Search Results: "ghe"

10 September 2025

John Goerzen: ARM is great, ARM is terrible (and so is RISC-V)

I ve long been interested in new and different platforms. I ran Debian on an Alpha back in the late 1990s and was part of the Alpha port team; then I helped bootstrap Debian on amd64. I ve got somewhere around 8 Raspberry Pi devices in active use right now, and the free NNCPNET Internet email service I manage runs on an ARM instance at a cloud provider. ARM-based devices are cheap in a lot of ways: they use little power and there are many single-board computers based on them that are inexpensive. My 8-year-old s computer is a Raspberry Pi 400, in fact. So I like ARM. I ve been looking for ARM devices that have accelerated AES (Raspberry Pi 4 doesn t) so I can use full-disk encryption with them. There are a number of options, since ARM devices are starting to go more mid-range. Radxa s ROCK 5 series of SBCs goes up to 32GB RAM. The Orange Pi 5 Max and Ultra have up to 16GB RAM, as does the Raspberry Pi 5. Pine64 s Quartz64 has up to 8GB of RAM. I believe all of these have the ARM cryptographic extensions. They re all small and most are economical. But I also dislike ARM. There is a terrible lack of standardization in the ARM community. They say their devices run Linux, but the default there is that every vendor has their own custom Debian fork, and quite likely kernel fork as well. Most don t maintain them very well. Imagine if you were buying x86 hardware. You might have to manage AcerOS, Dellbian, HPian, etc. Most of them have no security support (particularly for the kernel). Some are based on Debian 11 (released in 2021), some Debian 12 (released in 2023), and none on Debian 13 (released a month ago). That is exactly the situation we have on ARM. While Raspberry Pi 4 and below can run Debian trixie directly, Raspberry Pi has not bothered to upstream support for the Pi 5 yet, and Raspberry Pi OS is only based on Debian bookworm (released in 2023) and very explicitly does not support a key Debian feature: you can t upgrade from one Raspberry Pi OS release to the next, so it s a complete reinstall every 2 years instead of just an upgrade. OrangePiOS only supports Debian bookworm but notably, their kernel is mostly stuck at 5.10 for every image they have (bookworm shipped with 6.1 and bookworm-backports supports 6.12). Radxa has a page on running Debian on one specific board, they seem to actually not support Debian directly, but rather their fork Radxa OS. There s a different installer for every board; for instance, this one for the Rock 4D. Looking at it, I can see that it uses files from here and here, with custom kernel, gstreamer, u-boot, and they put zfs in main for some reason. From Pine64, the Quartz64 seems to be based on an ancient 4.6 or 4.19 kernel. Perhaps, though, one might be able to use Debian s Pine A64+ instructions on it. Trixie doesn t have a u-boot image for the Quartz64 but it does have device tree files for it. RISC-V seems to be even worse; not only do we have this same issue there, but support in trixie is more limited and so is performance among the supported boards. The alternative is x86-based mini PCs. There are a bunch based on the N100, N150, or Celeron. Many of them support AES-NI and the prices are roughly in line with the higher-end ARM units. There are some interesting items out there; for instance, the Radxa X4 SBC features both an N100 and a RP2040. Fanless mini PCs are available from a number of vendors. Companies like ZimaBoard have interesting options like the ZimaBlade also. The difference in power is becoming less significant; it seems the newer ARM boards need 20W or 30W power supplies, and that may put them in the range of the mini PCs. As for cost, the newer ARM boards need a heat sink and fan, so by the time you add SBC, fan, storage, etc. you re starting to get into the price range of the mini PCs. It is great to see all the options of small SBCs with ARM and RISC-V processors, but at some point you ve got to throw up your hands and go this ecosystem has a lot of problems and consider just going back to x86. I m not sure if I m quite there yet, but I m getting close. Update 2025-09-11: I found a performant encryption option for the Pi 4, but was stymied by serial console problems; see the update post.

9 September 2025

John Goerzen: btrfs on a Raspberry Pi

I m something of a filesystem geek, I guess. I first wrote about ZFS on Linux 14 years ago, and even before I used ZFS, I had used ext2/3/4, jfs, reiserfs, xfs, and no doubt some others. I ve also used btrfs. I last posted about it in 2014, when I noted it has some advantages over ZFS, but also some drawbacks, including a lot of kernel panics. Since that comparison, ZFS has gained trim support and btrfs has stabilized. The btrfs status page gives you an accurate idea of what is good to use on btrfs. Background: Moving towards ZFS and btrfs I have been trying to move everything away from ext4 and onto either ZFS or btrfs. There are generally several reasons for that:
  1. The checksums for every block help detect potential silent data corruption
  2. Instant snapshots make consistent backups of live systems a lot easier, and without the hassle and wasted space of LVM snapshots
  3. Transparent compression and dedup can save a lot of space in storage-constrained environments
For any machine with at least 32GB of RAM (plus my backup server, which has only 8GB), I run ZFS. While it lacks some of the flexibility of btrfs, it has polish. zfs list -o space shows a useful space accounting. zvols can be behind VMs. With my project simplesnap, I can easily send hourly backups with ZFS, and I choose to send them over NNCP in most cases. I have a few VMs in the cloud (running Debian, of course) that I use to host things like this blog, my website, my gopher site, the quux NNCP public relay, and various other things. In these environments, storage space can be expensive. For that matter, so can RAM. ZFS is RAM-hungry, so that rules out ZFS. I ve been running btrfs in those environments for a few years now, and it s worked out well. I do async dedup, lzo or zstd compression depending on the needs, and the occasional balance and defrag. Filesystems on the Raspberry Pi I run Debian trixie on all my Raspberry Pis; not Raspbian or Raspberry Pi OS for a number of reasons. My 8-yr-old uses a Raspberry Pi 400 as her primary computer and loves it! She doesn t do web browsing, but plays Tuxpaint, some old DOS games like Math Blaster via dosbox, and uses Thunderbird for a locked-down email account. But it was SLOW. Just really, glacially, slow, especially for Thunderbird. My first step to address that was to get a faster MicroSD card to hold the OS. That was a dramatic improvement. It s still slow, but a lot faster. Then, I thought, maybe I could use btrfs with LZO compression to reduce the amount of I/O and speed things up further? Analysis showed things were mostly slow due to I/O, not CPU, constraints. The conversion Rather than use the btrfs in-place conversion from ext4, I opted to dar it up (like tar), run mkfs.btrfs on the SD card, then unpack the archive back onto it. Easy enough, right? Well, not so fast. The MicroSD card is 128GB, and the entire filesystem is 6.2GB. But after unpacking 100MB onto it, I got an out of space error. btrfs has this notion of block groups. By default, each block group is dedicated to either data or metadata. btrfs fi df and btrfs fi usage will show you details about the block groups. btrfs allocates block groups greedily (the ssd_spread mount option I use may have exacerbated this). What happened was it allocated almost the entire drive to data block groups, trying to spread the data across it. It so happened that dar archived some larger files first (maybe /boot), so btrfs was allocating data and metadata blockgroups assuming few large files. But then it started unpacking one of the directories in /usr with lots of small files (maybe /usr/share/locale). It quickly filled up the metadata block group, and since the entire SD card had been allocated to different block groups, I got ENOSPC. Deleting a few files and running btrfs balance resolved it; now it allocated 1GB to metadata, which was plenty. I re-ran the dar extract and now everything was fine. See more details on btrfs balance and block groups. This was the only btrfs problem I encountered. Benchmarks I timed two things prior to switching to btrfs: how long it takes to boot (measured from the moment I turn on the power until the moment the XFCE login box is displayed), and how long it takes to start Thunderbird. After switching to btrfs with LZO compression, somewhat to my surprise, both measures were exactly the same! Why might this be? It turns out that SD cards are understood to be pathologically bad with random read performance. Boot and Thunderbird both are likely doing a lot of small random reads, not large streaming reads. Therefore, it may be that even though I have reduced the total I/O needed, the impact is unsubstantial because the real bottleneck is the seeks across the disk. Still, I gain the better backup support and silent data corruption prevention, so I kept btrfs. SSD mount options and MicroSD endurance btrfs has several mount options specifically relevant to SSDs. Aside from the obvious trim support, they are ssd and ssd_spread. The documentation on this is vague and my attempts to learn more about it found a lot of information that was outdated or unsubstantiated folklore. Some reports suggest that older SSDs will benefit from ssd_spread, but that it may have no effect or even a harmful effect on newer ones, and can at times cause fragmentation or write amplification. I could find nothing to back this up, though. And it seems particularly difficult to figure out what kind of wear leveling SSD firmware does. MicroSD firmware is likely to be on the less-advanced side, but still, I have no idea what it might do. In any case, with btrfs not updating blocks in-place, it should be better than ext4 in the most naive case (no wear leveling at all) but may have somewhat more write traffic for the pathological worst case (frequent updates of small portions of large files). One anecdotal report I read and can t find anymore, somehow was from a person that had set up a sort of torture test for SD cards, with reports that ext4 lasted a few weeks or months before the MicroSDs failed, while btrfs lasted years. If you are looking for a MicroSD card, by the way, The Great MicroSD Card Survey is a nice place to start. For longevity: I mount all my filesystems with noatime already, so I continue to recommend that. You can also consider limiting the log size in /etc/systemd/journald.conf, running daily fstrim (which may be more successful than live trims in all filesystems). Conclusion I ve been pretty pleased with btrfs. The concerns I have today relate to block groups and maintenance (periodic balance and maybe a periodic defrag). I m not sure I d be ready to say put btrfs on the computer you send to someone that isn t Linux-savvy because the chances of running into issues are higher than with ext4. Still, for people that have some tech savvy, btrfs can improve reliability and performance in other ways.

6 September 2025

Reproducible Builds: Reproducible Builds in August 2025

Welcome to the August 2025 report from the Reproducible Builds project! Welcome to the latest report from the Reproducible Builds project for August 2025. These monthly reports outline what we ve been up to over the past month, and highlight items of news from elsewhere in the increasingly-important area of software supply-chain security. If you are interested in contributing to the Reproducible Builds project, please see the Contribute page on our website. In this report:

  1. Reproducible Builds Summit 2025
  2. Reproducible Builds and live-bootstrap at WHY2025
  3. DALEQ Explainable Equivalence for Java Bytecode
  4. Reproducibility regression identifies issue with AppArmor security policies
  5. Rust toolchain fixes
  6. Distribution work
  7. diffoscope
  8. Website updates
  9. Reproducibility testing framework
  10. Upstream patches

Reproducible Builds Summit 2025 Please join us at the upcoming Reproducible Builds Summit, set to take place from October 28th 30th 2025 in Vienna, Austria!** We are thrilled to host the eighth edition of this exciting event, following the success of previous summits in various iconic locations around the world, including Venice, Marrakesh, Paris, Berlin, Hamburg and Athens. Our summits are a unique gathering that brings together attendees from diverse projects, united by a shared vision of advancing the Reproducible Builds effort. During this enriching event, participants will have the opportunity to engage in discussions, establish connections and exchange ideas to drive progress in this vital field. Our aim is to create an inclusive space that fosters collaboration, innovation and problem-solving. If you re interesting in joining us this year, please make sure to read the event page which has more details about the event and location. Registration is open until 20th September 2025, and we are very much looking forward to seeing many readers of these reports there!

Reproducible Builds and live-bootstrap at WHY2025 WHY2025 (What Hackers Yearn) is a nonprofit outdoors hacker camp that takes place in Geestmerambacht in the Netherlands (approximately 40km north of Amsterdam). The event is organised for and by volunteers from the worldwide hacker community, and knowledge sharing, technological advancement, experimentation, connecting with your hacker peers, forging friendships and hacking are at the core of this event . At this year s event, Frans Faase gave a talk on live-bootstrap, an attempt to provide a reproducible, automatic, complete end-to-end bootstrap from a minimal number of binary seeds to a supported fully functioning operating system . Frans talk is available to watch on video and his slides are available as well.

DALEQ Explainable Equivalence for Java Bytecode Jens Dietrich of the Victoria University of Wellington, New Zealand and Behnaz Hassanshahi of Oracle Labs, Australia published an article this month entitled DALEQ Explainable Equivalence for Java Bytecode which explores the options and difficulties when Java binaries are not identical despite being from the same sources, and what avenues are available for proving equivalence despite the lack of bitwise correlation:
[Java] binaries are often not bitwise identical; however, in most cases, the differences can be attributed to variations in the build environment, and the binaries can still be considered equivalent. Establishing such equivalence, however, is a labor-intensive and error-prone process.
Jens and Behnaz therefore propose a tool called DALEQ, which:
disassembles Java byte code into a relational database, and can normalise this database by applying Datalog rules. Those databases can then be used to infer equivalence between two classes. Notably, equivalence statements are accompanied with Datalog proofs recording the normalisation process. We demonstrate the impact of DALEQ in an industrial context through a large-scale evaluation involving 2,714 pairs of jars, comprising 265,690 class pairs. In this evaluation, DALEQ is compared to two existing bytecode transformation tools. Our findings reveal a significant reduction in the manual effort required to assess non-bitwise equivalent artifacts, which would otherwise demand intensive human inspection. Furthermore, the results show that DALEQ outperforms existing tools by identifying more artifacts rebuilt from the same code as equivalent, even when no behavioral differences are present.
Jens also posted this news to our mailing list.

