Just a Theory

By David E. Wheeler

Posts about CloudNativePG

Adventures in Extension Packaging

I gave a presentation at PGConf.dev last week, Adventures in Extension Packaging. It summarizes stuff I learned in the past year in developing the PGXN Meta v2 RFC, re-packaging all of the extensions on pgt.dev, and experimenting with the CloudNativePG community’s proposal to mount extension OCI images in immutable PostgreSQL containers.

Turns out a ton of work and experimentation remains to be done.

Previous work covers the first half of the talk, including:

The rest of the talk encompasses newer work. Read on for details.

Automated Packaging Challenges

Back in December I took over maintenance of the Trunk registry, a.k.a., pgt.dev, refactoring and upgrading all 200+ extensions and adding Postgres 17 builds. This experience opened my eyes to the wide variety of extension build patterns and configurations, even when supporting a single OS (Ubuntu 22.04 “Jammy”). Some examples:

  • pglogical requires an extra make param to build on PostgreSQL 17: make -C LDFLAGS_EX="-L/usr/lib/postgresql/17/lib"
  • Some pgrx extensions require additional params, for example:
  • pljava needs a pointer to libjvm: mvn clean install -Dpljava.libjvmdefault=/usr/lib/x86_64-linux-gnu/libjvm.so
  • plrust needs files to be moved around, a shell script to be run, and to be built from a subdirectory
  • bson also needs files to be moved around and a pointer to libbson
  • timescale requires an environment variable and shell script to run before building
  • Many extensions require patching to build for various configurations and OSes, like this tweak to build pguri on Postgres 17 and this patch to get duckdb_fdw to build at all

Doubtless there’s much more. These sorts of challenges led the RPM and APT packaging systems to support explicit scripting and patches for every package. I don’t think it would be sensible to support build scripting in the meta spec.

However, the PGXN meta SDK I developed last year supports the merging of multiple META.json files, so that downstream packagers could maintain files with additional configurations, including explicit build steps or lists of packages, to support these use cases.

Furthermore, the plan to add reporting to PGXN v2 means that downstream packages could report build failures, which would appear on PGXN, where they’d encourage some maintainers, at least, to fix issues within their control.

Dependency Resolution

Dependencies present another challenge. The v2 spec supports third party dependencies — those not part of Postgres itself or the ecosystem of extensions. Ideally, an extension like pguri would define its dependence on the uriparser library like so:

{
  "dependencies": {
    "postgres": { "version": ">= 9.3" },
    "packages": {
      "build": {
        "requires": {
          "pkg:generic/uriparser": 0,
        }
      }
    }
  }
}

An intelligent build client will parse the dependencies, provided as purls, to determine the appropriate OS packages to install to satisfy. For example, building on a Debian-based system, it would know to install liburiparser-dev to build the extension and require liburiparser1 to run it.

With the aim to support multiple OSes and versions — not to mention Postgres versions — the proposed PGXN binary registry would experience quite the combinatorial explosion to support all possible dependencies on all possible OSes and versions. While I propose to start simple (Linux and macOS, Postgres 14-18) and gradually grow, it could quickly get quite cumbersome.

So much so that I can practically hear Christoph’s and Devrim’s reactions from here:

Photo of Ronald Reagan and his team laughing uproariously with the white Impact Bold-style meme text at the top that reads, “AND THEN HE SAID…”, followed by large text at the bottom that reads, “WE’LL PACKAGE EVERY EXTENSION FOR EVERY PLATFORM!”

Photo of Christoph, Devrim, and other long-time packagers laughing at me.

Or perhaps:

Photo of two German shepherds looking at a pink laptop and appearing to laugh hysterically, with the white Impact Bold-style meme text at the top that reads, “AND THEN HE SAID…”, followed by large text at the bottom that reads, “UPSTREAM MAINTAINERS WILL FIX BUILD FAILURES!”

Photo of Christoph and Devrim laughing at me.

I hardly blame them.

A CloudNativePG Side Quest

Gabriele Bartolini blogged the proposal to deploy extensions to CloudNativePG containers without violating the immutability of the container. The introduction of the extension_control_path GUC in Postgres 18 and the ImageVolume feature in Kubernetes 1.33 enable the pattern, likely to be introduced in CloudNativePG v1.27. Here’s a sample CloudNativePG cluster manifest with the proposed extension configuration:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
apiVersion: postgresql.cnpg.io/v1
kind: Cluster
metadata:
  name: postgresql-with-extensions
spec:
  instances: 1
  imageName: ghcr.io/cloudnative-pg/postgresql-trunk:18-devel
  postgresql:
    extensions:
      - name: vector
        image:
          reference: ghcr.io/cloudnative-pg/pgvector-18-testing
  storage:
    storageClass: standard
    size: 1Gi

The extensions object at lines 9-12 configures pgvector simply by referencing an OCI image that contains nothing but the files for the extension. To “install” the extension, the proposed patch triggers a rolling update, replicas first. For each instance, it takes the following steps:

  • Mounts each extension as a read-only ImageVolume under /extensions; in this example, /extensions/vector provides the complete contents of the image

  • Updates LD_LIBRARY_PATH to include the path to the lib directory of the each extension, e.g., /extensions/vector/lib.

