You don't need to convince me that performance matters and it's good, but I was just thinking that might Nikolay: be. Search is considered part of application. It's not like we have such system Okay, let's in some cases. Okay if it's some If it's a lawyer who needs to pull some lawsuits

from history like Related to the topic they are working on right now They can wait a minute. No problem Or if it's Like bi report this actually means we overlooked 1 trend as well Analytical thing right so yeah market iceberg and and DuckDB and how to bring all this and Postgres couple of a few companies played good game and And we are acquired Michael: Yeah, I would say this was huge news, I think, let's

say last year. And then I think this year what happened as a result were possibly that 1 of the biggest analytic databases companies in the world acquired what was 1 of the most promising companies in this space. So that was, I actually, yeah I had that on my list to cover, I was thinking that's more community side of things, but I guess it's technical because it's it came about probably because of their progress on the analytics space. Nikolay: Yeah, so obviously Crunchy Data started to work with

analytical side of Postgres, which was always considered weak and achieved had great achievements and got acquired for a quarter of 1,000,000,000. And Neon, although it's, they didn't play a lot with analytics. They did something with DugDB, as I saw, like, they tried here and there things, but still it was purely an OLTP thing, right? For, with branching and super fast point-in-time recovery and serverless, everything.

And got acquired by Snowflake for 1,000,000,000. Michael: Databricks. Nikolay: Oh, Databricks, yes. Snowflake, Crunchy, Databricks, Neon, right. And they are, as we see, they both work to bring solutions which are combined, like analytics plus OLTP and everything. We could consider this also from a different angle, but this is also a game to conquer like enterprise Postgres topic, right?

Just coming from a political perspective first. And it's interesting. I like what's happening, And I also like that some people don't like to be not in open source environment and more talents are available. Michael: Oh, interesting. Nikolay: To be higher because of these acquisitions. Yeah, of course it's natural because both Snowflake and Databricks, I don't consider them as pro open source companies at all. And we know Neon, I haven't checked for a few weeks, but Neon

stopped showing any commits on GitHub since July, right? Michael: Yeah, that's the tricky thing with acquisitions like this, because if you read the posts about what the plan is and what they're announcing is going to happen, and then watch for 5 years, quite often those things don't align. So it's really difficult to assess these things within, you know, 6 months of them happening as to why did it happen? Is this a good thing for the community?

Is it a bad thing for the community? It's really hard to tell. It could be the best thing ever, or it could be awful. And it's really hard to tell this close to it, which of those is gonna be the case. Nikolay: Yeah, in this context, I would like to mention also a couple of companies which I like what's happening as well. Like you can see like some negative sides, but you can see a lot of positive sides in everywhere. For example, Planescale came to Postgres ecosystem.

And they resurrected the old topic, which some people tried to bring life into. It's let's use local NVMe disks. That's great, like absolutely great. And they also, like I like what they do a lot, in a lot of areas. I don't like their position with open source because I don't see what open source they produce at all. They just use Vitess, which was created before PlanetScale and that's it. Michael: And maintain it, I think, yeah. Nikolay: Well, yeah, yeah, that's, that's, that's of course,

but what else? So anyway, but I like a lot what they do and they brought absolutely different, I think different kind of views at things which Postgres ecosystem somehow missed. So like it's kind of blood of MySQL ecosystem was merged to Postgres ecosystem and this is great. Like I've different points of view and so on and This feels really

great and this year achievement, I think, in terms of Postgres ecosystem development. And another is, of course, Supabase, which is pure open source. They bet heavily on open source and Multi-model and OrioleDB and all things. And this is their, I think, strategy to do things open source. And their growth is... Well, and not just, I would Michael: say not just open source, but with permissive licenses Yeah, I think it's I think that's incredible is I think it's something that people don't really fully appreciate of how important