Reproducibility regression identifies issue with AppArmor security policies Tails developer intrigeri has tracked and followed a reproducibility regression in the generation of AppArmor policy caches, and has identified an issue with the 4.1.0 version of AppArmor. Although initially tracked on the Tails issue tracker, intrigeri filed an issue on the upstream bug tracker. AppArmor developer John Johansen replied, confirming that they can reproduce the issue and went to work on a draft patch. Through this, John revealed that it was caused by an actual underlying security bug in AppArmor that is to say, it resulted in permissions not (always) matching what the policy intends and, crucially, not merely a cache reproducibility issue. Work on the fix is ongoing at time of writing.

Rust toolchain fixes Rust Clippy is a linting tool for the Rust programming language. It provides a collection of lints (rules) designed to identify common mistakes, stylistic issues, potential performance problems and unidiomatic code patterns in Rust projects. This month, however, Sosth ne Gu don filed a new issue in the GitHub requesting a new check that would lint against non deterministic operations in proc-macros, such as iterating over a HashMap .

Distribution work In Debian this month: Lastly, Bernhard M. Wiedemann posted another openSUSE monthly update for their work there.

diffoscope diffoscope is our in-depth and content-aware diff utility that can locate and diagnose reproducibility issues. This month, Chris Lamb made the following changes, including preparing and uploading versions, 303, 304 and 305 to Debian:
  • Improvements:
    • Use sed(1) backreferences when generating debian/tests/control to avoid duplicating ourselves. [ ]
    • Move from a mono-utils dependency to versioned mono-devel mono-utils dependency, taking care to maintain the [!riscv64] architecture restriction. [ ]
    • Use sed over awk to avoid mangling dependency lines containing = (equals) symbols such as version restrictions. [ ]
  • Bug fixes:
    • Fix a test after the upload of systemd-ukify version 258~rc3. [ ]
    • Ensure that Java class files are named .class on the filesystem before passing them to javap(1). [ ]
    • Do not run jsondiff on files over 100KiB as the algorithm runs in O(n^2) time. [ ]
    • Don t check for PyPDF version 3 specifically; check for >= 3. [ ]
  • Misc:
    • Update copyright years. [ ][ ]
In addition, Martin Joerg fixed an issue with the HTML presenter to avoid crash when page limit is None [ ] and Zbigniew J drzejewski-Szmek fixed compatibility with RPM 6 [ ]. Lastly, John Sirois fixed a missing requests dependency in the trydiffoscope tool. [ ]

Website updates Once again, there were a number of improvements made to our website this month including:

Reproducibility testing framework The Reproducible Builds project operates a comprehensive testing framework running primarily at tests.reproducible-builds.org in order to check packages and other artifacts for reproducibility. In August, however, a number of changes were made by Holger Levsen, including:
  • reproduce.debian.net-related:
    • Run 4 workers on the o4 node again in order to speed up testing. [ ][ ][ ][ ]
    • Also test trixie-proposed-updates and trixie-updates etc. [ ][ ]
    • Gather seperate statistics for each tested release. [ ]
    • Support sources from all Debian suites. [ ]
    • Run new code from the prototype database rework branch for the amd64-pull184 pseudo-architecture. [ ][ ]
    • Add a number of helpful links. [ ][ ][ ][ ][ ][ ][ ][ ][ ]
    • Temporarily call debrebuild without the --cache argument to experiment with a new version of devscripts. [ ][ ][ ]
    • Update public TODO. [ ]
  • Installation tests:
    • Add comments to explain structure. [ ]
    • Mark more old jobs as old or dead . [ ][ ][ ]
    • Turn the maintenance job into a no-op. [ ]
  • Jenkins node maintenance:
    • Increase penalties if the osuosl5 or ionos7 nodes are down. [ ]
    • Stop trying to fix network automatically. [ ]
    • Correctly mark ppc64el architecture nodes when down. [ ]
    • Upgrade the remaining arm64 nodes to Debian trixie in anticipation of the release. [ ][ ]
    • Allow higher SSD temperatures on the riscv64 architecture. [ ]
  • Debian-related:
    • Drop the armhf architecture; many thanks to Vagrant for physically hosting the nodes for ten years. [ ][ ]
    • Add Debian forky, and archive bullseye. [ ][ ][ ][ ][ ][ ][ ]
    • Document the filesystem space savings from dropping the armhf architecture. [ ]
    • Exclude i386 and armhfr from JSON results. [ ]
    • Update TODOs for when Debian trixie and forky have been released. [ ][ ]
  • tests.reproducible-builds.org-related:
  • Misc:
    • Detect errors with openQA erroring out. [ ]
    • Drop the long-disabled openwrt_rebuilder jobs. [ ]
    • Use qa-jenkins-dev@alioth-lists.debian.net as the contact for jenkins.debian.net. [ ]
    • Redirect reproducible-builds.org/vienna25 to reproducible-builds.org/vienna2025. [ ]
    • Disable all OpenWrt reproducible CI jobs, in coordination with the OpenWrt community. [ ][ ]
    • Make reproduce.debian.net accessable via IPv6. [ ]
    • Ignore that the megacli RAID controller requires packages from Debian bookworm. [ ]
In addition,
  • James Addison migrated away from deprecated toplevel deb822 Python module in favour of debian.deb822 in the bin/reproducible_scheduler.py script [ ] and removed a note on reproduce.debian.net note after the release of Debian trixie [ ].
  • Jochen Sprickerhof made a huge number of improvements to the reproduce.debian.net statistics calculation [ ][ ][ ][ ][ ][ ] as well as to the reproduce.debian.net service more generally [ ][ ][ ][ ][ ][ ][ ][ ].
  • Mattia Rizzolo performed a lot of work migrating scripts to SQLAlchemy version 2.0 [ ][ ][ ][ ][ ][ ] in addition to making some changes to the way openSUSE reproducibility tests are handled internally. [ ]
  • Lastly, Roland Clobus updated the Debian Live packages after the release of Debian trixie. [ ][ ]

Upstream patches The Reproducible Builds project detects, dissects and attempts to fix as many currently-unreproducible packages as possible. We endeavour to send all of our patches upstream where appropriate. This month, we wrote a large number of such patches, including:

Finally, if you are interested in contributing to the Reproducible Builds project, please visit our Contribute page on our website. However, you can get in touch with us via:

19 August 2025

Russell Coker: Colmi P80 SmartWatch First Look

I just bought a Colmi P80 SmartWatch from Aliexpress for $26.11 based on this blog post reviewing it [1]. The main things I was after in this was a larger higher resolution screen because my vision has apparently deteriorated during the time I ve been wearing a Pinetime [2] and I now can t read messages on it when not wearing my reading glasses. The watch hardware is quite OK. It has a larger and higher resolution screen and looks good. The review said that GadgetBridge (the FOSS SmartWatch software in the F-Droid repository) connected when told that the watch was a P79 and in a recent release got support for sending notifications. In my tests with GadgetBridge it doesn t set the time, can t seem to send notifications, can t read the battery level, and seems not to do anything other than just say connected . So I installed the proprietary app, as an aside it s a neat feature to have the watch display a QR code for installing the app, maybe InfiniTime should have a similar QR code for getting GadgetBridge from the F-Droid repository. The proprietary app is quote OK for the basic functionality and a less technical relative who is using one is happy. For my use the proprietary app is utterly broken. One of my main uses is to get notifications of Jabber messages from the Conversations app (that s in F-Droid). I have Conversations configured to always have a notification of how many accounts are connected which prevents Android from killing it, with GadgetBridge that notification isn t reported but the actual message contents are (I don t know how/why that happens) but with the Colmi app I get repeated notifcation messages on the watch about the accounts being connected. Also the proprietary app has on/off settings for messages to go to the watch for a hard coded list of 16 common apps and an Others setting for the rest. GadgetBridge lists the applications that are actually installed so I can configure it not to notify me about Reddit, connecting to my car audio, and many other less common notifications. I prefer the GadgetBridge option to have an allow-list for apps that I want notifications from but it also has a configuration option to use a deny list so you could have everything other than the app that gives lots of low value notifications. The proprietary app has a wide range of watch faces that it can send to the watch which is a nice feature that would be good to have in InfiniTime and GadgetBridge. The P80 doesn t display a code on screen when it is paired via Bluetooth so if you have multiple smart watches then you are at risk of connecting to the wrong one and there doesn t seem to be anything stopping a hostile party from connecting to one. Note that hostile parties are not restricted to the normal maximum transmission power and can use a high gain antenna for reception so they can connect from longer distances than normal Bluetooth devices. Conclusion The Colmi P80 hardware is quite decent, the only downside is that the vibration has an annoying tinny feel. Strangely it has a rotation sensor for a rotating button (similar to analogue watches) but doesn t seem to have a use for it as the touch screen does everything. The watch firmware is quite OK (not great but adequate) but lacking a password for pairing is a significant lack. The Colmi Android app has some serious issues that make it unusable for what I do and the release version of GadgetBridge doesn t work with it, so I have gone back to the PineTime for actual use. The PineTime cost twice as much, has less features (no sensor for O2 level in blood), but seems more solidly constructed. I plan to continue using the P80 with GadgetBridge and Debian based SmartWatch software to help develop the Debian Mobile project. I expect that at some future time GadgetBridge and the programs written for non-Android Linux distributions will support the P80 and I will transition to it. I am confident that it will work well for me at some future time and that I will get $26.11 of value from it. At this time I recommend that people who do the sort of things I do get one of each and that less technical people get a Colmi P80.

3 August 2025

Ben Hutchings: FOSS activity in July 2025

In July I attended DebCamp and DebConf in Brest, France. I very much enjoyed the opportunity to reconnect with other Debian contributors in person. I had a number of interesting and fruitful conversations there, besides the formally organised BoFs and talks. I also gave my own talk on What s new in the Linux kernel (and what s missing in Debian). Here s the usual categorisation of activity:

2 August 2025

Russell Coker: Server CPU Sockets

I am always looking for ways of increasing the compute power I have at a reasonable price. I am very happy with my HP z840 dual CPU workstation [1] that I m using as a server and my HP z640 single CPU workstation [2]. Both of them were available second hand at quite reasonable prices and could be cheaply upgraded to faster CPUs. But if I can get something a lot faster for a reasonable price then I ll definitely get it. Socket LGA2011-v3 The home server and home workstation I currently use have socket LGA2011-v3 [3] which supports the E5-2699A v4 CPU which gives a rating of 26,939 according to Passmark [4]. That Passmark score is quite decent, you can get CPUs using DDR4 RAM that go up to almost double that but it s a reasonable speed and it works in systems that are readily available at low prices. The z640 is regularly on sale for less than $400AU and the z840 is occasionally below $600. The Dell PowerEdge T430 is an ok dual-CPU tower server using the same socket. One thing that s not well known is that is it limited to something like 135W per CPU when run with two CPUs. So it will work correctly with a single E5-2697A v4 with 145W TDP (I ve tested that) but will refuse to boot with two of them. In my test system I tried replacing the 495W PSUs with 750W PSUs and it made no difference, the motherboard has the limit. With only a single CPU you only get 8/12 DIMM sockets and not all PCIe slots work. There are many second hand T430s on sale with only a single CPU presumably because the T330 sucks. My T430 works fine with a pair of E5-2683 v4 CPUs. The Dell PowerEdge T630 also takes the same CPUs but supports higher TDP than the T430. They also support 18*3.5 disks or 32*2.5 but they are noisy. I wouldn t buy one for home use. AMD There are some nice AMD CPUs manufactured around the same time and AMD has done a better job of making multiple CPUs that fit the same socket. The reason I don t generally use AMD CPUs is that they are used in a minority of the server grade systems so as I want ECC RAM and other server features I generally can t find AMD systems at a reasonable price on ebay etc. There are people who really want second hand server grade systems with AMD CPUs and outbid me. This is probably a region dependent issue, maybe if I was buying in the US I could get some nice workstations with AMD CPUs at low prices. Socket LGA1151 Socket LGA1151 [5] is used in the Dell PowerEdge T330. It only supports 2 memory channels and 4 DIMMs compared to the 4 channels and 8 DIMMs in LGA2011, and it also has a limit of 64G total RAM for most systems and 128G for some systems. By today s standards even 128G is a real limit for server use, DDR4 RDIMMs are about $1/GB and when spending $600+ on system and CPU upgrade you wouldn t want to spend less than $130 on RAM. The CPUs with decent performance for that socket like the i9-9900K aren t supported by the T330 (possibly they don t support ECC RAM). The CPUs that Dell supports perform very poorly. I suspect that Dell deliberately nerfed the T330 to drive sales of the T430. The Lenovo P330 uses socket LGA1151-2 but has the same issues of taking slow CPUs in addition to using UDIMMs which are significantly more expensive on the second hand market. Socket LGA2066 The next Intel socket after LGA2011-v3 is LGA2066 [6]. That is in The Dell Precision 5820 and HP Z4 G4. It takes an i9-10980XE for 32,404 on Passmark or a W-2295 for 30,906. The variant of the Dell 5820 that supports the i9 CPUs doesn t seem to support ECC RAM so it s not a proper workstation. The single thread performance difference between the W-2295 and the E5-2699A v4 is 2640 to 2055, a 28% increase for the W-2295. There are High Frequency Optimized cpus for socket LGA2011-v3 but they all deliver less than 2,300 on the Passmark single-thread tests which is much less than what you can get from socket LGA2066. The W-2295 costs $1000 on ebay and the E5-2699A v4 is readily available for under $400 and a few months ago I got a matched pair for a bit over $400. Note that getting a matched pair of Intel CPUs is a major pain [7]. Comparing sockets LGA2011-v3 and LGA2066 for a single-CPU system is a $300 system (HP x640) + $400 CPU (E5-2699A v4) vs $500 system (Dell Precision 5820) + $1000 CPU (W-2295), so more than twice the price for a 30% performance benefit on some tasks. The LGA2011-v3 and USB-C both launched in 2014 so LGA2011-v3 systems don t have USB-C sockets, a $20 USB-C PCIe card doesn t change the economics. Socket LGA3647 Socket LGA3647 [8] is used in the Dell PowerEdge T440. It supports 6 channels of DDR4 RAM which is a very nice feature for bigger systems. According to one Dell web page the best CPU Dell officially supports for this is the Xeon Gold 5120 which gives performance only slightly better than the E5-2683 v4 which has a low enough TDP that a T430 can run two of them. But according to another Dell web page they support 16 core CPUs which means performance better than a T430 but less than a HP z840. The T440 doesn t seem like a great system, if I got one cheap I could find a use for it but I wouldn t pay the prices that they go for on ebay. The Dell PowerEdge T640 has the same socket and is described as supporting up to 28 core CPUs. But I anticipate that it would be as loud as the T630 and it s also expensive. This socket is also used in the HP Z6 G4 which takes a W-3265 or Xeon Gold 6258R CPU for the high end options. The HP Z6 G4 systems on ebay are all above $1500 and the Xeon Gold 6258R is also over $1000 so while the Xeon Gold 6258R in a Z6 G4 will give 50% better performance on multithreaded operations than the systems I currently have it s costing almost 3* as much. It has 6 DIMM sockets which is a nice improvement over the 4 in the z640. The Z6 G4 takes a maximum of 768G of RAM with the optional extra CPU board (which is very expensive both new and on ebay) compared to my z840 which has 512G and half it s DIMM slots empty. The HP Z8 G4 has the same socket and takes up to 3TB of RAM if used with CPUs that support it (most CPUs only support 768G and you need a M variant to support more). The higher performance CPUs supported in the Z6 G4 and Z8 G4 don t have enough entries in the Passmark database to be accurate, but going from 22 cores in the E5-2699A v4 to 28 in the Xeon Platinum 8180 when using the same RAM technology doesn t seem like a huge benefit. The Z6 and Z8 G4 systems run DDR4 RAM at up to 2666 speed while the z640 and z840 only to 2400, a 10% increase in RAM speed is nice but not a huge difference. I don t think that any socket LGA3647 systems will ever be ones I want to buy. They don t offer much over LGA2011-v3 but are in newer and fancier systems that will go for significantly higher prices. DDR5 I think that DDR5 systems will be my next step up in tower server and workstation performance after the socket LGA2011-v3 systems. I don t think anything less will offer me enough of a benefit to justify a change. I also don t think that they will be in the price range I am willing to pay until well after DDR6 is released, some people are hoping for DDR6 to be released late this year but next year seems more likely. So maybe in 2027 there will be some nice DDR5 systems going cheap. CPU Benchmark Results Here are the benchmark results of CPUs I mentioned in this post according to passmark.com [9]. I didn t reference results of CPUs that only had 1 or 2 results posted as they aren t likely to be accurate.
CPU Single Thread Multi Thread TDP
E5-2683 v4 1,713 17,591 120W
Xeon Gold 5120 1,755 18,251 105W
i9-9900K 2,919 18,152 95W
E5-2697A v4 2,106 21,610 145W
E5-2699A v4 2,055 26,939 145W
W-3265 2,572 30,105 205W
W-2295 2,642 30,924 165W
i9-10980XE 2,662 32,397 165W
Xeon Gold 6258R 2,080 40,252 205W

1 August 2025

puer-robustus: My Google Summer of Code '25 at Debian

I ve participated in this year s Google Summer of Code (GSoC) program and have been working on the small (90h) autopkgtests for the rsync package project at Debian.

Writing my proposal Before you can start writing a proposal, you need to select an organization you want to work with. Since many organizations participate in GSoC, I ve used the following criteria to narrow things down for me:
  • Programming language familiarity: For me only Python (preferably) as well as shell and Go projects would have made sense. While learning another programming language is cool, I wouldn t be as effective and helpful to the project as someone who is proficient in the language already.
  • Standing of the organization: Some of the organizations participating in GSoC are well-known for the outstanding quality of the software they produce. Debian is one of them, but so is e.g. the Django Foundation or PostgreSQL. And my thinking was that the higher the quality of the organization, the more there is to learn for me as a GSoC student.
  • Mentor interactions: Apart from the advantage you get from mentor feedback when writing your proposal (more on that further below), it is also helpful to gauge how responsive/helpful your potential mentor is during the application phase. This is important since you will be working together for a period of at least 2 months; if the mentor-student communication doesn t work, the GSoC project is going to be difficult.
  • Free and Open-Source Software (FOSS) communication platforms: I generally believe that FOSS projects should be built on FOSS infrastructure. I personally won t run proprietary software when I want to contribute to FOSS in my spare time.
  • Be a user of the project: As Eric S. Raymond has pointed out in his seminal The Cathedral and the Bazaar 25 years ago
    Every good work of software starts by scratching a developer s personal itch.
Once I had some organizations in mind whose projects I d be interested in working on, I started writing proposals for them. Turns out, I started writing my proposals way too late: In the end I only managed to hand in a single one which is risky. Competition for the GSoC projects is fierce and the more quality (!) proposals you send out, the better your chances are at getting one. However, don t write proposals for the sake of it: Reviewers get way too many AI slop proposals already and you will not do yourself a favor with a low-quality proposal. Take the time to read the instructions/ideas/problem descriptions the project mentors have provided and follow their guidelines. Don t hesitate to reach out to project mentors: In my case, I ve asked Samuel Henrique a few clarification questions whereby the following (email) discussion has helped me greatly in improving my proposal. Once I ve finalized my proposal draft, I ve sent it to Samuel for a review, which again led to some improvements to the final proposal which I ve uploaded to the GSoC program webpage.

Community bonding period Once you get the information that you ve been accepted into the GSoC program (don t take it personally if you don t make it; this was my second attempt after not making the cut in 2024), get in touch with your prospective mentor ASAP. Agree upon a communication channel and some response times. Put yourself in the loop for project news and discussions whatever that means in the context of your organization: In Debian s case this boiled down to subscribing to a bunch of mailing lists and IRC channels. Also make sure to setup a functioning development environment if you haven t done so for writing the proposal already.

Payoneer setup The by far most annoying part of GSoC for me. But since you don t have a choice if you want to get the stipend, you will need to signup for an account at Payoneer. In this iteration of GSoC all participants got a personalized link to open a Payoneer account. When I tried to open an account by following this link, I got an email after the registration and email verification that my account is being blocked because Payoneer deems the email adress I gave a temporary one. Well, the email in question is most certainly anything but temporary, so I tried to get in touch with the Payoneer support - and ended up in an LLM-infused kafkaesque support hell. Emails are answered by an LLM which for me meant utterly off-topic replies and no help whatsoever. The Payoneer website offers a real-time chat, but it is yet another instance of a bullshit-spewing LLM bot. When I at last tried to call them (the support lines are not listed on the Payoneer website but were provided by the GSoC program), I kid you not, I was being told that their platform is currently suffering from technical problems and was hung up on. Only thanks to the swift and helpful support of the GSoC administrators (who get priority support from Payoneer) I was able to setup a Payoneer account in the end. Apart from showing no respect to customers, Payoneer is also ripping them off big time with fees (unless you get paid in USD). They charge you 2% for currency conversions to EUR on top of the FX spread they take. What worked for me to avoid all of those fees, was to open a USD account at Wise and have Payoneer transfer my GSoC stipend in USD to that account. Then I exchanged the USD to my local currency at Wise for significantly less than Payoneer would have charged me. Also make sure to close your Payoneer account after the end of GSoC to avoid their annual fee.

Project work With all this prelude out of the way, I can finally get to the actual work I ve been doing over the course of my GSoC project.

Background The upstream rsync project generally sees little development. Nonetheless, they released version 3.4.0 including some CVE fixes earlier this year. Unfortunately, their changes broke the -H flag. Now, Debian package maintainers need to apply those security fixes to the package versions in the Debian repositories; and those are typically a bit older. Which usually means that the patches cannot be applied as is but will need some amendments by the Debian maintainers. For these cases it is helpful to have autopkgtests defined, which check the package s functionality in an automated way upon every build. The question then is, why should the tests not be written upstream such that regressions are caught in the development rather than the distribution process? There s a lot to say on this question and it probably depends a lot on the package at hand, but for rsync the main benefits are twofold:
  1. The upstream project mocks the ssh connection over which rsync is most typically used. Mocking is better than nothing but not the real thing. In addition to being a more realisitic test scenario for the typical rsync use case, involving an ssh server in the test would automatically extend the overall resilience of Debian packages as now new versions of the openssh-server package in Debian benefit from the test cases in the rsync reverse dependency.
  2. The upstream rsync test framework is somewhat idiosyncratic and difficult to port to reimplementations of rsync. Given that the original rsync upstream sees little development, an extensive test suit further downstream can serve as a threshold for drop-in replacements for rsync.

Goal(s) At the start of the project, the Debian rsync package was just running (a part of) the upstream tests as autopkgtests. The relevant snippet from the build log for the rsync_3.4.1+ds1-3 package reads:
114s ------------------------------------------------------------
114s ----- overall results:
114s 36 passed
114s 7 skipped
Samuel and I agreed that it would be a good first milestone to make the skipped tests run. Afterwards, I should write some rsync test cases for local calls, i.e. without an ssh connection, effectively using rsync as a more powerful cp. And once that was done, I should extend the tests such that they run over an active ssh connection. With these milestones, I went to work.

Upstream tests Running the seven skipped upstream tests turned out to be fairly straightforward:
  • Two upstream tests concern access control lists and extended filesystem attributes. For these tests to run they rely on functionality provided by the acl and xattr Debian packages. Adding those to the Build-Depends list in the debian/control file of the rsync Debian package repo made them run.
  • Four upstream tests required root privileges to run. The autopkgtest tool knows the needs-root restriction for that reason. However, Samuel and I agreed that the tests should not exclusively run with root privileges. So, instead of just adding the restiction to the existing autopkgtest test, we created a new one which has the needs-root restriction and runs the upstream-tests-as-root script - which is nothing else than a symlink to the existing upstream-tests script.
The commits to implement these changes can be found in this merge request. The careful reader will have noticed that I only made 2 + 4 = 6 upstream test cases run out of 7: The leftover upstream test is checking the functionality of the --ctimes rsync option. In the context of Debian, the problem is that the Linux kernel doesn t have a syscall to set the creation time of a file. As long as that is the case, this test will always be skipped for the Debian package.

Local tests When it came to writing Debian specific test cases I started of a completely clean slate. Which is a blessing and a curse at the same time: You have full flexibility but also full responsibility. There were a few things to consider at this point in time:
  • Which language to write the tests in? The programming language I am most proficient in is Python. But testing a CLI tool in Python would have been weird: it would have meant that I d have to make repeated subprocess calls to run rsync and then read from the filesystem to get the file statistics I want to check. Samuel suggested I stick with shell scripts and make use of diffoscope - one of the main tools used and maintained by the Reproducible Builds project - to check whether the file contents and file metadata are as expected after rsync calls. Since I did not have good reasons to use bash, I ve decided to write the scripts to be POSIX compliant.
  • How to avoid boilerplate? If one makes use of a testing framework, which one? Writing the tests would involve quite a bit of boilerplate, mostly related to giving informative output on and during the test run, preparing the file structure we want to run rsync on, and cleaning the files up after the test has run. It would be very repetitive and in violation of DRY to have the code for this appear in every test. Good testing frameworks should provide convenience functions for these tasks. shunit2 comes with those functions, is packaged for Debian, and given that it is already being used in the curl project, I decided to go with it.
  • Do we use the same directory structure and files for every test or should every test have an individual setup? The tradeoff in this question being test isolation vs. idiosyncratic code. If every test has its own setup, it takes a) more work to write the test and b) more work to understand the differences between tests. However, one can be sure that changes to the setup in one test will have no side effects on other tests. In my opinion, this guarantee was worth the additional effort in writing/reading the tests.
Having made these decisions, I simply started writing tests and ran into issues very quickly.

rsync and subsecond mtime diffs When testing the rsync --times option, I observed a weird phenomenon: If the source and destination file have modification times which differ only in the nanoseconds, an rsync --times call will not synchronize the modification times. More details about this behavior and examples can be found in the upstream issue I raised. In the Debian tests we had to occasionally work around this by setting the timestamps explicitly with touch -d.
diffoscope regression In one test case, I was expecting a difference in the modification times but diffoscope would not report a diff. After a good amount of time spent on debugging the problem (my default, and usually correct, assumption is that something about my code is seriously broken if I run into issues like that), I was able to show that diffoscope only displayed this behavior in the version in the unstable suite, not on Debian stable (which I am running on my development machine). Since everything pointed to a regression in the diffoscope project and with diffoscope being written in Python, a language I am familiar with, I wanted to spend some time investigating (and hopefully fixing) the problem. Running git bisect on the diffoscope repo helped me in identifying the commit which introduced the regression: The commit contained an optimization via an early return for bit-by-bit identical files. Unfortunately, the early return also caused an explicitly requested metadata comparison (which could be different between the files) to be skipped. With a nicely diagnosed issue like that, I was able to go to a local hackerspace event, where people work on FOSS together for an evening every month. In a group, we were able to first, write a test which showcases the broken behavior in the latest diffoscope version, and second, make a fix to the code such that the same test passes going forward. All details can be found in this merge request.
shunit2 failures At some point I had a few autopkgtests setup and passing, but adding a new one would throw me totally inexplicable errors. After trying to isolate the problem as much as possible, it turns out that shunit2 doesn t play well together we the -e shell option. The project mentions this in the release notes for the 2.1.8 version1, but in my opinion a constraint this severe should be featured much more prominently, e.g. in the README.