  • Updates the extension_control_path and dynamic_library_path GUCs to point to the share and lib directories of each extension, in this example:

    extension_control_path = '$system:/extensions/vector/share'
    dynamic_library_path   = '$libdir:/extensions/vector/lib'
    

This works! Alas, the pod restart is absolutely necessary, whether or not any extension requires it,1, because:

  • Kubernetes resolves volume mounts, including ImageVolumes, at pod startup
  • The dynamic_library_path and extension_control_path GUCs require a Postgres restart
  • Each extension requires another path to be appended to both of these GUCs, as well as the LD_LIBRARY_PATH

Say we wanted to use five extensions. The extensions part of the manifest would look something like this:

extensions:
  - name: vector
    image:
      reference: ghcr.io/cloudnative-pg/pgvector-18-testing
  - name: semver
    image:
      reference: ghcr.io/example/semver:0.40.0
  - name: auto_explain
    image:
      reference: ghcr.io/example/auto_explain:18
  - name: bloom
    image:
      reference: ghcr.io/example/bloom:18
  - name: postgis
    image:
      reference: ghcr.io/example/postgis:18

To support this configuration, CNPG must configure the GUCs like so:

extension_control_path = '$system:/extensions/vector/share:/extensions/semver/share:/extensions/auto_explain/share:/extensions/bloom/share:/extensions/postgis/share'

dynamic_library_path   = '$libdir:/extensions/vector/lib:/extensions/semver/lib:/extensions/auto_explain/lib:/extensions/bloom/lib:/extensions/postgis/lib'

And also LD_LIBRARY_PATH:

LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/extensions/vector/lib:/extensions/semver/lib:/extensions/auto_explain/lib:/extensions/"

In other words, every additional extension requires another prefix to be appended to each of these configurations. Ideally we could use a single prefix for all extensions, avoiding the need to update these configs and therefore to restart Postgres. Setting aside the ImageVolume limitation2 for the moment, this pattern would require no rolling restarts and no GUC updates unless a newly-added extension requires pre-loading via shared_preload_libraries.

Getting there, however, requires a different extension file layout than PostgreSQL currently uses.

RFC: Extension Packaging and Lookup

Imagine this:

  • A single extension search path GUC
  • Each extension in its own eponymous directory
  • Pre-defined subdirectory names used inside each extension directory

The search path might look something like:

extension_search_path = '$system:/extensions:/usr/local/extensions'

Looking at one of these directories, /extensions, its contents would be extension directories:

❯ ls -1 extensions
auto_explain
bloom
postgis
semver
vector

And the contents of one these extension directories would be something like:

❯ tree extensions/semver
extensions/semver
├── doc
│   └── semver.md
├── lib
│   └── semver.so
├── semver.control
└── sql
    ├── semver--0.31.0--0.31.1.sql
    ├── semver--0.31.1--0.31.2.sql
    ├── semver--0.31.2--0.32.0.sql
    └── semver--0.5.0--0.10.0.sql

For this pattern, Postgres would look for the appropriately-named directory with a control file in each of the paths. To find the semver extension, for example, it would find /extensions/semver/semver.control.

All the other files for the extension would live in specifically-named subdirectories: doc for documentation files, lib for shared libraries, sql for SQL deployment files, plus bin, man, html, include, locale, and any other likely resources.

With all of the files required for an extension bundled into well-defined subdirectories of a single directory, it lends itself to the layout of the proposed binary distribution format. Couple it with OCI distribution and it becomes a natural fit for ImageVolume deployment: simply map each extension OCI image to a subdirectory of the desired search path and you’re done. The extensions object in the CNPG Cluster manifest remains unchanged, and CNPG no longer needs to manipulate any GUCs.

Some might recognize this proposal from a previous RFC post. It not only simplifies the CloudNativePG use cases, but because it houses all of the files for an extension in a single bundle, it also vastly simplifies installation on any system:

  1. Download the extension package
  2. Validate its signature & contents
  3. Unpack its contents into a directory named for the extension in the extension search path

Simple!

Fun With Dependencies

Many extensions depend on external libraries, and rely on the OS to find them. OS packagers follow the dependency patterns of their packaging systems: require the installation of other packages to satisfy the dependencies.

How could a pattern be generalized by the Trunk Packaging Format to work on all OSes? I see two potential approaches:

  1. List the dependencies as purls that the installing client translates to the appropriate OS packages it installs.
  2. Bundle dependencies in the Trunk package itself

Option 1 will work well for most use cases, but not immutable systems like CloudNativePG. Option 2 could work for such situations. But perhaps you noticed the omission of LD_LIBRARY_PATH manipulation in the packaging and lookup discussion above. Setting aside the multitude of reasons to avoid LD_LIBRARY_PATH3, how else could the OS find shared libraries needed by an extension?

Typically, one installs shared libraries in one of a few directories known to tools like ldconfig, which must run after each install to cache their locations. But one cannot rely on ldconfig in immutable environments, because the cache of course cannot be mutated.

We could, potentially, rely on rpath, a feature of modern dynamic linkers that reads a list of known paths from the header of a binary file. In fact, most modern OSes support $ORIGIN as an rpath value4 (or @loader_path on Darwin/macOS), which refers to the same directory in which the binary file appears. Imagine this pattern:

  • The Trunk package for an extension includes dependency libraries alongside the extension module
  • The module is compiled with rpath=$ORIGIN

To test this pattern, let’s install the Postgres 18 beta and try the pattern with the pguri extension. First, remove the $libdir/ prefix (as discussed previously) and patch the extension for Postgres 17+:

perl -i -pe 's{\$libdir/}{}' pguri/uri.control pguri/*.sql
perl -i -pe 's/^(PG_CPPFLAGS.+)/$1 -Wno-int-conversion/' pguri/Makefile

Then compile it with CFLAGS to set rpath and install it with a prefix parameter:

make CFLAGS='-Wl,-rpath,\$$ORIGIN'
make install prefix=/usr/local/postgresql

With the module installed, move the liburiparser shared library from OS packaging to the lib directory under the prefix, resulting in these contents:

❯ ls -1 /usr/local/postgresql/lib
liburiparser.so.1
liburiparser.so.1.0.30
uri.so

The chrpath utility shows that the extension module, uri.so, has its RUNPATH (the modern implementation of rparth) properly configured:

❯ chrpath /usr/local/postgresql/lib/uri.so 
uri.so: RUNPATH=$ORIGIN

Will the OS be able to find the dependency? Use ldd to find out:

❯ ldd /usr/local/postgresql/lib/uri.so 
	linux-vdso.so.1
	liburiparser.so.1 => /usr/local/postgresql/lib/liburiparser.so.1
	libc.so.6 => /lib/aarch64-linux-gnu/libc.so.6
	/lib/ld-linux-aarch64.so.1

The second line of output shows that it does in fact find liburiparser.so.1 where we put it. So far so good. Just need to tell the GUCs where to find them and restart Postgres:

extension_control_path = '$system:/usr/local/postgresql/share'
dynamic_library_path   = '$libdir:/usr/local/postgresql/lib'

And then it works!

❯ psql -c "CREATE EXTENSION uri"
CREATE EXTENSION
❯ psql -c "SELECT 'https://example.com/'::uri"
         uri          
----------------------
 https://example.com/

Success! So we can adopt this pattern, yes?

A Wrinkle

Well, maybe. Try it with a second extension, http, once again building it with rpath=$ORIGIN and installing it in the custom lib directory:

perl -i -pe 's{$libdir/}{}g' *.control
make CFLAGS='-Wl,-rpath,\$$ORIGIN'
make install prefix=/usr/local/postgresql

Make sure it took:

❯ chrpath /usr/local/postgresql/lib/http.so 
http.so: RUNPATH=$ORIGIN

Now use ldd to see what shared libraries it needs:

❯ ldd /usr/local/postgresql/lib/http.so
	linux-vdso.so.1 
	libcurl.so.4 => not found
	libc.so.6 => /lib/aarch64-linux-gnu/libc.so.6

Naturally it needs libcurl; let’s copy it from another system and try again:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
❯ scp dev:libcurl.so.4 /usr/local/postgresql/lib/
❯ ldd /usr/local/postgresql/lib/http.so
	linux-vdso.so.1
	libcurl.so.4 => /usr/local/postgresql/lib/libcurl.so.4
	libc.so.6 => /lib/aarch64-linux-gnu/libc.so.6
	/lib/ld-linux-aarch64.so.1
	libnghttp2.so.14 => not found
	libidn2.so.0 => /lib/aarch64-linux-gnu/libidn2.so.0
	librtmp.so.1 => not found
	libssh.so.4 => not found
	libpsl.so.5 => not found
	libssl.so.3 => /lib/aarch64-linux-gnu/libssl.so.3
	libcrypto.so.3 => /lib/aarch64-linux-gnu/libcrypto.so.3
	libgssapi_krb5.so.2 => /lib/aarch64-linux-gnu/libgssapi_krb5.so.2
	libldap.so.2 => not found
	liblber.so.2 => not found
	libzstd.so.1 => /lib/aarch64-linux-gnu/libzstd.so.1
	libbrotlidec.so.1 => not found
	libz.so.1 => /lib/aarch64-linux-gnu/libz.so.1

Line 4 shows it found libcurl.so.4 where we put it, but the rest of the output lists a bunch of new dependencies that need to be satisfied. These did not appear before because the http.so module doesn’t depend on them; the libcurl.so library does. Let’s add libnghttp2 and try again:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
❯ scp dev:libnghttp2.so.14 /usr/local/postgresql/lib/
❯ ldd /usr/local/postgresql/lib/http.so
	linux-vdso.so.1
	libcurl.so.4 => /usr/local/postgresql/lib/libcurl.so.4
	libc.so.6 => /lib/aarch64-linux-gnu/libc.so.6
	/lib/ld-linux-aarch64.so.1
	libnghttp2.so.14 => not found
	libidn2.so.0 => /lib/aarch64-linux-gnu/libidn2.so.0
	librtmp.so.1 => not found
	libssh.so.4 => not found
	libpsl.so.5 => not found
	libssl.so.3 => /lib/aarch64-linux-gnu/libssl.so.3
	libcrypto.so.3 => /lib/aarch64-linux-gnu/libcrypto.so.3
	libgssapi_krb5.so.2 => /lib/aarch64-linux-gnu/libgssapi_krb5.so.2
	libldap.so.2 => not found
	liblber.so.2 => not found
	libzstd.so.1 => /lib/aarch64-linux-gnu/libzstd.so.1
	libbrotlidec.so.1 => not found
	libz.so.1 => /lib/aarch64-linux-gnu/libz.so.1

Sadly, as line 7 shows, it still can’t find libnghttp2.so.

It turns out that rpath works only for immediate dependencies. To solve this problem, liburl and all other shared libraries must also be compiled with rpath=$ORIGIN — which means we can’t simply copy those libraries from OS packages5. In th meantime, only deirect dependencies could be bundled with an extension.

Project Status

The vision of accessible, easy-install extensions everywhere remains intact. I’m close to completing a first release of the PGXN v2 build SDK with support for meta spec v1 and v2, PGXS, and pgrx extensions. I expect the first deliverable to be a command-line client to complement and eventuallly replace the original CLI. It will be put to work building all the extensions currently distributed on PGXN, which will surface new issues and patterns that inform the development and completion of the v2 meta spec.