it was for PostgreSQL to have a perm, not just an open source license, but a really permissive open source license. Nikolay: Yeah. Like Apache, MIT. Michael: Exactly. Exactly. And. Nikolay: Postgres license. Michael: Yeah. All of them extremely permissive. And, And I mean, they're not copy left. They're not AGPL. Like they're actively trying to encourage collaboration and allowing people to use it for commercial purposes is is you Nikolay: know, common misconception here at GPL and AGPL, they

are more open source than these permissive licenses, right? But somehow people avoid them. Because they cannot build commercial cloud things on top of them. Michael: I'm not trying to get into an argument as to what's more or less open source, I'm saying that the fact that permissive I think is, Nikolay: is... Freedom, yeah. Michael: Yeah, and I think it's good for everybody, it's good for the rest Nikolay: of us. Unprotected freedom, this is how you can see it. Because somebody can clone and create commercial software based on that, easy. Michael: Well, and I think it's about betting on the long term.

Like, I think it's about saying commercial things will come and go, but this will still be here. Nikolay: It's off topic, Michael: yes. I apologize. Nikolay: I wanted to credit them for that. Michael: To wrap it up, Nikolay: I just wanted to mention that Supabase growth is absolutely crazy. And why? Because a lot of tools which are also growing crazy, They basically bring a lot of development activities even from non-developers,

vibe coding, right? Many platforms grow, grow, grow. And Postgres became, it's obvious this year, Postgres is default database for vibe coding. This is 100% Like as of end of 2025 it's so and Supabase growth and Neon mentioning that 80% or something of databases created are created because of vibe coding and AI agents and so on. Michael: On Neon yeah. What about Replit?

I actually didn't finish reading the what did you see the post that Mike was responding to? Is that Postgres? Nikolay: Yeah well yeah it's Postgres yeah so Postgres everywhere. Michael: Yeah nice. I would say the thing about Supabase and PlanetScale and people like that, they're also amazing at marketing and I think that's underrated in the Postgres ecosystem in terms of being good for the project. They've kind of brought some of that energy of early MongoDB that people are actually excited to use it, developers want to

use it, they think it's an easy thing to do. You know what I mean? It's that kind of like, it's cool all of a sudden. And I don't know if I've been here long enough that to remember the last time that it was actually cool so Nikolay: I don't know yeah let's mention friend Pashol who left Postgres world but he says not fully left also had an episode right yeah yeah and and joined MongoDB, so that's also interesting.

Last time I used MongoDB was 12 years ago, I don't know. Last time I listened to them was 2018 in VLDB Los Angeles. Michael: But if you go back, if you look at startups that are 12 years old, a lot of them will be on Mongo, like, or 11 years old, 10 years old. And I think we're going to see this, like, startups that started now, like last year, this year, next year, a lot of them are gonna be on Postgres, and I think that's really healthy for the ecosystem. Nikolay: Yeah, yeah, yeah. And also challenging because people don't realize what is it.

Michael: Fair. Nikolay: What do Michael: you mean the vibe code is? Nikolay: Right, so I think there are challenges in the area of explaining basics. It's I think demand on understanding like simple basics like for us vacuum is basic but maybe it's not basics. Relational things are basics like sometimes I see different platforms go at different levels there some people like Supabase they expose whole Postgres you can play with tables and so on. Some like Gadget, for example, they expose concept of indexes,

but they don't let you to connect to Postgres directly. So more like vibe side. So different levels here. It's so cool. You can choose how deep you can go, but if you go and you don't understand, you need, I think, demand for good education of relational databases and Postgres is growing because of all these activities. And on 1 hand, on another hand, demand for good maintenance, operational tools and practices and Methodologies also growing

because imagine how many databases are created now every day a year ago. It was much much less right Postgres Like number of databases is exploding this year. Literally. Do you think, but Michael: don't you think their shelf life is also reducing? Like, Nikolay: of course, most of them won't survive. Yeah. We're just experimenting, wipe coding, like throwing out later, but some of them will survive. They need a path for good health.