Tests over an ssh connection The centrepiece of this project; everything else has in a way only been preparation for this. Obviously, the goal was to reuse the previously written local tests in some way. Not only because lazy me would have less work to do this way, but also because of a reduced long-term maintenance burden of one rather than two test sets. As it turns out, it is actually possible to accomplish that: The remote-tests script doesn t do much apart from starting an ssh server on localhost and running the local-tests script with the REMOTE environment variable set. The REMOTE environment variable changes the behavior of the local-tests script in such a way that it prepends "$REMOTE": to the destination of the rsync invocations. And given that we set REMOTE=rsync@localhost in the remote-tests script, local-tests copies the files to the exact same locations as before, just over ssh. The implementational details for this can be found in this merge request.

proposed-updates Most of my development work on the Debian rsync package took place during the Debian freeze as the release of Debian Trixie is just around the corner. This means that uploading by Debian Developers (DD) and Debian Maintainers (DM) to the unstable suite is discouraged as it makes migrating the packages to testing more difficult for the Debian release team. If DDs/DMs want to have the package version in unstable migrated to testing during the freeze they have to file an unblock request. Samuel has done this twice (1, 2) for my work for Trixie but has asked me to file the proposed-updates request for current stable (i.e. Debian Bookworm) myself after I ve backported my tests to bookworm.

Unfinished business To run the upstream tests which check access control list and extended file system attributes functionality, I ve added the acl and xattr packages to Build-Depends in debian/control. This, however, will only make the packages available at build time: If Debian users install the rsync package, the acl and xattr packages will not be installed alongside it. For that, the dependencies would have to be added to Depends or Suggests in debian/control. Depends is probably to strong of a relation since rsync clearly works well in practice without, but adding them to Suggests might be worthwhile. A decision on this would involve checking, what happens if rsync is called with the relevant options on a host machine which has those packages installed, but where the destination machine lacks them. Apart from the issue described above, the 15 tests I managed to write are are a drop in the water in light of the infinitude of rsync options and their combinations. Most glaringly, not all options of the --archive option are covered separately (which would help indicating what code path of rsync broke in a regression). To increase the likelihood of catching regressions with the autopkgtests, the test coverage should be extended in the future.

Conclusion Generally, I am happy with my contributions to Debian over the course of my small GSoC project: I ve created an extensible, easy to understand, and working autopkgtest setup for the Debian rsync package. There are two things which bother me, however:
  1. In hindsight, I probably shouldn t have gone with shunit2 as a testing framework. The fact that it behaves erratically with the -e flag is a serious drawback for a shell testing framework: You really don t want a shell command to fail silently and the test to continue running.
  2. As alluded to in the previous section, I m not particularly proud of the number of tests I managed to write.
On the other hand, finding and fixing the regression in diffoscope - while derailing me from the GSoC project itself - might have a redeeming quality.

DebConf25 By sheer luck I happened to work on a GSoC project at Debian over a time period during which the annual Debian conference would take place close enough to my place of residence. Samuel pointed the opportunity to attend DebConf out to me during the community bonding period and since I could make time for the event in my schedule, I signed up. DebConf was a great experience which - aside from gaining more knowledge about Debian development - allowed me to meet the actual people usually hidden behind email adresses and IRC nicks. I can wholeheartedly recommend attending a DebConf to every interested Debian user! For those who have missed this year s iteration of the conference, I can recommend the following recorded talks: While not featuring as a keynote speaker (understandably so as the newcomer to Debian community that I am), I could still contribute a bit to the conference program.

GSoC project presentation The Debian Outreach team has scheduled a session in which all GSoC and Outreachy students over the past year had the chance to present their work in a lightning talk. The session has been recorded and is available online, just like my slides and the source for them.

Debian install workshop Additionally, with so many Debian experts gathering in one place while KDE s End of 10 campaign is ongoing, I felt it natural to organize a Debian install workhop. In hindsight I can say that I underestimated how much work it would be, especially for me who does not speak a word of French. But although the turnout of people who wanted us to install Linux on their machines was disappointingly low, it was still worth it: Not only because the material in the repo can be helpful to others planning install workshops but also because it was nice to meet a) the person behind the Debian installer images and b) the local Brest/Finist re Linux user group as well as the motivated and helpful people at Infini.

Credits I want to thank the Open Source team at Google for organizing GSoC: The highly structured program with a one-to-one mentorship is a great avenue to start contributing to well established and at times intimidating FOSS projects. And as much as I disagree with Google s surveillance capitalist business model, I have to give it to them that the company at least takes its responsibility for FOSS (somewhat) seriously - unlike many other businesses which rely on FOSS and choose to freeride of it. Big thanks to the Debian community! I ve experienced nothing but friendliness in my interactions with the community. And lastly, the biggest thanks to my GSoC mentor Samuel Henrique. He has dealt patiently and competently with all my stupid newbie questions. His support enabled me to make - albeit small - contributions to Debian. It has been a pleasure to work with him during GSoC and I m looking forward to working together with him in the future.

  1. Obviously, I ve only read them after experiencing the problem.

29 July 2025

Christoph Berg: The Debian Conference 2025 in Brest

It's Sunday and I'm now sitting in the train from Brest to Paris where I will be changing to Germany, on the way back from the annual Debian conference. A full week of presentations, discussions, talks and socializing is laying behind me and my head is still spinning from the intensity.
Pollito and the gang of DebConf mascots wearing their conference badgesPollito and the gang of DebConf mascots wearing their conference badges (photo: Christoph Berg)
Sunday, July 13th It started last Sunday with traveling to the conference. I got on the Eurostar in Duisburg and we left on time, but even before reaching Cologne, the train was already one hour delayed for external reasons, collecting yet another hour between Aachen and Liege for its own technical problems. "The train driver is working on trying to fix the problem." My original schedule had well over two hours for changing train stations in Paris, but being that late, I missed the connection to Brest in Montparnasse. At least in the end, the total delay was only one hour when finally arriving at the destination. Due to the French julliet quatorze fireworks approaching, buses in Brest were rerouted, but I managed to catch the right bus to the conference venue, already meeting a few Debian people on the way. The conference was hosted at the IMT Atlantique Brest campus, giving the event a nice university touch. I arrived shortly after 10 in the evening and after settling down a bit, got on one of the "magic" buses for transportation to the camping site where half of the attendees where stationed. I shared a mobile home with three other Debianites, where I got a small room for myself. Monday, July 14th Next morning, we took the bus back to the venue with a small breakfast and the opening session where Enrico Zini invited me to come to his and Nicolas Dandrimont's session about Debian community governance and curation, which I gladly did. Many ideas about conflict moderation and community steering were floated around. I hope some of that can be put into effect to make flamewars on the mailing lists less heated and more directed. After that, I attended Olly Betts' "Stemming with Snowball" session, which is the stemmer used also in PostgreSQL. Text search is one of the areas in PostgreSQL that I never really looked closely at, including the integration into the postgresql-common package, so it was nice to get more information about that. In preparation for the conference, a few of us Ham radio operators in Debian had decided to bring some radio gear to DebConf this year in order to perhaps spark more interest for our hobby among the fellow geeks. In the afternoon after the talks, I found a quieter spot just outside of the main hall and set up a shortwave antenna by attaching a 10m mast to one of the park benches there. The 40m band was still pretty much closed, but I could work a few stations from England, just across the channel from Bretagne, answering questions from interested passing-by Debian people between the contacts. Over time, the band opened and more European stations got into the log.
F/DF7CB in Brest (photo: Evangelos Ribeiro Tzaras)
Tuesday, July 15th Tuesday started with Helmut Grohne's session about "Reviving (un)schroot". The schroot program has been Debian's standard way of managing build chroots for a long time, but it is more and more being regarded as obsolete with all kinds of newer containerization and virtualization technologies taking over. Since many bits of Debian infrastructure depend on schroot, and its user interface is still very useful, Helmut reimplemented it using Linux namespaces and the "unshare" systemcall. I had already worked with him at the Hamburg Minidebconf to replace the apt.postgresql.org buildd machinery with the new system, but we were not quite there yet (network isolation is nice, but we still sometimes need proper networking), so it was nice to see the effort is still progressing and I will give his new scripts a try when I'm back home. Next, Stefano Rivera and Colin Watson presented Debusine, a new package repository and workflow management system. It looks very promising for anyone running their own repository, so perhaps yet another bit of apt.postgresql.org infrastructure to replace in the future. After that, I went to the Debian LTS BoF session by Santiago Ruano Rinc n and Bastien Roucari s - Debian releases plus LTS is what we are covering with apt.postgresql.org. Then there were bits from the DPL (Debian Project Leader), and a session moderated by Stefano Rivera interesting to me as a member of the Debian Technical Committee on the future structure of the packages required for cross-building in Debian, a topic which had been brought to TC a while ago. I am happy that we could resolve the issue without having to issue a formal TC ruling as the involved parties (kernel, glibc, gcc and the cross-build people) found a promising way forward themselves. DebConf is really a good way to get such issues unstuck. Ten years ago at the 2015 Heidelberg DebConf, Enrico had given a seminal "Semi-serious stand-up comedy" talk, drawing parallels between the Debian Open Source community and the BDSM community - "People doing things consensually together". (Back then, the talk was announced as "probably unsuitable for people of all ages".) With his unique presentation style and witty insights, the session made a lasting impression on everyone attending. Now, ten years later (and he and many in the audience being ten years older), he gave an updated version of it. We are now looking forward to the sequel in 2035. The evening closed with the famous DebConf tradition of the Cheese & Wine party in a old fort next to the coast, just below the conference venue. Even when he's a fellow Debian Developer, Ham and also TC member, I had never met Paul Tagliamonte in person before, but we spent most of the evening together geeking out on all things Debian and Ham radio.
The northern coast of Ushant (photo: Christoph Berg)
Wednesday, July 16th Wednesday already marked the end of the first half of the week, the day of the day trips. I had chosen to go to Ouessant island (Ushant in English) which marks the Western end of French mainland and hosts one of the lighthouses yielding the way into the English channel. The ferry trip included surprisingly big waves which left some participants seasick, but everyone recovered fast. After around one and a half hours we arrived, picked up the bicycles, and spent the rest of the day roaming the island. The weather forecast was originally very cloudy and 18 C, but over noon this turned into sunny and warm, so many got an unplanned sunburn. I enjoyed the trip very much - it made up for not having time visiting the city during the week. After returning, we spent the rest of the evening playing DebConf's standard game, Mao (spoiler alert: don't follow the link if you ever intend to play).
Having a nice day (photo: Christoph Berg)
Thursday, July 17th The next day started with the traditional "Meet the Technical Committee" session. This year, we trimmed the usual slide deck down to remove the boring boilerplate parts, so after a very short introduction to the work of the committee by our chairman Matthew Vernon, we opened up the discussion with the audience, with seven (out of 8) TC members on stage. I think the format worked very well, with good input from attendees. Next up was "Don't fear the TPM" by Jonathan McDowell. A common misconception in the Free Software community is that the TPM is evil DRM hardware working against the user, but while it could be used in theory that way, the necessary TPM attestations seem to impossible to attain in practice, so that wouldn't happen anyway. Instead, it is a crypto coprocessor present in almost all modern computers that can be used to hold keys, for example to be used for SSH. It will also be interesting to research if we can make use of it for holding the Transparent Data Encryption keys for CYBERTEC's PostgreSQL Enterprise Edition. Aigars Mahinovs then directed everyone in place for the DebConf group picture, and Lucas Nussbaum started a discussion about archive-wide QA tasks in Debian, an area where I did a lot of work in the past and that still interests me. Antonio Terceiro and Paul Gevers followed up with techniques to track archive-wide rebuilding and testing of packages and in turn filing a lot of bugs to track the problems. The evening ended with the conference dinner, again in the fort close by the coast. DebConf is good for meeting new people, and I incidentally ran into another Chris, who happened to be one of the original maintainers of pgaccess, the pre-predecessor of today's pgadmin. I admit still missing this PostgreSQL frontend for its simplicity and ability to easily edit table data, but it disappeared around 2004. Friday, July 18th On Friday, I participated in discussion sessions around contributors.debian.org (PostgreSQL is planning to set up something similar) and the New Member process which I had helped to run and reform a decade or two ago. Agathe Porte (also a Ham radio operator, like so many others at the conference I had no idea of) then shared her work on rust-rewriting the slower parts of Lintian, the Debian package linter. Craig Small talked about "Free as in Bytes", the evolution of the Linux procps free command. Over the time and many kernel versions, the summary numbers printed became better and better, but there will probably never be a version that suits all use cases alike. Later over dinner, Craig (who is also a TC member) and I shared our experiences with these numbers and customers (not) understanding them. He pointed out that for PostgreSQL and looking at used memory in the presence of large shared memory buffers, USS (unique set size) and PSS (proportional set size) should be more realistic numbers than the standard RSS (resident set size) that the top utility is showing by default. Antonio Terceiro and Paul Gevers again joined to lead a session, now on ci.debian.net and autopkgtest, the test driver used for running tests on packages after then have been installed on a system. The PostgreSQL packages are heavily using this to make sure no regressions creep in even after builds have successfully completed and test re-runs are rescheduled periodically. The day ended with Bdale Garbee's electronics team BoF and Paul Tagliamonte and me setting up the radio station in the courtyard, again answering countless questions about ionospheric conditions and operating practice. Saturday, July 19th Saturday was the last conference day. In the first session, Nikos Tsipinakis and Federico Vaga from CERN announced that the LHC will be moving to Debian for the accelerator's frontend computers in their next "long shutdown" maintenance period in the next year. CentOS broke compatibility too often, and Debian trixie together with the extended LTS support will cover the time until the next long shutdown window in 2035, until when the computers should have all been replaced with newer processors covering higher x86_64 baseline versions. The audience was very delighted to hear that Debian is now also being used in this prestige project. Ben Hutchings then presented new Linux kernel features. Particularly interesting for me was the support for atomic writes spanning more than one filesystem block. When configured correctly, this would mean PostgreSQL didn't have to record full-page images in the WAL anymore, increasing throughput and performance. After that, the Debian ftp team discussed ways to improve review of new packages in the archive, and which of their processes could be relaxed with new US laws around Open Source and cryptography algorithms export. Emmanuel Arias led a session on Salsa CI, Debian's Gitlab instance and standard CI pipeline. (I think it's too slow, but the runners are not under their control.) Julian Klode then presented new features in APT, Debian's package manager. I like the new display format (and a tiny bit of that is also from me sending in wishlist bugs). In the last round of sessions this week, I then led the Ham radio BoF with an introduction into the hobby and how Debian can be used. Bdale mentioned that the sBitx family of SDR radios is natively running Debian, so stock packages can be used from the radio's touch display. We also briefly discussed his involvement in ARDC and the possibility to get grants from them for Ham radio projects. Finally, DebConf wrapped up with everyone gathering in the main auditorium and cheering the organizers for making the conference possible and passing Pollito, the DebConf mascot, to the next organizer team.
Pollito on stage (photo: Christoph Berg)
Sunday, July 20th Zoom back to the train: I made it through the Paris metro and I'm now on the Eurostar back to Germany. It has been an intense week with all the conference sessions and meeting all the people I had not seen so long. There are a lot of new ideas to follow up on both for my Debian and PostgreSQL work. Next year's DebConf will take place in Santa Fe, Argentina. I haven't yet decided if I will be going, but I can recommend the experience to everyone! The post The Debian Conference 2025 in Brest appeared first on CYBERTEC PostgreSQL Services & Support.