In the future, I’d also like to:

  • Finish working out Trunk format and dependency patterns
  • Develop and submit the prroposed extension_search_path patch
  • Submit ImageVolume feedback to Kubernetes to allow runtime mounting
  • Start building and distributing OCI Trunk packages
  • Make the pattern available for distributed registries, so anyone can build their own Trunk releases!
  • Hack fully-dynamic extension loading into CloudNativePG

Let’s Talk

I recognize the ambition here, but feel equal to it. Perhaps not every bit will work out, but I firmly believe in setting a clear vision and executing toward it while pragmatically revisiting and revising it as experience warrants.

If you’d like to contribute to the project or employ me to continue working on it, let’s talk! Hit me up via one of the services listed on the about page.


  1. The feature does not yet support pre-loading shared libraries. Presumably a flag will be introduced to add the extension to shared_preload_libraries↩︎

  2. Though we should certainly request the ability to add new ImageVolume mounts without a restart. We can’t be the only ones thinking about kind of feature, right? ↩︎

  3. In general, one should avoid LD_LIBRARY_PATH for variety of reasons, not least of which its bluntness. For various security reasons, macOS ignores it unless sip is disabled, and SELinux prevents its propagation to new processes. ↩︎

  4. Although not Windows, alas. ↩︎

  5. Unless packagers could be pursuaded to build all libraries with rpath=$ORIGIN, which seems like a tall order. ↩︎

Mini Summit 5 Transcript: Improving the PostgreSQL Extensions Experience in Kubernetes with CloudNativePG

Orange card with large black text reading “Extension Management in CNPG”. Smaller text below reads “Gabriele Bartolini (EDB)” and that is the date, “05.07.2025”.

The final PostgresSQL Extension Mini-Summit took place on May 7. Gabriele Bartolini gave an overview of PostgreSQL extension management in CloudNativePG (CNPG). This talk brings together the topics of several previous Mini-Summits — notably Peter Eisentraut on implementing an extension search path — to look at the limitations of extension support in CloudNativePG and the possibilities enabled by the extension search path feature and the Kubernetes 1.33 ImageVolume feature. Check it out:

Or read on for the full transcript with thanks to Floor Drees for putting it together.

Introduction

Floor Drees.

On May 7 we hosted the last of five (5) virtual Mini-Summits that lead up to the big one at the Postgres Development Conference (PGConf.Dev), taking place next week, in Montreal, Canada. Gabriele Bartolini, CloudNativePG maintainer, PostgreSQL Contributor, and VP Cloud Native at EDB, joined to talk about improving the Postgres extensions experience in Kubernetes with CloudNativePG.

The organizers:

The stream and the closed captions available for the recording are supported by PGConf.dev and their gold level sponsors, Google, AWS, Huawei, Microsoft, and EDB.

Improving the Postgres extensions experience in Kubernetes with CloudNativePG

Gabriele Bartolini.

Hi everyone. Thanks for this opportunity, and thank you Floor and David for inviting me today.

I normally start every presentation with a question, and this is actually the question that has been hitting me and the other maintainers of CloudNativePG — and some are in this call — from the first day. We know that extensions are important in Kubernetes, in Postgres, and we’ve always been asking how can we deploy extensions, without breaking the immutability of the container.

So today I will be telling basically our story, and hopefully providing good insights in the future about how with CloudNativePG we are trying to improve the experience of Postgres extensions when running databases, including issues.

I’ve been using Postgres for 25 years. I’m one of the co-founders of 2ndQuadrant, which was bought by a EDB in 2020. And because of my contributions, I’ve been recognized as a Postgres contributor and I’m really grateful for that. And I’m also “Data on Kubernetes ambassador”; my role is to promote the usage of stateful workloads in Kubernetes. I’m also DevOps evangelist. I always say this: DevOps is the reason why I encountered Kubernetes, and it will also be the reason why I move away one day from Kubernetes. It’s about culture and I’ll explain this later.

In the past I’ve been working with Barman; I’m one of the creators of Barman. And since 2022, I’m one of the maintainers of CloudNativePG. I want to thank my company, EDB, for being the major contributor in Postgres history in terms of source code. And right now we are also the creators of CloudNativePG. And as we’ll see, the company donated the IP to the CNCF. So it’s something that is quite rare, and I’m really grateful for that.

What I plan to cover tonight is first, set the context and talk about immutable application containers, which have been kind of a dogma for us from day one. Then, how we are handling right now extensions in Kubernetes with CNPG. This is quite similar to the way other operators deal with it. Then the future and key takeaways.

First, we’re talking about Kubernetes. If you’re not familiar, it’s an orchestration system for containers. It’s not just an executor of containers, but it’s a complex system that also manages infrastructure. When it manages infrastructure, it also manages cloud native applications that are also called workloads. When we’re thinking about Postgres in Kubernetes, the database is a workload like the others. That, I think, is the most important mind shift among Postres users that I have faced myself, that I’ve always treated Postgres differently from the rest. Here in Kubernetes is it’s just another workload.

Then of course, it’s not like any other workload, and that’s where operators come into play, and I think the work that we are doing even tonight is in the direction to improve how databases is run in Kubernetes in general, and for everyone.

It was open sourced in 2014, and, it’s owned by the CNCF, and it’s actually the first project that graduated, and graduated is the most advanced stage in the graduation process of the CNCF, which starts with sandbox, then incubation and then graduation.

CloudNativePG is an operator for Postgres. It’s production-ready — what we say is level five. Level five is kind of an utopic, and unbounded level, the highest one as defined by the operator development framework. It’s used by all these players including Tembo, IBM Cloud Paks, Google Cloud, Azure, Akamai, and so on. CNPG is a CNCF project since January. It’s distributed under Apache License 2.0 and the IP — the Intellectual Property — is owned by the community and protected by the CNCF. It therefore is a vendor neutral and openly governed project. This is kind of a guarantee that it will always be free. This is also, in my opinion, a differentiation between CloudNativePG and the rest.

