Using and configuring unattended-upgrades under Debian Bookworm - Part 1: Preparations
Photo by Markus Winkler: https://www.pexels.com/photo/wood-writing-mathematics-typography-19915915/
Preface
At the time of this writing Debian Bookworm is the stable release of Debian. It utilizes Systemd timers for all automation tasks like the updating of the package lists and execution of the actual apt-get upgrade. Therefore we won't need to configure APT-Parameters in files like /etc/apt/apt.conf.d/02periodic. In fact some of these files don't even exist on my systems. Keep that in mind if you read this article along with others, who might do things differently - or for older/newer releases of Debian.
Part 2 can be found here: Using and configuring unattended-upgrades under Debian Bookworm - Part 2: Practise
unattended-upgrades is not your enemy
I've read and heard my fair share of people condemning unattended-upgrades as it:
- Made things unnecessary complicated
- Broke the system
- Ignored APT-Pinnings and caused problems because of that
- Just didn't work
- Caused problems with High Availability (HA) services
- Made the database fail
- Split-Brains occurred only due to unattended-upgrades restarting services
- Restarted services during work hours causing disruptions and annoyances
You should get the gist. And then there is my experience, running unattended-upgrades on over 6000 Debian servers running all sort of services. From Apaches, Tomcats, JBoss and HAProxy to DRBD, GlusterFS, Corosync, Pacemaker some NoSQL-DBs and services like keepalived and ipvsadm for Layer2/3-Loadbalancing or Quagga/FRR for BGP-Route announcements. And in all those years unattended-upgrades wasn't the root cause of a single problem, outage or incident.
Therefore I suspect that many of the people whose complaints I've read just installed it and never cared about it again. Or, at least, didn't care enough. Presumably they just wanted a quick solution to help them get their systems updated. Well yes, unattended-upgrades is the right tool for this. But as always:
Know your system and plan accordingly!
Things to consider beforehand
Let me give you a list of things to consider before you even install unattended-upgrades.
- Are all of your APT-Repositories in order?
- GPG-keys valid and present?
- Correct repository for your Debian release?
- Do you use only official Debian repositories?
- Do you use Debian repositories from other projects/vendors?
- Are your repositories in sync on all your servers?
- Preferably you use the same Debian Mirror on all of your systems. At least make sure you don't use different repositories for security and system updates across all your Debian systems.
- This prevents problems from outdated mirrors. Happens rarely, but can happen.
- Trust me: When 2 machines out of a cluster of 5 behave differently, your first thought or troubleshooting step won't be to check if the different Apt-Repositories have the same content
- Remember: Those are often provided as-is by volunteers. They are NOT a commercial service. And yet most of these volunteers do an exceptional & awesome job despite not being paid for it.
- Best case scenario: Your company has an internal Debian Mirror. Saving you money on bandwidth usage.
- Have a look at https://www.debian.org/mirror/ftpmirror.en.html or Aptly if you plan on setting up a mirror.
- Basically: If an
apt-get updateprints out anything other then your configured repositories, followed by the lineReading package lists... Done: Fix your repositories! - Yes, you can have vendors with broken Debian repositories. Most often the Release file will be buggy or missing at all. If you can't get your vendor to fix it, well, then the best option is to not specify the repository at all and ensure you have another form of automation to update those packages.
- Sadly a
wget http://some-company.tld/some.deb && dpkg -i some.debis still considered valuable quality work at far too many enterprise software companies out there..- Or you just work around that nuisance, create an internal repository yourself and upload the packages there.
- A good time to remind them that you pay them and that you need a fully working Debian repository, following the Debian guidelines so you can automatically patch all your systems.
- Sadly a
- Do NOT continue otherwise. You have been warned.
- What kind of services does your system provide?
- Here it is essential to know what the system does. Technically and organizationally.
- What services are provided, why, to whom?
- Are there any Service Level Agreements (SLAs) in place?
- Do downtimes of a service need to be scheduled first?
- Or can service restarts happen at any time?
- Also automatically? Or is human supervision required by law/standards? (Looking at you PCI DSS (Wikipedia) 😉)
- Or is there already an agreed upon maintenance window during which the service is allowed to be unresponsive?
- Do all services have High Availability (HA)?
- Do the service(s) survive if the primary/master/main system is shut down?
- How many systems can be unavailable at the same time?