28 July 2025

Dimitri John Ledkov: Achieving actually full disk encryption of UEFI ESP at rest with TCG OPAL, FIPS, LUKS

Achieving full disk encryption using FIPS, TCG OPAL and LUKS to encrypt UEFI ESP on bare-metal and in VMs
Many security standards such as CIS and STIG require to protect information at rest. For example, NIST SP 800-53r5 SC-28 advocate to use cryptographic protection, offline storage and TPMs to enhance protection of information confidentiality and/or integrity.Traditionally to satisfy such controls on portable devices such as laptops one would utilize software based Full Disk Encryption - Mac OS X FileVault, Windows Bitlocker, Linux cryptsetup LUKS2. In cases when FIPS cryptography is required, additional burden would be placed onto these systems to operate their kernels in FIPS mode.Trusted Computing Group works on establishing many industry standards and specifications, which are widely adopted to improve safety and security of computing whilst keeping it easy to use. One of their most famous specifications them is TCG TPM 2.0 (Trusted Platform Module). TPMs are now widely available on most devices and help to protect secret keys and attest systems. For example, most software full disk encryption solutions can utilise TCG TPM to store full disk encryption keys providing passwordless, biometric or pin-base ways to unlock the drives as well as attesting that system have not been modified or compromised whilst offline.TCG Storage Security Subsystem Class: Opal Specification is a set of specifications for features of data storage devices. The authors and contributors to OPAL are leading and well trusted storage manufacturers such as Samsung, Western Digital, Seagate Technologies, Dell, Google, Lenovo, IBM, Kioxia, among others. One of the features that Opal Specification enables is self-encrypting drives which becomes very powerful when combined with pre-boot authentication. Out of the box, such drives always and transparently encrypt all disk data using hardware acceleration. To protect data one can enter UEFI firmware setup (BIOS) to set NVMe single user password (or user + administrator/recovery passwords) to encrypt the disk encryption key. If one's firmware didn't come with such features, one can also use SEDutil to inspect and configure all of this. Latest release of major Linux distributions have SEDutil already packaged.Once password is set, on startup, pre-boot authentication will request one to enter password - prior to booting any operating systems. It means that full disk is actually encrypted, including the UEFI ESP and all operating systems that are installed in case of dual or multi-boot installations. This also prevents tampering with ESP, UEFI bootloaders and kernels which with traditional software-based encryption often remain unencrypted and accessible. It also means one doesn't have to do special OS level repartitioning, or installation steps to ensure all data is encrypted at rest.What about FIPS compliance? Well, the good news is that majority of the OPAL compliant hard drives and/or security sub-chips do have FIPS 140-3 certification. Meaning they have been tested by independent laboratories to ensure they do in-fact encrypt data. On the CMVP website one can search for module name terms "OPAL" or "NVMe" or name of hardware vendor to locate FIPS certificates.Are such drives widely available? Yes. For example, a common Thinkpad X1 gen 11 has OPAL NVMe drives as standard, and they have FIPS certification too. Thus, it is likely in your hardware fleet these are already widely available. Use sedutil to check if MediaEncrypt and LockingSupported features are available.Well, this is great for laptops and physical servers, but you may ask - what about public or private cloud? Actually, more or less the same is already in-place in both. On CVMP website all major clouds have their disk encryption hardware certified, and all of them always encrypt all Virtual Machines with FIPS certified cryptography without an ability to opt-out. One is however in full control of how the encryption keys are managed: cloud-provider or self-managed (either with a cloud HSM or KMS or bring your own / external). See these relevant encryption options and key management docs for GCP, Azure, AWS. But the key takeaway without doing anything, at rest, VMs in public cloud are always encrypted and satisfy NIST SP 800-53 controls.What about private cloud? Most Linux based private clouds ultimately use qemu typically with qcow2 virtual disk images. Qemu supports user-space encryption of qcow2 disk, see this manpage. Such encryption encrypts the full virtual machine disk, including the bootloader and ESP. And it is handled entirely outside of the VM on the host - meaning the VM never has access to the disk encryption keys. Qemu implements this encryption entirely in userspace using gnutls, nettle, libgcrypt depending on how it was compiled. This also means one can satisfy FIPS requirements entirely in userspace without a Linux kernel in FIPS mode. Higher level APIs built on top of qemu also support qcow2 disk encryption, as in projects such as libvirt and OpenStack Cinder.If you carefully read the docs, you may notice that agent support is explicitly sometimes called out as not supported or not mentioned. Quite often agents running inside the OS may not have enough observability to them to assess if there is external encryption. It does mean that monitoring above encryption options require different approaches - for example monitor your cloud configuration using tools such as Wiz and Orca, rather than using agents inside individual VMs. For laptop / endpoint security agents, I do wish they would start gaining capability to report OPAL SED availability and status if it is active or not.What about using software encryption none-the-less on top of the above solutions? It is commonly referred to double or multiple encryption. There will be an additional performance impact, but it can be worthwhile. It really depends on what you define as data at rest for yourself and which controls you need. If one has a dual-boot laptop, and wants to keep one OS encrypted whilst booted into the other, it can perfectly reasonable to encrypted the two using separate software encryption keys. In addition to the OPAL encryption of the ESP. For more targeted per-file / per-folder encryption, one can look into using gocryptfs which is the best successor to the once popular, but now deprecated eCryptfs (amazing tool, but has fallen behind in development and can lead to data loss).All of the above mostly talks about cryptographic encryption, which only provides confidentially but not data integrity. To protect integrity, one needs to choose how to maintain that. dm-verity is a good choice for read-only and rigid installations. For read-write workloads, it may be easier to deploy ZFS or Btrfs instead. If one is using filesystems without a built-in integrity support such as XFS or Ext4, one can retrofit integrity layer to them by using dm-integrity (either standalone, or via dm-luks/cryptsetup --integrity option).

If one has a lot of estate and a lot of encryption keys to keep track off a key management solution is likely needed. The most popular solution is likely the one from Thales Group marketed under ChiperTrust Data Security Platform (previously Vormetric), but there are many others including OEM / Vendor / Hardware / Cloud specific or agnostic solutions.

I hope this crash course guide piques your interest to learn and discover modern confidentially and integrity solutions, and to re-affirm or change your existing controls w.r.t. to data protection at rest.

Full disk encryption, including UEFI ESP /boot/efi is now widely achievable by default on both baremetal machines and in VMs including with FIPS certification. To discuss more let's connect on Linkedin.