The project was originally created by EDB, but specifically at that time, by 2ndQuadrant. And, as I always like to recall, it was Simon Riggs that put me in charge of the initiative. I’ll always be grateful to Simon, not only for that, but for everything he has done for me and the team.

CNPG can be installed in several ways. As you can see, it’s very popular in terms of stars. There’s more than 4,000 commits. And what’s impressive is the number of downloads in three years, which is 78 million, which means that it’s used the way we wanted it to be used: with CICD pipelines.

This is the CNCF landscape; these are the CNCF projects. As you can see, there are only five projects in the CNCF in the database area, and CloudNativePG is the only one for Postgres. Our aim for 2025 and 2026 is to become incubating. If you’re using CNPG and you want to help with the process, get in touch with me and Floor.

I think to understand again, what, why we’ve done all this process, that led to the patch that, you’ve seen in Postgres 18, it’s important to understand what cloud native has meant to us since we started in 2019. We’ve got our own definition, but I think it still applies. For us it’s three things, Cloud native. It’s people that work following DevOps culture. For example, there are some capabilities that come from DevOps that apply to the cloud native world. I selected some of them like in user infrastructure, infrastructure abstraction, version control. These three form the infrastructure-as-code principle, together with the declarative configuration.

A shift left on security. You’ll see with CloudNativePG, we rarely mention security because it’s pretty much everywhere. It’s part of the process. Then continuous delivery.

The second item is immutable application containers, which kind of led the immutable way of thinking about extensions. And then the third one is that these application containers must be orchestrated via an infrastructure-as-code by an orchestrator, and the standard right now is Kubernetes.

For us it’s these three things, and without any of them, you cannot achieve cloud native.

So what are these immutable application containers? To explain immutability I’d like to talk about immutable infrastructure, which is probably what the majority of people that have historically worked with Postgres are used to. I’m primarily referring to traditional environments like VMs and bare metal where the main ways we deploy Postgres is through packages, maybe even managed by configuration managers, but still, packages are the main artifacts. The infrastructure is seen as a long-term kind of project. Changes happen over time and are incremental updates, updates on an existing infrastructure. So if you want to know the history of the infrastructure over time, you need to check all the changes that have applied. In case of failure of a system, systems are healed. So that’s the pets concept that comes from DevOps.

On the other hand, immutable infrastructure relies on OCI container images. OCI is a standard, the Open Container Initiative and it’s part of the Linux Foundation as well. Immutable infrastructure is founded on continuous delivery, which is the foundation of GitOps practices. In an immutable infrastructure, releasing a new version of an application is not updating the system’s application, it is building a new image and publishing it on a public registry and then deploying it. Changes in the system happen in an atomic way: the new version of a container is pulled from the registry and the existing image is almost instantaneously replaced by the new one. This is true for stateless applications and we’ll see, in the case of stateful applications like Postgres, is not that instantaneous because we need to perform a switchover or restart — in any case, generate a downtime.

When it comes to Kubernetes, the choice was kind of obvious to go towards that immutable infrastructure. So no incremental updates, and in the case of stateful workloads where you cannot change the content of the container, you can use data volumes or persistent volumes. These containers are not changed. If you want to change even a single file or a binary in a container image, you need to create a new one. This is very important for security and change management policies in general.

But what I really like about this way of managing our infrastructure is that, at any time, Kubernetes knows exactly what software is running in your infrastructure. All of this is versioned in an SCM, like Git or whatever. This is something that in the mutable world is less easy to obtain. Again, for security, this is the foundational thing because this is how you can control CVEs, the vulnerabilities in your system. This is a very basic representation of how you build, contain — let’s say the lifecycle of a container image. You create a Dockerfile, you put it in Git, for example, then there’s an action or a pipeline that creates the container image, maybe even run some tests and then pushes it to the container registry.

I walked you through the concepts of mutable and immutable containers, what are, these immutable application containers? If you go back and read what we were rising before CloudNativePG was famous or was even used, we were always putting in immutable application containers as one of the principles we could not lose.

For an immutable application container, it means that there’s only a single application running; that’s why it’s called “application”. If you have been using Docker, you are more familiar with system containers: you run a Debian system, you just connect and then you start treating it like a VM. Application containers are not like that. And then they are immutable — read-only — so you cannot even make any change or perform updates of packages. But in CloudNativePG, because we are managing databases, we need to put the database files in separate persistent volumes. Persistent volumes are standard resources provided by Kubernetes. This is where we put PGDATA and, if you want, a separate volume for WAL files with different storage specifications and even an optional number of table spaces.

CloudNativePG orchestrates what we call “operand images”. These are very important to understand. They contain the Postgres binaries and they’re orchestrated via what we call the “instance manager”. The instance manager is just the process that runs and controlled Postgres; I’ss the PID 1 — or the entry point — of the container.

There’s no other, like SSHD or other, other applications work. There’s just the instance manager that then controls everything else. And this is the project of the operating images. This is one open source project, and every week we rebuild the Postgres containers. We recently made some changes to the flavors of these images and I’ll talk about it shortly.

We mentioned the database, we mentioned the binaries, but what about extensions? This is the problem. Postgres extensions in Kubernetes with CloudNativePG is the next section, and it’s kind of a drama. I’m not hiding this. The way we are managing extensions in Kubernetes right now, in my opinion, is not enough. It works, but it’s got several limitations — mostly limitations in terms of usage.