- Or do other tasks need to be done (manually) on secondary/slave/standby systems?
- Does your failover work?
- Is the failover tested regularly?
- How are you informed if the service(s) stop responding?
- Is your monitoring set to check right after updates did happen? (You can specify the time when unattended-upgrades does install the updates.)
- This will enable you to classify all your installed packages into the following categories for unattended-upgrades:
- Can be updated anytime
- Can be updated at certain times / Only a certain number of systems is allowed to be unavailable at any time
- See:
systemctl list-timersespeciallyapt-daily-upgrade.timerand maybeapt-daily.timer
- See:
- Only manual updates (Blacklisting)
Unattended-Upgrade::Package-Blacklistin/etc/apt/apt.conf.d/50unattended-upgradesis your friend- Then execute
apt-get update && apt-get upgrademanually to update the blacklisted packages
- Never update this package / We need a specific version
- This will need blacklisting AND APT-Pinnings to be in place
- As an
apt-get update && apt-get upgradecan still be executed manually
- In certain situations (HA-Nodes, manually triggered fail-over, etc.) you will need to run specific commands which need to be executed before or after an package update.
- This is something unattended-upgrades can't help you with
- Sometimes these are commands which should be included in the Debian package itself. Namely in the preinst, prerm, postinst and postrm files. The so-called package maintainer scripts (Official Debian documentation).
- But more often the commands are customer/environment specific and adding them to the maintainer script makes no sense or can even cause disruptions in the service.
- A feasible workaround is the utilisation of a drop-in file if your service is started via a Systemd Unit-File, see https://www.freedesktop.org/software/systemd/man/latest/systemd.unit.html and search for "drop-in". The ArchWiki also has a good article: https://wiki.archlinux.org/title/systemd#Drop-in_files - but keep in mind that Arch is based on Gentoo and hence maybe has different paths than Debian
- If the correct procedure is missing from the package, or isn't suitable for your environment.. Then its best to exclude the packages from unattended-upgrades and get yourself some tool like Ansible, Puppet or Rundeck to automate the execution of manual tasks. (Good how I love Rundeck. 🥰) There you are able to ensure everything is valid and verified before switching your primary cluster node and running package updates.
- If you have some kind of ITIL ChangeManagement process in place this, of course, also can't be checked by unattended-upgrades
- The only viable solution would be to have different repositories based on your environment classifications (development, QA testing, production) or approval classifications (untested, testing, approved) and push packages to the corresponding repository once the ChangeManagement process is completed
- Then you can install the updates automatically from there
- Unattended-Upgrades is great! But you can't automate every single scenario by just using it!
- Sometimes this even means to go back to the drawing board and optimise your internal company processes first. Especially in situations where approvals have to be given manually by real humans
- When you start rolling out unattended-upgrades it's better to have an overview first on how many patches are missing on each server. The more updates you install, the more likely it is that things will change or even break. Or may it even just those informal log messages that some parameter or option will be deprecated in the next major release.
- I advise to start with non-critical systems first
- Ideally systems which are somewhat up-to-date so you can identify & troubleshoot problems easier
- It's perfectly fine to update all systems manually first and only then enable unattended-upgrades
- As then you will have a common ground for all your systems and bugs are easier to identify as the same bug will affect all systems at somewhat the same time
- You need to have a quick & easy, bureaucracy-free process to add packages to the blacklist
- There once were broken corosync and keepalived packages in Debian in the early 2010s
- Once we saw that, we immediately added them to the blacklist and downgraded the other systems where the update was already installed
- You don't want to spent one hour on the phone frantically trying to reach a member of your Change Advisory Board to greenlight the change which lets you modify the blacklist while unattended-upgrades happily wrecks one system after another
- Point 8 sets a requirement: How do you roll out a new blacklist/config for unattended-upgrades to 6000 servers in under 15 minutes?
- You are too slow to do that manually, no matter how fast you can type
for HOST in $(cat host-list.txt); do ssh $HOST some-command; done? Yeah... No. It works, sure... But.. Come on, really? And this does only one host at a time.. Still taking too long.- You do want something like Puppet, Ansible or Rundeck for this very reason.
- Change the configuration in hiera
- Commit it to git
- Log in to Rundeck and execute the "Do a manually triggered Puppet/Ansible run" job
- Then drink a coffee while you watch Rundeck doing it's job and can care for the 10 or so servers who will fail for various other reasons.