20 July 2025

Michael Prokop: What to expect from Debian/trixie #newintrixie

Trixie Banner, Copyright 2024 Elise Couper Update on 2025-07-28: added note about Debian 13/trixie support for OpenVox (thanks, Ben Ford!) Debian v13 with codename trixie is scheduled to be published as new stable release on 9th of August 2025. I was the driving force at several of my customers to be well prepared for the upcoming stable release (my efforts for trixie started in August 2024). On the one hand, to make sure packages we care about are available and actually make it into the release. On the other hand, to ensure there are no severe issues that make it into the release and to get proper and working upgrades. So far everything is looking pretty well and working fine, the efforts seemed to have payed off. :) As usual with major upgrades, there are some things to be aware of, and hereby I m starting my public notes on trixie that might be worth for other folks. My focus is primarily on server systems and looking at things from a sysadmin perspective. Further readings As usual start at the official Debian release notes, make sure to especially go through What s new in Debian 13 + issues to be aware of for trixie (strongly recommended read!). Package versions As a starting point, let s look at some selected packages and their versions in bookworm vs. trixie as of 2025-07-20 (mainly having amd64 in mind):
Package bookworm/v12 trixie/v13
ansible 2.14.3 2.19.0
apache 2.4.62 2.4.64
apt 2.6.1 3.0.3
bash 5.2.15 5.2.37
ceph 16.2.11 18.2.7
docker 20.10.24 26.1.5
dovecot 2.3.19 2.4.1
dpkg 1.21.22 1.22.21
emacs 28.2 30.1
gcc 12.2.0 14.2.0
git 2.39.5 2.47.2
golang 1.19 1.24
libc 2.36 2.41
linux kernel 6.1 6.12
llvm 14.0 19.0
lxc 5.0.2 6.0.4
mariadb 10.11 11.8
nginx 1.22.1 1.26.3
nodejs 18.13 20.19
openjdk 17.0 21.0
openssh 9.2p1 10.0p1
openssl 3.0 3.5
perl 5.36.0 5.40.1
php 8.2+93 8.4+96
podman 4.3.1 5.4.2
postfix 3.7.11 3.10.3
postgres 15 17
puppet 7.23.0 8.10.0
python3 3.11.2 3.13.5
qemu/kvm 7.2 10.0
rsync 3.2.7 3.4.1
ruby 3.1 3.3
rust 1.63.0 1.85.0
samba 4.17.12 4.22.3
systemd 252.36 257.7-1
unattended-upgrades 2.9.1 2.12
util-linux 2.38.1 2.41
vagrant 2.3.4 2.3.7
vim 9.0.1378 9.1.1230
zsh 5.9 5.9
Misc unsorted apt The new apt version 3.0 brings several new features, including: systemd systemd got upgraded from v252.36-1~deb12u1 to 257.7-1 and there are lots of changes. Be aware that systemd v257 has a new net.naming_scheme, v257 being PCI slot number is now read from firmware_node/sun sysfs file. The naming scheme based on devicetree aliases was extended to support aliases for individual interfaces of controllers with multiple ports. This might affect you, see e.g. #1092176 and #1107187, the Debian Wiki provides further useful information. There are new systemd tools available: The tools provided by systemd gained several new options: Debian s systemd ships new binary packages: Linux Kernel The trixie release ships a Linux kernel based on latest longterm version 6.12. As usual there are lots of changes in the kernel area, including better hardware support, and this might warrant a separate blog entry. To highlight some changes with Debian trixie: See Kernelnewbies.org for further changes between kernel versions. Configuration management For puppet users, Debian provides the puppet-agent (v8.10.0), puppetserver (v8.7.0) and puppetdb (v8.4.1) packages. Puppet s upstream does not provide packages for trixie, yet. Given how long it took them for Debian bookworm, and with their recent Plans for Open Source Puppet in 2025, it s unclear when (and whether at all) we might get something. As a result of upstream behavior, also the OpenVox project evolved, and they already provide Debian 13/trixie support (https://apt.voxpupuli.org/openvox8-release-debian13.deb). FYI: the AIO puppet-agent package for bookworm (v7.34.0-1bookworm) so far works fine for me on Debian/trixie. Be aware that due to the apt-key removal you need a recent version of the puppetlabs-apt for usage with trixie. The puppetlabs-ntp module isn t yet ready for trixie (regarding ntp/ntpsec), if you should depend on that. ansible is available and made it with version 2.19 into trixie. Prometheus stack Prometheus server was updated from v2.42.0 to v2.53, and all the exporters that got shipped with bookworm are still around (in more recent versions of course). Trixie gained some new exporters: Virtualization docker (v26.1.5), ganeti (v3.1.0), libvirt (v11.3.0, be aware of significant changes to libvirt packaging), lxc (v6.0.4), podman (v5.4.2), openstack (see openstack-team on Salsa), qemu/kvm (v10.0.2), xen (v4.20.0) are all still around. Proxmox already announced their PVE 9.0 BETA, being based on trixie and providing 6.14.8-1 kernel, QEMU 10.0.2, LXC 6.0.4, OpenZFS 2.3.3. Vagrant is available in version 2.3.7, but Vagrant upstream does not provide packages for trixie yet. Given that HashiCorp adopted the BSL, the future of vagrant in Debian is unclear. If you re relying on VirtualBox, be aware that upstream doesn t provide packages for trixie, yet. VirtualBox is available from Debian/unstable (version 7.1.12-dfsg-1 as of 2025-07-20), but not shipped with stable release since quite some time (due to lack of cooperation from upstream on security support for older releases, see #794466). Be aware that starting with Linux kernel 6.12, KVM initializes virtualization on module loading by default. This prevents VirtualBox VMs from starting. In order to avoid this, either add kvm.enable_virt_at_load=0 parameter into kernel command line or unload the corresponding kvm_intel / kvm_amd module. If you want to use Vagrant with VirtualBox on trixie, be aware that Debian s vagrant package as present in trixie doesn t support the VirtualBox package version 7.1 as present in Debian/unstable (manually patching vagrant s meta.rb and rebuilding the package without Breaks: virtualbox (>= 7.1) is known to be working). util-linux The are plenty of new options available in the tools provided by util-linux: Now no longer present in util-linux as of trixie: The following binaries got moved from util-linux to the util-linux-extra package: And the util-linux-extra package also provides new tools: OpenSSH OpenSSH was updated from v9.2p1 to 10.0p1-5, so if you re interested in all the changes, check out the release notes between those versions (9.3, 9.4, 9.5, 9.6, 9.7, 9.8, 9.9 + 10.0). Let s highlight some notable behavior changes in Debian: There are some notable new features: Thanks to everyone involved in the release, looking forward to trixie + and happy upgrading!
Let s continue with working towards Debian/forky. :)

12 July 2025

Bits from Debian: Debconf25 welcomes its sponsors

DebConf25 logo DebConf25, the 26th edition of the Debian conference is taking place in Brest Campus of IMT Atlantique Bretagne-Pays de la Loire, France. We appreciate the organizers for their hard work, and hope this event will be highly beneficial for those who attend in person as well as online. This event would not be possible without the help from our generous sponsors. We would like to warmly welcome the sponsors of DebConf 25, and introduce them to you. We have five Platinum sponsors. Our Gold sponsors are: Our Silver sponsors are: Bronze sponsors: And finally, our Supporter level sponsors: A special thanks to the IMT Atlantique Bretagne-Pays de la Loire, our Venue Partner and our Network Partner ResEl! Thanks to all our sponsors for their support! Their contributions enable a diverse global community of Debian developers and maintainers to collaborate, support one another, and share knowledge at DebConf25.

Christian Kastner: Easy dynamic dispatch using GLIBC Hardware Capabilities

TL;DR With GLIBC 2.33+, you can build a shared library multiple times targeting various optimization levels, and the dynamic linker/loader will pick the highest version supported by the current CPU. For example, with the layout below, on a Ryzen 9 5900X, x86-64-v3/libfoo0.so would be loaded:
/usr/lib/glibc-hwcaps/x86-64-v4/libfoo0.so
/usr/lib/glibc-hwcaps/x86-64-v3/libfoo0.so
/usr/lib/glibc-hwcaps/x86-64-v2/libfoo0.so
/usr/lib/libfoo0.so
Longer Version GLIBC Hardware Capabilities or "hwcaps" are an easy, almost trivial way to add a simple form of dynamic dispatch to any amd64 or POWER build, provided that either the build target or the compiler's optimizations can make use of certain CPU extensions. Mo Zhou pointed me towards this when I was faced with the challenge of creating a performant Debian package for ggml, the tensor library behind llama.cpp and whisper.cpp.
The Challenge A performant yet universally loadable library needs to make use of some form of dynamic dispatch to leverage the most effective SIMD extensions available on any given CPU it may run on. Last January, when I first started with the packaging of ggml for Debian, ggml did have support for this through its GGML_CPU_ALL_VARIANTS=ON option, but this was limited to amd64. This meant that on all the other architectures that Debian supports, I would need to target some ancient baseline, thus effectively crippling the package there.
Dynamic Dispatch using hwcaps hwcaps were introduced in GLIBC 2.33 and replace the (now) Legacy Hardware Capabilities, which were removed in 2.37. The way hwcaps work is delightfully simple: the dynamic linker/loader will look for a shared library not just in the standard library paths, but also in subdirectories thereof of the form hwcaps/<level>, starting with the highest <level> that the current CPU supports. The levels are predefined. I'm using the amd64 levels below. For ggml, this meant that I simply could build the library in multiple passes, each time targeting a different <level>, and install the result in the corresponding subdirectory, which resulted in the following layout (reduced to libggml.so for brevity):
/usr/lib/x86_64-linux-gnu/ggml/glibc-hwcaps/x86-64-v4/libggml.so
/usr/lib/x86_64-linux-gnu/ggml/glibc-hwcaps/x86-64-v3/libggml.so
/usr/lib/x86_64-linux-gnu/ggml/glibc-hwcaps/x86-64-v2/libggml.so
/usr/lib/x86_64-linux-gnu/ggml/libggml.so
In practice, this means that on a CPU supporting AVX512, the linker/loader would load x86-64-v4/libggml.so if it existed, and otherwise continue to look for the other levels, all the way down to the lowest one. On a CPU which supported only SSE4.2, the lookup process would be the same, ending with picking x86-64-v2/libggml.so. With QEMU, all of this was quickly verified. Note that the lowest-level library, targeting x86-64-v1, is not installed to a subdirectory, but to the path where the library would normally have been installed. This has the nice property that on systems not using GLIBC, and thus not having hwcaps available, package installation will still result in a loadable library, albeit the version with the worst performance. And a careful observer might have noticed that in the example above, the library is installed to a private ggml/ directory, so this mechanism also works when using RUNPATH or LD_LIBRARY_PATH. As mentioned above, Debian's ggml package will soon switch to GGML_CPU_ALL_VARIANTS=ON, but this was still quite the useful feature to discover.

11 July 2025

Jamie McClelland: Avoiding Apache Max Request Workers Errors

Wow, I hate this error:
AH00484: server reached MaxRequestWorkers setting, consider raising the MaxRequestWorkers setting
For starters, it means I have to relearn how MaxRequestWorkers functions in Apache:
For threaded and hybrid servers (e.g. event or worker), MaxRequestWorkers restricts the total number of threads that will be available to serve clients. For hybrid MPMs, the default value is 16 (ServerLimit) multiplied by the value of 25 (ThreadsPerChild). Therefore, to increase MaxRequestWorkers to a value that requires more than 16 processes, you must also raise ServerLimit.
Ok remind me what ServerLimit refers to?
For the prefork MPM, this directive sets the maximum configured value for MaxRequestWorkers for the lifetime of the Apache httpd process. For the worker and event MPMs, this directive in combination with ThreadLimit sets the maximum configured value for MaxRequestWorkers for the lifetime of the Apache httpd process. For the event MPM, this directive also defines how many old server processes may keep running and finish processing open connections. Any attempts to change this directive during a restart will be ignored, but MaxRequestWorkers can be modified during a restart. Special care must be taken when using this directive. If ServerLimit is set to a value much higher than necessary, extra, unused shared memory will be allocated. If both ServerLimit and MaxRequestWorkers are set to values higher than the system can handle, Apache httpd may not start or the system may become unstable. With the prefork MPM, use this directive only if you need to set MaxRequestWorkers higher than 256 (default). Do not set the value of this directive any higher than what you might want to set MaxRequestWorkers to. With worker, use this directive only if your MaxRequestWorkers and ThreadsPerChild settings require more than 16 server processes (default). Do not set the value of this directive any higher than the number of server processes required by what you may want for MaxRequestWorkers and ThreadsPerChild. With event, increase this directive if the process number defined by your MaxRequestWorkers and ThreadsPerChild settings, plus the number of gracefully shutting down processes, is more than 16 server processes (default).
Got it? In other words, you can consider raising the MaxRequestWorkers setting all you want, but you can t just change that setting, you have to read about several other compliated settings, do some math, and spend a lot of time wondering if you are going to remember what you just did and how to undo it if you blow up your server. On the plus side, typically, nobody should increase this limit - because if the server runs out of connections, it usually means something else is wrong. In our case, on a shared web server running Apache2 and PHP-FPM, it s usually because a single web site has gone out of control. But wait! How can that happen, we are using PHP-FPM s max_children setting to prevent a single PHP web site from taking down the server? After years of struggling with this problem I have finally made some headway. Our PHP pool configuration typically looks like this:
user = site342999writer
group = site342999writer
listen = /run/php/8.1-site342999.sock
listen.owner = www-data
listen.group = www-data
pm = ondemand
pm.max_children = 12
pm.max_requests = 500
php_admin_value[memory_limit] = 256M
And we invoke PHP-FPM via this apache snippet:
<FilesMatch \.php$>
        SetHandler "proxy:unix:/var/run/php/8.1-site342999.sock fcgi://localhost"
</FilesMatch>
With these settings in place, what happens when we use up all 12 max_children? According to the docs:
By default, mod_proxy will allow and retain the maximum number of connections that could be used simultaneously by that web server child process. Use the max parameter to reduce the number from the default. The pool of connections is maintained per web server child process, and max and other settings are not coordinated among all child processes, except when only one child process is allowed by configuration or MPM design.
The max parameter seems to default to the ThreadsPerChild, so it seems that the default here is to allow any web site to consume ThreadsPerChild (25) x ServerLimit (16), which is also the max number of over all connections. Not great. To make matter worse, there is another setting available which is mysteriously called acquire:
If set, this will be the maximum time to wait for a free connection in the connection pool, in milliseconds. If there are no free connections in the pool, the Apache httpd will return SERVER_BUSY status to the client.
By default this is not set which seems to suggest Apache will just hang on to connections forever until a free PHP process becomes available (or some other time out happens). So, let s try something different:
 <Proxy "fcgi://localhost">
    ProxySet acquire=1 max=12
  </proxy>
This snippet is the way you can configure the proxy configuration we setup in the SetHandler statement above. It s documented on the Apache mod_proxy page. Now we limit the maximum pool size per process to half of what is available for the entire server and we tell Apache to immediately throw a 503 error if we have exceeded our maximum number of connecitons. Now, if a site is overwhelmed with traffic, instead of maxing out the available Apache connections while leaving user with constantly spinning browsers, the users will get 503 ed and the server will be able to server other sites.

23 June 2025

Russell Coker: PFAs

For some time I ve been noticing news reports about PFAs [1]. I hadn t thought much about that issue, I grew up when leaded petrol was standard, when almost all thermometers had mercury, when all small batteries had mercury, and I had generally considered that I had already had so many nasty chemicals in my body that as long as I don t eat bottom feeding seafood often I didn t have much to worry about. I already had a higher risk of a large number of medical issues than I d like due to decisions made before I was born and there s not much to do about it given that there are regulations restricting the emissions of lead, mercury etc. I just watched a Veritasium video about Teflon and the PFA poisoning related to it s production [2]. This made me realise that it s more of a problem than I realised and it s a problem that s getting worse. PFA levels in the parts-per-trillion range in the environment can cause parts-per-billion in the body which increases the risks of several cancers and causes other health problems. Fortunately there is some work being done on water filtering, you can get filters for a home level now and they are working on filters that can work at a sufficient scale for a city water plant. There is a map showing PFAs in the environment in Australia which shows some sites with concerning levels that are near residential areas [3]. One of the major causes for that in Australia is fire retardant foam Australia has never had much if any Teflon manufacturing AFAIK. Also they noted that donating blood regularly can decrease levels of PFAs in the bloodstream. So presumably people who have medical conditions that require receiving donated blood regularly will have really high levels.