For example, we cannot place them in the data files or in persistent volumes because these volumes are not read-only in any way. In any case, they cannot be strictly immutable. So we discarded this option to have persistent volume where you could kind of deploy extensions and maybe you can even download on the fly or use the package manager to download them or these kind of operations. We discarded this from the start and we embraced the operand image solution. Essentially what we did was placing these extensions in the same operand image that contains the Postgres binaries. This is a typical approach of also the other operators. If you think about also Zalando we call it “the Spilo way”. Spilo contained all the software that would run with the Zalando operator.

Our approach was a bit different, in that we wanted lighter images, so we created a few flavors of images, and also selected some extensions that we placed in the images. But in general, we recommended to build custom images. We provided instructions and we’ve also provided the requirements to build container images. But as you can see, the complexity of the operational layer is quite high, it’s not reasonable to ask any user or any customer to build their own images.

This is how they look now, although this is changing as I was saying:

A stack of boxes with “Debian base image” at the top, then “PostgreSQL”, then “Barman Cloud”, and finally  three “Extension” boxes at the bottom.

You’ve got a base image, for example, the Debian base image. You deploy the Postgres binaries. Then — even right now though it’s changing — CloudNativePG requires Barman Cloud to be installed. And then we install the extensions that we think are needed. For example, I think we distribute pgAudit, if I recall correctly, pgvector and pg_failover_slots. Every layer you add, of course, the image is heavier and we still rely on packages for most extensions.

The problem is, you’ve got a cluster that is already running and you want, for example, to test an extension that’s just come out, or you want to deploy it in production. If that extension is not part of the images that we build, you have to build your own image. Because of the possible combinations of extensions that exist, it’s impossible to build all of these combinations. You could build, for example, a system that allows you to select what extensions you want and then build the image, but in our way of thinking, this was not the right approach. And then you’ve got system dependencies and, if an extension brings a vulnerability that affects the whole image and requires more updates — not just of the cluster, but also of the builds of the image.

We wanted to do something else, but we immediately faced some limitations of the technologies. One was on Postgres, the other one was on Kubernetes. In Postgres, extensions need to be placed in a single folder. It’s not possible to define multiple locations, but thanks to the work that Peter and this team have done, now we’ve got extension_control_path in version 18.

Kubernetes could not allow until, 10 days ago, to mount OCI artifacts as read-only volumes. There’s a new feature that is now part of Kubernetes 1.33 that allows us to do it.

This is the patch that I was talking about, by Peter Eisentraut. I’m really happy that CloudNativePG is mentioned as one of the use cases. And there’s also mentioned for the work that, me, David, and Marco and, primarily Marco and Niccolò from CloudNativePG have done.

This is the patch that introduced VolumeSource in Kubernetes 1.33.

The idea is that with Postgres 18 now we can set in the configuration where we can look up for extensions in the file system. And then, if there are libraries, we can also use the existing dynamic_library_path GUC.

So, you remember, this is where we come from [image above]; the good thing is we have the opportunity to build Postgres images that are minimal, that only contain Postgres.

Three stacks of boxes. On the left, “Debian base image” on top of “PostgreSQL”. On the right, “Debian base image” on top of “Barman Cloud”. On the lower right, a single box for an extension.

Instead of recreating them every week — because it’s very likely that something has some dependency, has a CVE, and so recreate them for everyone, forcing everyone to update their Postgres systems — we can now release them maybe once a month, and pretty much follow the Postgres cadence patch releases, and maybe if there are CVEs it’s released more frequently.

The other good thing is that now we are working to remove the dependency on Barman Cloud for CloudNativePG. CloudNativePG has a new plugin interface and with 1.26 with — which is expected in the next weeks — we are suggesting people start moving new workloads to the Barman Cloud plugin solution. What happens is that Barman Cloud will be in that sidecar image. So it will be distributed separately, and so its lifecycle is independent from the rest. But the biggest advantage is that any extension in Postgres can be distributed — right now we’ve got packages — The idea is that they are distributed also as images.

If we start thinking about this approach, if I write an extension for Postgres, until now I’ve been building only packages for Debian or for RPM systems. If I start thinking about also building container images, they could be immediately used by the new way of CloudNativePG to manage extensions. That’s my ultimate goal, let’s put it that way.

This is how things will change at run time without breaking immutability.

A box labeled “PostgreSQL Pod” with four separate boxes inside, labeled “Container Postgres”, “Sidecar Barman Cloud”, “Volume Extension 1”, and “Volume Extension 2”.

There will be no more need to think about all the possible combinations of extensions. There will be the Postgres pod that runs, for example, a primary or standby, that will have the container for Postgres. If you’re using Barman Cloud, the sidecar container managed by the plugin with Barman Cloud. And then, for every extension you have, you will have a different image volume that is read-only, very light, only containing the files distributed in the container image of the extension, and that’s all.

Once you’ve got these, we can then coordinate the settings for external extension_control_path and dynamic_library_path. What we did was, starting a fail fast pilot project within EDB to test the work that Peter was doing on the extension_control_path. For that we used the Postgres Trunk Containers project, which is a very interesting project that we have at CloudNativePG. Every day it rebuilds the latest snapshot of the master branch of Postgres so that we are able to catch, at an early stage, problems with the new version of Postgres in CloudNativePG. But there’s also an action that builds container images for a specific, for example, Commitfest patch. So we use that.