Understanding apt-cache policy
The next thing that will help you vastly in providing a smoothly running unattended-upgrades service is understanding and using the apt-cache tool, especially with the policy argument: apt-cache policy. Even more if you use Debian-Repositories from other projects or vendors.
Why? unattended-upgrades comes with some default configured Origins-Pattern in /etc/apt/apt.conf.d/50unattended-upgrades. These are useable for the official Debian Security updates and when new point releases of Debian are published (12.5, 12.6, etc.). These incorporate all updates since the last point release. Non-Security Updates for installed packages between point releases won't be installed with the default configuration.
If you want these when they are ready, you have to uncomment the line for "origin=Debian,codename=${distro_codename}-updates";.
${distro_codename}-proposed-updates are updates which may not be stable yet. Read https://wiki.debian.org/StableProposedUpdates for the details. Personally I keep it disabled. Especially on production systems.
For Debian Bookworm the default Origins-Pattern are:
Unattended-Upgrade::Origins-Pattern {
[...]
// "origin=Debian,codename=${distro_codename}-updates";
// "origin=Debian,codename=${distro_codename}-proposed-updates";
"origin=Debian,codename=${distro_codename},label=Debian";
"origin=Debian,codename=${distro_codename},label=Debian-Security";
"origin=Debian,codename=${distro_codename}-security,label=Debian-Security";
[...]
};
This means only packages matching these Origins-Pattern will be updated. Therefore you have to verify that these patterns include all packages you want to update and, additionally, that packages you do not wish to update are excluded. Although the use of the blacklist might be easier here, as this simply works on the name of the package.
Also note that codename=${distro_codename} is always part of an Origins-Pattern. It is important to not just use terms like stable or oldstable as these do not identify a unique distribution. Those change whenever a new major Debian release happens. (This is also true if you are referencing OS images in things like Docker. debian:latest is most certainly a bad idea, debian-bookworm:latest is better.)
Matching Origin-Patterns to apt-cache policy's output
How do you translate these Origins-Patterns into the lines apt-cache policy gives us?
Short side-note:
apt-cacheuses the metadata (repository information) obtained viaapt-get update. Therefore it also works if the configured repositories are offline, but can also show outdated data if you haven't updated the package list information viaapt-get updaterecently.
If executed you will see output like the following:
root@lanadmin:~# apt-cache policy
Package files:
100 /var/lib/dpkg/status
release a=now
500 http://debian.tu-bs.de/debian bookworm-updates/non-free-firmware amd64 Packages
release v=12-updates,o=Debian,a=stable-updates,n=bookworm-updates,l=Debian,c=non-free-firmware,b=amd64
origin debian.tu-bs.de
500 http://debian.tu-bs.de/debian bookworm-updates/main amd64 Packages
release v=12-updates,o=Debian,a=stable-updates,n=bookworm-updates,l=Debian,c=main,b=amd64
origin debian.tu-bs.de
500 http://security.debian.org/debian-security bookworm-security/non-free-firmware amd64 Packages
release v=12,o=Debian,a=stable-security,n=bookworm-security,l=Debian-Security,c=non-free-firmware,b=amd64
origin security.debian.org
500 http://security.debian.org/debian-security bookworm-security/main amd64 Packages
release v=12,o=Debian,a=stable-security,n=bookworm-security,l=Debian-Security,c=main,b=amd64
origin security.debian.org
500 http://debian.tu-bs.de/debian bookworm/non-free-firmware amd64 Packages
release v=12.5,o=Debian,a=stable,n=bookworm,l=Debian,c=non-free-firmware,b=amd64
origin debian.tu-bs.de
500 http://debian.tu-bs.de/debian bookworm/main amd64 Packages
release v=12.5,o=Debian,a=stable,n=bookworm,l=Debian,c=main,b=amd64
origin debian.tu-bs.de
Pinned packages:
root@lanadmin:~#For the sake of easiness, let us look at just this single line:
500 http://security.debian.org/debian-security bookworm-security/non-free-firmware amd64 Packages
release v=12,o=Debian,a=stable-security,n=bookworm-security,l=Debian-Security,c=non-free-firmware,b=amd64
origin security.debian.orgI wont go over each and every single listed value. If you want to know more: man apt_preferences probably has all the details and apt-cache policy packagename lists all policies for a single package.