15 June 2025

Sahil Dhiman: A Look at .UA ccTLD Authoritative Name Servers

I find the case of the .UA country code top level domain (ccTLD) interesting simply because of the different name server secondaries they have now. Post Russian invasion, the cyber warfare peaked, and critical infrastructure like getting one side ccTLD down would be big news in anycase. Most (g/cc)TLDs are served by two (and less likely) by three or more providers. Even in those cases, not all authoritative name servers are anycasted. Take, example of .NL ccTLD name servers:
$ dig ns nl +short
ns1.dns.nl.
ns3.dns.nl.
ns4.dns.nl.
ns1.dns.nl is SIDN which also manages their registry. ns3.dns.nl is ReCodeZero/ipcom, another anycast secondary. ns4.dns.nl is CIRA, anycast secondary. That s 3 diverse, anycast networks to serve the .NL ccTLD. .DE has a bit more at name servers at 6 but only 3 seems anycasted. Now let s take a look at .UA. Hostmaster LLC is the registry operator of the .UA ccTLD since 2001.
$ dig soa ua +short
in1.ns.ua. domain-master.cctld.ua. 2025061434 1818 909 3024000 2020
Shows in1.ns.ua as primary nameserver (which can be intentionally deceptive too). I used bgp.tools for checking anycast and dns.coffee for timeline of when secondary nameserver was added. dns.coffee only has data going back till 2011 though. Let s deep dive at who s hosting each of the name servers:

in1.ns.ua by Intuix LLC
  • 74.123.224.40
  • 2604:ee00:0:101:0:0:0:40
  • unicast
  • Serving .UA since 13/12/2018.
  • Company by Dmitry Kohmanyuk and Igor Sviridov who re administrative and technical contacts for .UA zone as well as the IANA DB.

ho1.ns.ua by Hostmaster LLC
  • 195.47.253.1
  • 2001:67c:258:0:0:0:0:1
  • bgp.tools doesn t mark the prefix as anycast but basis test from various location, this is indeed anycasted (visible in atleast DE, US, UA etc.). Total POPs unknown.
  • Serving .UA atleast since 2011.
  • The registry themselves.

bg.ns.ua by ClouDNS
  • 185.136.96.185 and 185.136.97.185
  • 2a06:fb00:1:0:0:0:4:185 and 2a06:fb00:1:0:0:0:2:185
  • anycast
  • Serving .UA since 01/03/2022.
  • atleast 62 PoPs

cz.ns.ua by NIC.cz

nn.ns.ua by Netnod
  • 194.58.197.4
  • 2a01:3f1:c001:0:0:0:0:53
  • anycast
  • atleast 80 PoPs.
  • Serving .UA since 01/12/2022.
  • Netnod has the distinction of being one of the 13 root server operator (i.root-servers.net) and .SE operator.

pch.ns.ua by PCH
  • 204.61.216.12
  • 2001:500:14:6012:ad:0:0:1
  • anycast
  • atleast 328 POPs.
  • Serving .UA atleast since 2011.
  • With more than 36 years of production anycast DNS experience, two of the root name server operators and more than 172 top-level domain registries using our infrastructure, and more than 120 million resource records in service from https://www.pch.net/services/anycast.

rcz.ns.ua by RcodeZero
  • 193.46.128.10
  • 2a02:850:ffe0:0:0:0:0:10
  • anycast
  • Atleast 56 PoPs via 2 different cloud providers.
  • Serving .UA since 04/02/2022.
  • sister company of nic.at (.AT operator).

Some points to note
  • That s 1 unicast and 6 anycast name servers with hundreds of POPs from 7 different organizations.
  • Having X number of Point of Presence (POP) doesn t always mean each location is serving the .UA nameserver prefix.
  • Number of POPs keeps going up or down based on operational requirements and optimizations.
  • Highest concentration of DNS queries for a ccTLD would essentially originate in the country (or larger region) itself. If one of the secondaries doesn t have POP inside UA, the query might very well be served from outside the country, which can affect resolution and may even stop during outages and fiber cuts (which have become common there it seems). - Global POPs do help in faster resolutions for others/outside users though and ofcourse availability.
  • Having this much diversity does lessen the chance of the ccTLD going down. Theoretically, the adversary has to bring down 7 different networks/setups before resolution starts failing (post TTLs expiry).

11 June 2025

Freexian Collaborators: Monthly report about Debian Long Term Support, May 2025 (by Roberto C. S nchez)

Like each month, have a look at the work funded by Freexian s Debian LTS offering.

Debian LTS contributors In May, 22 contributors have been paid to work on Debian LTS, their reports are available:
  • Abhijith PA did 8.0h (out of 0.0h assigned and 8.0h from previous period).
  • Adrian Bunk did 26.0h (out of 26.0h assigned).
  • Andreas Henriksson did 1.0h (out of 15.0h assigned and 3.0h from previous period), thus carrying over 17.0h to the next month.
  • Andrej Shadura did 3.0h (out of 10.0h assigned), thus carrying over 7.0h to the next month.
  • Bastien Roucari s did 20.0h (out of 20.0h assigned).
  • Ben Hutchings did 8.0h (out of 20.0h assigned and 4.0h from previous period), thus carrying over 16.0h to the next month.
  • Carlos Henrique Lima Melara did 12.0h (out of 11.0h assigned and 1.0h from previous period).
  • Chris Lamb did 15.5h (out of 0.0h assigned and 15.5h from previous period).
  • Daniel Leidert did 25.0h (out of 26.0h assigned), thus carrying over 1.0h to the next month.
  • Emilio Pozuelo Monfort did 21.0h (out of 16.75h assigned and 11.0h from previous period), thus carrying over 6.75h to the next month.
  • Guilhem Moulin did 11.5h (out of 8.5h assigned and 6.5h from previous period), thus carrying over 3.5h to the next month.
  • Jochen Sprickerhof did 3.5h (out of 8.75h assigned and 17.5h from previous period), thus carrying over 22.75h to the next month.
  • Lee Garrett did 26.0h (out of 12.75h assigned and 13.25h from previous period).
  • Lucas Kanashiro did 20.0h (out of 18.0h assigned and 2.0h from previous period).
  • Markus Koschany did 20.0h (out of 26.25h assigned), thus carrying over 6.25h to the next month.
  • Roberto C. S nchez did 20.75h (out of 24.0h assigned), thus carrying over 3.25h to the next month.
  • Santiago Ruano Rinc n did 15.0h (out of 12.5h assigned and 2.5h from previous period).
  • Sean Whitton did 6.25h (out of 6.0h assigned and 2.0h from previous period), thus carrying over 1.75h to the next month.
  • Sylvain Beucler did 26.25h (out of 26.25h assigned).
  • Thorsten Alteholz did 15.0h (out of 15.0h assigned).
  • Tobias Frost did 12.0h (out of 12.0h assigned).
  • Utkarsh Gupta did 1.0h (out of 15.0h assigned), thus carrying over 14.0h to the next month.

Evolution of the situation In May, we released 54 DLAs. The LTS Team was particularly active in May, publishing a higher than normal number of advisories, as well as helping with a wide range of updates to packages in stable and unstable, plus some other interesting work. We are also pleased to welcome several updates from contributors outside the regular team.
  • Notable security updates:
    • containerd, prepared by Andreas Henriksson, fixes a vulnerability that could cause containers launched as non-root users to be run as root
    • libapache2-mod-auth-openidc, prepared by Moritz Schlarb, fixes a vulnerability which could allow an attacker to crash an Apache web server with libapache2-mod-auth-openidc installed
    • request-tracker4, prepared by Andrew Ruthven, fixes multiple vulnerabilities which could result in information disclosure, cross-site scripting and use of weak encryption for S/MIME emails
    • postgresql-13, prepared by Bastien Roucari s, fixes an application crash vulnerability that could affect the server or applications using libpq
    • dropbear, prepared by Guilhem Moulin, fixes a vulnerability which could potentially result in execution of arbitrary shell commands
    • openjdk-17, openjdk-11, prepared by Thorsten Glaser, fixes several vulnerabilities, which include denial of service, information disclosure or bypass of sandbox restrictions
    • glibc, prepared by Sean Whitton, fixes a privilege escalation vulnerability
  • Notable non-security updates:
    • wireless-regdb, prepared by Ben Hutchings, updates information reflecting changes to radio regulations in many countries
This month s contributions from outside the regular team include the libapache2-mod-auth-openidc update mentioned above, prepared by Moritz Schlarb (the maintainer of the package); the update of request-tracker4, prepared by Andrew Ruthven (the maintainer of the package); and the updates of openjdk-17 and openjdk-11, also noted above, prepared by Thorsten Glaser. Additionally, LTS Team members contributed stable updates of the following packages:
  • rubygems and yelp/yelp-xsl, prepared by Lucas Kanashiro
  • simplesamlphp, prepared by Tobias Frost
  • libbson-xs-perl, prepared by Roberto C. S nchez
  • fossil, prepared by Sylvain Beucler
  • setuptools and mydumper, prepared by Lee Garrett
  • redis and webpy, prepared by Adrian Bunk
  • xrdp, prepared by Abhijith PA
  • tcpdf, prepared by Santiago Ruano Rinc n
  • kmail-account-wizard, prepared by Thorsten Alteholz
Other contributions were also made by LTS Team members to packages in unstable:
  • proftpd-dfsg DEP-8 tests (autopkgtests) were provided to the maintainer, prepared by Lucas Kanashiro
  • a regular upload of libsoup2.4, prepared by Sean Whitton
  • a regular upload of setuptools, prepared by Lee Garrett
Freexian, the entity behind the management of the Debian LTS project, has been working for some time now on the development of an advanced CI platform for Debian-based distributions, called Debusine. Recently, Debusine has reached a level of feature implementation that makes it very usable. Some members of the LTS Team have been using Debusine informally, and during May LTS coordinator Santiago Ruano Rinc n has made a call for the team to help with testing of Debusine, and to help evaluate its suitability for the LTS Team to eventually begin using as the primary mechanism for uploading packages into Debian. Team members who have started using Debusine are providing valuable feedback to the Debusine development team, thus helping to improve the platform for all users. Actually, a number of updates, for both bullseye and bookworm, made during the month of May were handled using Debusine, e.g. rubygems s DLA-4163-1. By the way, if you are a Debian Developer, you can easily test Debusine following the instructions found at https://wiki.debian.org/DebusineDebianNet. DebConf, the annual Debian Conference, is coming up in July and, as is customary each year, the week preceding the conference will feature an event called DebCamp. The DebCamp week provides an opportunity for teams and other interested groups/individuals to meet together in person in the same venue as the conference itself, with the purpose of doing focused work, often called sprints . LTS coordinator Roberto C. S nchez has announced that the LTS Team is planning to hold a sprint primarily focused on the Debian security tracker and the associated tooling used by the LTS Team and the Debian Security Team.

Thanks to our sponsors Sponsors that joined recently are in bold.

5 June 2025

Matthew Garrett: Twitter's new encrypted DMs aren't better than the old ones

(Edit: Twitter could improve this significantly with very few changes - I wrote about that here. It's unclear why they'd launch without doing that, since it entirely defeats the point of using HSMs)

When Twitter[1] launched encrypted DMs a couple
of years ago, it was the worst kind of end-to-end
encrypted - technically e2ee, but in a way that made it relatively easy for Twitter to inject new encryption keys and get everyone's messages anyway. It was also lacking a whole bunch of features such as "sending pictures", so the entire thing was largely a waste of time. But a couple of days ago, Elon announced the arrival of "XChat", a new encrypted message platform built on Rust with (Bitcoin style) encryption, whole new architecture. Maybe this time they've got it right?

tl;dr - no. Use Signal. Twitter can probably obtain your private keys, and admit that they can MITM you and have full access to your metadata.

The new approach is pretty similar to the old one in that it's based on pretty straightforward and well tested cryptographic primitives, but merely using good cryptography doesn't mean you end up with a good solution. This time they've pivoted away from using the underlying cryptographic primitives directly and into higher level abstractions, which is probably a good thing. They're using Libsodium's boxes for message encryption, which is, well, fine? It doesn't offer forward secrecy (if someone's private key is leaked then all existing messages can be decrypted) so it's a long way from the state of the art for a messaging client (Signal's had forward secrecy for over a decade!), but it's not inherently broken or anything. It is, however, written in C, not Rust[2].

That's about the extent of the good news. Twitter's old implementation involved clients generating keypairs and pushing the public key to Twitter. Each client (a physical device or a browser instance) had its own private key, and messages were simply encrypted to every public key associated with an account. This meant that new devices couldn't decrypt old messages, and also meant there was a maximum number of supported devices and terrible scaling issues and it was pretty bad. The new approach generates a keypair and then stores the private key using the Juicebox protocol. Other devices can then retrieve the private key.