Niccolò wrote a pilot patch, an exploratory patch, for the operator to define the extensions stanza inside the cluster resource. He also built some bare container images for a few extensions. We make sure to include a very simple one and the most complex one, which is PostGIS. This is the patch that — it’s still a draft — and the idea is to have it in the next version, 1.27 for CloudNativePG. This is how it works:

apiVersion: postgresql.cnpg.io/v1
kind: Cluster
metadata:
  name: postgresql-with-extensions
spec:
  instances: 1
  imageName: ghcr.io/cloudnative-pg/postgresql-trunk:18-devel
  postgresql:
    extensions:
      - name: pgvector
        image:
          reference: ghcr.io/cloudnative-pg/pgvector-18-testing:latest
  storage:
    storageClass: standard
    size: 1Gi

We have the extensions section in the cluster definition. We name the extension. Theoretically we could also define the version and we point to the image. What’s missing in this pilot patch is support for image catalogs, but that’s something else that we can worry about later.

What happens under the hood is that when you update, or when you add a new extension in the cluster definition, a rolling update is initiated. So there’s this short downtime, but the container image is loaded in the replicas first, and then in the primary. n image volume is mounted for each extension in, let’s say, /extensions/$name_of_extension folder and CNPG updates, these two parameters. It’s quite clean, quite neat. It works, but most of the work needs to happen here. So that’s been my call, I mean to call container images as a first class artifacts. If these changes, we have a new way to distribute images.

Just to approach the conclusion, if you want to know more about the whole story, I wrote this blog article that recaps everything, and the key takeaway for me — and then we go more on the patch if you want to, and also address the questions. But what is important for me? Being in the Postgres community for a long time, I think this is a good way, a good moment for us to challenge the status quo of the extension distribution ecosystem.

I think we have an opportunity now to define a standard, which, I just want to be clear, I’m focusing myself primarily on CNPG, but this is in general, even for other operators. I’m sure that this will benefit everyone and overall it will reduce the waste that we collectively create when distributing these extensions in Kubernetes. If this becomes a standard way to distribute extensions, the benefits will be much better operational work for everyone, primarily also easier testing and validation of extensions. I mean, right now, if you see an extension, ideally that extension — and it’s very easy to build — if you’re in GitHub, to build the container images. GitHub, for example, already provides the whole infrastructure for you to easily build container images.

So if we find a standard way to define a GitHub action to build Postgres extensions, I think, if you’re a developer of an extension, you can just use it and then you find a registry in your project directly that continuously publishes or periodically publishes this extension. Any user can just reference that image URL and then without having to build images, they’re just one rolling update away from testing a patch, testing also the upgrade paths.

I think there are some unknown unknowns that kind of scare me, in general, about upgrades, upgrades of extensions. This is, in my opinion, one of the biggest issues. It’s not that they’re not solved, but they require more attention and more testing if you’re using them in an immutable world. All of these will, in my opinion, will be much, much better with the approach we’ve proposed. Images will be lighter, and the lighter image is also safer and more secure, so less prone to have CVEs,lLess prone to require frequent updates, and also they reduce the usage of bandwidth, for an organization in general. What I was saying before, any extension project can be fully independent, have their own way to build images and publish them.

One last point. I keep hearing many signs, that all of the stuff that we are proposing right now seem like a kind of a limitation of Kubernetes. The way I see it, in my view, that it’s not actually a limitation, it’s that these problems have never been addressed before. The biggest mistake we can do is focus on the specific problem of managing extensions without analyzing the benefits that the entire stack brings to an organization. Kubernetes brings a lot of benefits in terms of security, velocity, change management and, operations that any organization must consider right now. Any Postgres DBA, any Postgres user, my advice is, if you haven’t done it yet, start taking Kubernetes, seriously.

Discussion

Floor: I do think that David, you wanted to talk maybe a little bit about the mutable volume pattern?

David: Well, if people are interested, in your early slide where you were looking at alternatives, one you were thinking of was putting extensions on a mutable volume and you decided not to do that. But at Tembo we did do that and I did a bunch of work trying to improve it and try to minimize image size and all that in the last couple months. Tembo Cloud is shutting down now, so I had to stop before I finished it, but I made quite a bit of progress. I’m happy to kind of talk through the ideas there. But I think that this approach is a better long term solution, fundamentally.

Gabriele: I would like if Marco and Niccolò, if you want to talk about the actual work you’ve done. Meanwhile, Peter asks, “why does an installation of an extension require a small downtime?” The reason is that at the moment, the image volume patch, if you add a new image volume, it requires the pod to restart. Nico or Marco, Jonathan, if you want to correct me on that.

Nico or Marco or Jonathan: It provides a rolling update of the cluster right now.

Gabriele: So that’s the reason. That’s the only drawback, but the benefits in my opinion, are…

David: My understanding is that, to add a new extension, it’s mounted it in a different place. And because every single extension is its own mount, you have to add it to both those GUCs. And at least one of them requires a restart.

Gabriele: But then for example, we’ve had this conversation at EDB for example, we’re planning to have flavors of predefined extensions. For example, you can choose a flavor and we distribute those extensions. For example, I dunno, for AI we place some AI kind of extensions in the same image, so it would be different.

But otherwise I’m considering the most extreme case of one extension, one container image, which in my opinion, for the open source world is the way that hopefully will happen. Because this way, think about that – I haven’t mentioned this — if I write an extension, I can then build the image and then run automated tests using Kubernetes to assess my extension on GitHub. If those tests fail, my commit will never be merged on main. This is trunk development, continuous delivery. This is, in my opinion, a far better way of delivering and developing software. This is, again, the reason why we ended up in Kubernetes. It’s not because it’s a technology we like, it’s a toy or so, it’s because it solves bigger problems than database problems.