We want to pay attention to the 2nd line starting with release. Here we have the relevant values. These are defined in the Release-File for each Debian Release/Repository. A documentation can be found here: https://wiki.debian.org/DebianRepository/Format#A.22Release.22_files
But what do they mean? man apt_preferences also explains them, but as they are relevant, let's make a short table.
| Field | Alias | unattended-upgrades variable | Description |
| Version | v | N/A | Contains the release version |
| Origin | o | ${distro_id} |
Originator of the packages. (Debian if used for packages from the Debian project. If you have commercial packages most likely the company or product name will show up here.) |
| Archive or Suite | a | N/A | Names the archive to which all the packages in the directory tree belong (on the repository server) |
| Codename | n | ${distro_codename} |
Codename of the Debian release. In our case: bookworm |
| Label | l | N/A | Names the label of the packages in the directory tree of the Release file. |
| Component | c | N/A | The licensing component associated with the packages in the directory tree. You may know this as: main, contrib, non-free-firmware, etc. |
| Architecture | b | N/A | The processor architecture for which a package is compiled. (amd64, i386, arm, etc.) |
| Site | N/A | N/A | FQDN for the APT-Repository. Useful when fields like Archive/Suite/Label are empty or ambiguous. (Example with component: "site=www.example.com,component=main";) |
All this information is usually listed in the first 30 lines of /etc/apt/apt.conf.d/50unattended-upgrades. Therefore: Read it, to ensure you get the currently valid information.
Understanding this makes it easy to match the configured Origin-Patterns to your configure Debian-Repositories.
If you are in doubt: Browse your Debian repository via HTTP/HTTPS and have a look at the Release file, for example: http://debian.tu-bs.de/debian/dists/bookworm/Release. The first 5 lines are:
Origin: DebianLabel: DebianSuite: stableVersion: 12.5Codename: bookworm
Filling in the variables we see that only the following Origin-Pattern matches:
"origin=Debian,codename=${distro_codename},label=Debian";
All others either have a different label and/or codename.
Looking at http://debian.tu-bs.de/debian/dists/bookworm-updates/Release we can see that this matches the following commented out Origin-Pattern:
"origin=Debian,codename=${distro_codename}-updates";
This means:
http://debian.tu-bs.de/debian/dists/bookworm/ holds all packages for the current point release of Debian Bookworm.
http://debian.tu-bs.de/debian/dists/bookworm-updates/ has all published updates which came out before a new point release is made. That is the reason why I always enable this repository too. But depending on your operating strategy going only with updates when a new point release is published is also fine.
Security updates are always published via the http://security.debian.org/debian-security/ repository and will be installed when available.
Key points / Story time
What you should have understood is:
- Each repository must be uniquely identifiable.
- This means: Each Origins-Pattern should only match one of all your configured repositories
- Of course you can have multiple repositories matching the same Origins-Pattern, but keep the possible implications in mind!
- Packages that share the same name must have the same content
- And by content I mean: Their hashsums must be identical for each given version
If it doesn't you are potentially in for a wild ride.
At a former company we used many internal repositories. And this was fine for a long time. Suddenly some developer started pushing Debian packages with the exact same name as official Debian packages to those internal repositories. We immediately complained. Laying out how that could wreck havoc on our infrastructure as we have to use those repositories and at the same time we can't blacklist that package, as blacklisting works solely on name of a package.
You can't do things like: "Blacklist package test-foo, but only if its coming from repository on repo.coolhost.tld"
We urged him to simply rename those packages or upload them to another repository - as those packages had to share the same name as they contained a not-yet included fix for a bug the company encountered. He wouldn't as he saw no problem with his approach. "Just don't use them." (Yeah thanks.. That's not how it works with automation.. Especially not if you - sort of - hijack repositories which are used for something entirely different..)
The workaround we made was by utilizing the apt priority for each repository, so packages from our internal official Debian mirror took priority. And that worked, but it was still annoying.
Some weeks later those packages caused an incident and in the root cause analysis the problem was identified and those packages were moved to a separate repository.
Lesson learned. Care for your repositories.
And this ends the first part of this post. The next part will focus on the practical side. We will look at the Systemd Unit- and Timer-File, how we can add new repositories to unattended-upgrades, for example to also upgrade our Proxmox installation using the Proxmox Debian repositories, how to blacklist packages and more.