That's why I think self-driving is also a new trend which is going to stay and It's great to have examples of previous attempts as Oracle Autonomous, we can learn from them. And others. Yeah, okay. We'll definitely discuss this topic not once in future, I'm quite sure. Well, again, like trend is obvious trend started here in 2025 Michael: Okay Wait, well, but yeah I think that it would be interesting

to see how important it turns out to be in 2026 My opinion is actually that the ones that survive can then be given a bit more attention. So like, it's gonna be not that, I think it's still difficult to get a business up and running and to a huge scale. The ones that do it are still incredible and they're still not that many. I don't see that there's gonna be tens of thousands more of those really successful huge companies But they might get there quicker and with fewer technical resources in the beginning So they'll need support for sure But I don't see the argument that there's

going to be a board of magnitude more them. Nikolay: My team feels it really well. Regular, imagine enterprises, regular startups, they are different, right? And there is new breed AI startups. These guys move really fast. They like database expertise even more. They are very often very smart, but they are different from regular startups and I feel it just from perspective how we how like work with databases organized. So this trend will only grow in the future, it started this year

I think, We noticed it this year, but it will not disappear. I think this is something new. Michael: And you don't think they'll hire, like, you think they won't hire for those, like, specialties once they grow? Nikolay: I think they will, but they move so fast. You can hire but then what like it's too late. Like you mentioned this case about partitioning right with regular methods this new hire with absolutely great knowledge will come but it will be too late.

It already should be done a year ago. Michael: But it's not too late. Success generally brings enough money to solve these problems. Like hire a good consultancy to help out. Nikolay: I'm not joking. I heard let's migrate to a different database system maybe. Michael: Okay, so you think for Postgres is survival? Nikolay: It's a challenge. Yeah, yeah. So like, if database cannot scale without like, too much attention and too much manual work, it won't survive its hyper growth.

We had this term hyper scalers. AI is going to bring new level of hyper scaling. So we need to adjust. Thank you so much. Let's keep, let's say happy Christmas, happy new year to everyone. And let's continue new year with all these trends and good Postgres health, good Postgres tools, good open source. It's great to be in this ecosystem and this community in broader

meaning. And I really enjoy making this podcast with you, Michael, and also happy new year and Merry Christmas. Michael: Absolutely, likewise. Thank you everybody. Thank you Nik.

Open the transcript file

Hosts and Guests

Hosts

  • Michael Christofides Host Feed-stated
  • Nikolay Samokhvalov Host Feed-stated

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Postgres 19

    Nov 19, 202653 minAudio

    Tip: watch on YouTube. Since episode 193, the video shows charts and illustrations as we talk. https://youtu.be/VcDU-NRE3hI Michael, Nik, and Kirk discuss Postgres 19: why the release came in November and not in...

    Credit the people on this episode

  2. The hidden costs of CREATE TEMP TABLE

    Nov 13, 202639 minAudio

    Tip: watch on YouTube. Since episode 193, the video shows charts and illustrations as we talk. https://www.youtube.com/watch?v=gAdhpYewLZ8 Nik and Kirk discuss the hidden costs of CREATE TEMP TABLE: dead tuples and...

    Credit the people on this episode

  3. Extensions on demand

    Nov 6, 20261 hr 9 minAudio

    Tip: watch on YouTube. Since episode 193, the video shows charts and illustrations as we talk. https://www.youtube.com/watch?v=7oeAAqU00zk Nik and Kirk are joined by Álvaro Hernández to discuss how Postgres in...

    Credit the people on this episode

  4. UUIDv4 vs UUIDv7

    Oct 30, 202648 minAudio

    Tip: watch on YouTube. Since episode 193, the video shows charts and illustrations as we talk. https://www.youtube.com/watch?v=gf8e3JkQ_8M Nik is joined by Kirk Wolak to discuss what a random UUIDv4 primary key really...

    Credit the people on this episode

  5. Dump/restore done. Now what?

    Oct 23, 202651 minAudio

    Nik is joined by Kirk Wolak, Software Architect at KiraSoft, to discuss what a database still needs after pg_restore finishes and before it takes production traffic - what pg_dump and pg_restore leave behind, why...

    Credit the people on this episode