Even when we talk about databases, there’s still work that needs to be done, needs to be improved. I’m really happy that we have more people that know Postgres nowadays that are joining CloudNativePG, and are elevating the discussions more and more on the database level. Because before it was primarily on Kubernetes level, but now we see people that know Postgres better than me get in CloudNativePG and propose new ideas, which is great. Which is the way it needs to be, in my opinion.

But I remember, Tembo approached us because we actually talked a lot with them. Jonathan, Marco, I’m sure that you recall, when they were evaluating different operators and they chose CloudNativePG. I remember we had these discussions where they asked us to break immutability and we said, “no way”. That’s why I think Tembo had to do the solution you described, because we didn’t want to do it upstream.

I think, to be honest, and to be fair, if image volumes were not added, we would’ve probably gone down that path, because this way of managing extensions, as I was saying, is not scalable, the current one. Because we want to always improve, I think that the approach we need to be critical on what we do. So, I don’t know, Niccolò, Marco, I would like you to, if you want, explain briefly.

[A bit of chatter, opened this Dockerfile.]

FROM ghcr.io/cloudnative-pg/postgresql-trunk:18-devel AS builder

USER 0

COPY . /tmp/pgvector

RUN set -eux; \
	mkdir -p /opt/extension && \
	apt-get update && \
	apt-get install -y --no-install-recommends build-essential clang-16 llvm-16-dev && \
	cd /tmp/pgvector && \
	make clean && \
	make OPTFLAGS="" && \
	make install datadir=/opt/extension/share/ pkglibdir=/opt/extension/lib/

FROM scratch

COPY --from=builder /opt/extension/lib/* /lib/
COPY --from=builder /opt/extension/share/extension/* /share/

Niccolò: I forked, for example, pgvector, That’s what we can do basically for every simple extensions that we can just build. This is a bit more complicated because we have to build from a trunk version of Postgres 18. So we have to compile pgvector from source, and then in a scratch layer we just archive the libraries and every other content that was previously built. But ideally whenever PG 18 comes out as a stable version of Postgres, we just need to apt install pgvector and grab the files from the path. Where it gets a bit more tricky is in the case of PostGIS, or TimescaleDB, or any extension whose library requires third party libraries. For example, PostGIS has a strong requirement on the geometric libraries, so you need to import them as well inside the mount volume. I can link you an example of the PostGIS one.

Gabriele: I think it’s important, we’ve got, I think Peter here, David as well, I mean, for example, if we could get standard ways in Postgres to generate Dockerfiles for extensions, that could be great. And as I said, these extensions can be used by any operator, not only CNPG.

David: That’s my POC does. It’s a patch against the PGXS that would build a trunk image.

Gabriele: This is the work that Niccolò had to do to make PostGIS work in the pilot project: he had to copy everything.

Niccolò: I think we can make it a little bit smoother and dynamically figure out everything from the policies library, so we don’t have to code everything like this, but this is just a proof of concept that it can work.

David: So you installed all those shared libraries that were from packages.

Niccolò: Yeah, they’re being copied in the same MountVolume where the actual extensions are copied as well. And then the pilot patch is able to set up the library path inside the pod so that it makes the libraries available to the system because of course, these libraries are only part of the MountVolume. They’re not injected inside the system libraries of the pod, so we have to set up the library path to make them available to Postgres. That’s how we’re able to use them.

David: So they end up in PKGLIBDIR but they still work.

Niccolò: Yeah.

Gabriele: I mean, there’s better ideas, better ways. As Niccolò also said, it was a concept.

David: Probably a lot of these shared libraries could be shared with other extensions. So you might actually want other OCI images that just have some of the libraries that shared between.

Gabriele: Yeah, absolutely. So we could work on a special kind of, extensions or even metadatas so that we can place, you know…

So, yeah, that’s it.

Jonathan: I think it’s important to invite everyone to try and test this, especially the Postgres trunk containers, when they want to try something new stuff, new like this one, just because we always need people testing. When more people review and test, it’s amazing. Because every time we release something, probably we’ll miss something, some extension like PostGIS missing one of the libraries that wasn’t included in the path. Even if we can try to find a way to include it, it will not be there. So testing, please! Test all the time!

Gabriele: Well, we’ve got this action now, they’re failing. I mean, it’s a bit embarrassing. [Cross talk.] We already have patch to fix it.

But I mean, this is a great project as I mentioned before, because it allows us to test the current version of Postgres, but also if you want to build from a Commitfest or if you’ve got your own Postgres repository with sources, you can compile, you can get the images from using this project.

Floor: Gabriele, did you want to talk about SBOMs?

Gabriele: I forgot to mention Software Bill of Materials. They’re very important. It’s kind of now basic for any container image. There’s also the possibility to add them to these container images too. This is very important. Again, in a change manager for security and all of that — in general supply chain. And signatures too. But we’ve got signature for packages as well. There’s also a attestation of provenance.

Floor: Very good, thanks everyone!

Mini Summit 5: Extension Management in CNPG

Orange card with large black text reading “Extension Management in CNPG”. Smaller text below reads “Gabriele Bartolini (EDB)” and that is the date, “05.07.2025”.

The last Extension Ecosystem Mini-Summit is upon us. How did that happen?

Join us for a virtual conference session featuring Gabriele Bartolini, who will be discussing Extension Management in CNPG. I’m psyched for this one, as the PostgresSQL community has contributed quite a lot to improving extensions management in CloudNativePG in the past year, some of which we covered in previously. If you miss it, the video, slides, and transcript will appear here soon.

Though it may be a week or two to get the transcripts done, considering that PGConf.dev is next week, and featuring the Extension Ecosystem Summit on Tuesday, 13 May in Montreál, CA. Hope to see you there; be sure to say “hi!”