Doesn't this mean Twitter has the private key? Well, no. There's a PIN involved, and the PIN is used to generate an encryption key. The stored copy of the private key is encrypted with that key, so if you don't know the PIN you can't decrypt the key. So we brute force the PIN, right? Juicebox actually protects against that - before the backend will hand over the encrypted key, you have to prove knowledge of the PIN to it (this is done in a clever way that doesn't directly reveal the PIN to the backend). If you ask for the key too many times while providing the wrong PIN, access is locked down.

But this is true only if the Juicebox backend is trustworthy. If the backend is controlled by someone untrustworthy[3] then they're going to be able to obtain the encrypted key material (even if it's in an HSM, they can simply watch what comes out of the HSM when the user authenticates if there's no validation of the HSM's keys). And now all they need is the PIN. Turning the PIN into an encryption key is done using the Argon2id key derivation function, using 32 iterations and a memory cost of 16MB (the Juicebox white paper says 16KB, but (a) that's laughably small and (b) the code says 16 * 1024 in an argument that takes kilobytes), which makes it computationally and moderately memory expensive to generate the encryption key used to decrypt the private key. How expensive? Well, on my (not very fast) laptop, that takes less than 0.2 seconds. How many attempts to I need to crack the PIN? Twitter's chosen to fix that to 4 digits, so a maximum of 10,000. You aren't going to need many machines running in parallel to bring this down to a very small amount of time, at which point private keys can, to a first approximation, be extracted at will.

Juicebox attempts to defend against this by supporting sharding your key over multiple backends, and only requiring a subset of those to recover the original. I can't find any evidence that Twitter's does seem to be making use of this,Twitter uses three backends and requires data from at least two, but all the backends used are under x.com so are presumably under Twitter's direct control. Trusting the keystore without needing to trust whoever's hosting it requires a trustworthy communications mechanism between the client and the keystore. If the device you're talking to can prove that it's an HSM that implements the attempt limiting protocol and has no other mechanism to export the data, this can be made to work. Signal makes use of something along these lines using Intel SGX for contact list and settings storage and recovery, and Google and Apple also have documentation about how they handle this in ways that make it difficult for them to obtain backed up key material. Twitter has no documentation of this, and as far as I can tell does nothing to prove that the backend is in any way trustworthy. (Edit to add: The Juicebox API does support authenticated communication between the client and the HSM, but that relies on you having some way to prove that the public key you're presented with corresponds to a private key that only exists in the HSM. Twitter gives you the public key whenever you communicate with them, so even if they've implemented this properly you can't prove they haven't made up a new key and MITMed you the next time you retrieve your key)

On the plus side, Juicebox is written in Rust, so Elon's not 100% wrong. Just mostly wrong.

But ok, at least you've got viable end-to-end encryption even if someone can put in some (not all that much, really) effort to obtain your private key and render it all pointless? Actually no, since you're still relying on the Twitter server to give you the public key of the other party and there's no out of band mechanism to do that or verify the authenticity of that public key at present. Twitter can simply give you a public key where they control the private key, decrypt the message, and then reencrypt it with the intended recipient's key and pass it on. The support page makes it clear that this is a known shortcoming and that it'll be fixed at some point, but they said that about the original encrypted DM support and it never was, so that's probably dependent on whether Elon gets distracted by something else again. And the server knows who and when you're messaging even if they haven't bothered to break your private key, so there's a lot of metadata leakage.

Signal doesn't have these shortcomings. Use Signal.

[1] I'll respect their name change once Elon respects his daughter

[2] There are implementations written in Rust, but Twitter's using the C one with these JNI bindings

[3] Or someone nominally trustworthy but who's been compelled to act against your interests - even if Elon were absolutely committed to protecting all his users, his overarching goals for Twitter require him to have legal presence in multiple jurisdictions that are not necessarily above placing employees in physical danger if there's a perception that they could obtain someone's encryption keys

comment count unavailable comments

29 May 2025

Arthur Diniz: Bringing Kubernetes Back to Debian

I ve been part of the Debian Project since 2019, when I attended DebConf held in Curitiba, Brazil. That event sparked my interest in the community, packaging, and how Debian works as a distribution. In the early years of my involvement, I contributed to various teams such as the Python, Golang and Cloud teams, packaging dependencies and maintaining various tools. However, I soon felt the need to focus on packaging software I truly enjoyed, tools I was passionate about using and maintaining. That s when I turned my attention to Kubernetes within Debian.

A Broken Ecosystem The Kubernetes packaging situation in Debian had been problematic for some time. Given its large codebase and complex dependency tree, the initial packaging approach involved vendorizing all dependencies. While this allowed a somewhat functional package to be published, it introduced several long-term issues, especially security concerns. Vendorized packages bundle third-party dependencies directly into the source tarball. When vulnerabilities arise in those dependencies, it becomes difficult for Debian s security team to patch and rebuild affected packages system-wide. This approach broke Debian s best practices, and it eventually led to the abandonment of the Kubernetes source package, which had stalled at version 1.20.5. Due to this abandonment, critical bugs emerged and the package was removed from Debian s testing channel, as we can see in the package tracker.

New Debian Kubernetes Team Around this time, I became a Debian Maintainer (DM), with permissions to upload certain packages. I saw an opportunity to both contribute more deeply to Debian and to fix Kubernetes packaging. In early 2024, just before DebConf Busan in South Korea, I founded the Debian Kubernetes Team. The mission of the team was to repackage Kubernetes in a maintainable, security-conscious, and Debian-compliant way. At DebConf, I shared our progress with the broader community and received great feedback and more visibility, along with people interested in contributing to the team. Our first tasks was to migrate existing Kubernetes-related tools such as kubectx, kubernetes-split-yaml and kubetail into a dedicated namespace on Salsa, Debian s GitLab instance. Many of these tools were stored across different teams (like the Go team), and consolidating them helped us organize development and focus our efforts.

De-vendorizing Kubernetes Our main goal was to un-vendorize Kubernetes and bring it up-to-date with upstream releases. This meant:
  • Removing the vendor directory and all embedded third-party code.
  • Trimming the build scope to focus solely on building kubectl, Kubernetes CLI.
  • Using Files-Excluded in debian/copyright to cleanly drop unneeded files during source imports.
  • Rebuilding the dependency tree, ensuring all Go modules were separately packaged in Debian.
We used uscan, a standard Debian packaging tool that fetches upstream tarballs and prepares them accordingly. The Files-Excluded directive in our debian/copyright file instructed uscan to automatically remove unnecessary files during the repackaging process:
$ uscan
Newest version of kubernetes on remote site is 1.32.3, specified download version is 1.32.3
Successfully repacked ../v1.32.3 as ../kubernetes_1.32.3+ds.orig.tar.gz, deleting 30616 files from it.
The results were dramatic. By comparing the original upstream tarball with our repackaged version, we can see that our approach reduced the tarball size by over 75%:
$ du -h upstream-v1.32.3.tar.gz kubernetes_1.32.3+ds.orig.tar.gz
14M	upstream-v1.32.3.tar.gz
3.2M	kubernetes_1.32.3+ds.orig.tar.gz
This significant reduction wasn t just about saving space. By removing over 30,000 files, we simplified the package, making it more maintainable. Each dependency could now be properly tracked, updated, and patched independently, resolving the security concerns that had plagued the previous packaging approach.

Dependency Graph To give you an idea of the complexity involved in packaging Kubernetes for Debian, the image below is a dependency graph generated with debtree, visualizing all the Go modules and other dependencies required to build the kubectl binary. kubectl-depgraph This web of nodes and edges represents every module and its relationship during the compilation process of kubectl. Each box is a Debian package, and the lines connecting them show how deeply intertwined the ecosystem is. What might look like a mess of blue spaghetti is actually a clear demonstration of the vast and interconnected upstream world that tools like kubectl rely on. But more importantly, this graph is a testament to the effort that went into making kubectl build entirely using Debian-packaged dependencies only, no vendoring, no downloading from the internet, no proprietary blobs.

Upstream Version 1.32.3 and Beyond After nearly two years of work, we successfully uploaded version 1.32.3+ds of kubectl to Debian unstable. kubernetes/-/merge_requests/1 The new package also includes:
  • Zsh, Fish, and Bash completions installed automatically
  • Man pages and metadata for improved discoverability
  • Full integration with kind and docker for testing purposes

Integration Testing with Autopkgtest To ensure the reliability of kubectl in real-world scenarios, we developed a new autopkgtest suite that runs integration tests using real Kubernetes clusters created via Kind. Autopkgtest is a Debian tool used to run automated tests on binary packages. These tests are executed after the package is built but before it s accepted into the Debian archive, helping catch regressions and integration issues early in the packaging pipeline. Our test workflow validates kubectl by performing the following steps:
  • Installing Kind and Docker as test dependencies.
  • Spinning up two local Kubernetes clusters.
  • Switching between cluster contexts to ensure multi-cluster support.
  • Deploying and scaling a sample nginx application using kubectl.
  • Cleaning up the entire test environment to avoid side effects.
  • debian/tests/kubectl.sh

Popcon: Measuring Adoption To measure real-world usage, we rely on data from Debian s popularity contest (popcon), which gives insight into how many users have each binary installed. popcon-graph popcon-table Here s what the data tells us:
  • kubectl (new binary): Already installed on 2,124 systems.
  • golang-k8s-kubectl-dev: This is the Go development package (a library), useful for other packages and developers who want to interact with Kubernetes programmatically.
  • kubernetes-client: The legacy package that kubectl is replacing. We expect this number to decrease in future releases as more systems transition to the new package.
Although the popcon data shows activity for kubectl before the official Debian upload date, it s important to note that those numbers represent users who had it installed from upstream source-lists, not from the Debian repositories. This distinction underscores a demand that existed even before the package was available in Debian proper, and it validates the importance of bringing it into the archive.
Also worth mentioning: this number is not the real total number of installations, since users can choose not to participate in the popularity contest. So the actual adoption is likely higher than what popcon reflects.

Community and Documentation The team also maintains a dedicated wiki page which documents:
  • Maintained tools and packages
  • Contribution guidelines
  • Our roadmap for the upcoming Debian releases
https://debian-kubernetes.org

Looking Ahead to Debian 13 (Trixie) The next stable release of Debian will ship with kubectl version 1.32.3, built from a clean, de-vendorized source. This version includes nearly all the latest upstream features, and will be the first time in years that Debian users can rely on an up-to-date, policy-compliant kubectl directly from the archive. By comparing with upstream, our Debian package even delivers more out of the box, including shell completions, which the upstream still requires users to generate manually. In 2025, the Debian Kubernetes team will continue expanding our packaging efforts for the Kubernetes ecosystem. Our roadmap includes:
  • kubelet: The primary node agent that runs on each node. This will enable Debian users to create fully functional Kubernetes nodes without relying on external packages.
  • kubeadm: A tool for creating Kubernetes clusters. With kubeadm in Debian, users will then be able to bootstrap minimum viable clusters directly from the official repositories.
  • helm: The package manager for Kubernetes that helps manage applications through Kubernetes YAML files defined as charts.
  • kompose: A conversion tool that helps users familiar with docker-compose move to Kubernetes by translating Docker Compose files into Kubernetes resources.

Final Thoughts This journey was only possible thanks to the amazing support of the debian-devel-br community and the collective effort of contributors who stepped up to package missing dependencies, fix bugs, and test new versions. Special thanks to:
  • Carlos Henrique Melara (@charles)
  • Guilherme Puida (@puida)
  • Jo o Pedro Nobrega (@jnpf)
  • Lucas Kanashiro (@kanashiro)
  • Matheus Polkorny (@polkorny)
  • Samuel Henrique (@samueloph)
  • Sergio Cipriano (@cipriano)
  • Sergio Durigan Junior (@sergiodj)
I look forward to continuing this work, bringing more Kubernetes tools into Debian and improving the developer experience for everyone.

Arthur Diniz: Bringing Kubernetes Back to Debian

I ve been part of the Debian Project since 2019, when I attended DebConf held in Curitiba, Brazil. That event sparked my interest in the community, packaging, and how Debian works as a distribution. In the early years of my involvement, I contributed to various teams such as the Python, Golang and Cloud teams, packaging dependencies and maintaining various tools. However, I soon felt the need to focus on packaging software I truly enjoyed, tools I was passionate about using and maintaining. That s when I turned my attention to Kubernetes within Debian.

A Broken Ecosystem The Kubernetes packaging situation in Debian had been problematic for some time. Given its large codebase and complex dependency tree, the initial packaging approach involved vendorizing all dependencies. While this allowed a somewhat functional package to be published, it introduced several long-term issues, especially security concerns. Vendorized packages bundle third-party dependencies directly into the source tarball. When vulnerabilities arise in those dependencies, it becomes difficult for Debian s security team to patch and rebuild affected packages system-wide. This approach broke Debian s best practices, and it eventually led to the abandonment of the Kubernetes source package, which had stalled at version 1.20.5. Due to this abandonment, critical bugs emerged and the package was removed from Debian s testing channel, as we can see in the package tracker.

New Debian Kubernetes Team Around this time, I became a Debian Maintainer (DM), with permissions to upload certain packages. I saw an opportunity to both contribute more deeply to Debian and to fix Kubernetes packaging. In early 2024, just before DebConf Busan in South Korea, I founded the Debian Kubernetes Team. The mission of the team was to repackage Kubernetes in a maintainable, security-conscious, and Debian-compliant way. At DebConf, I shared our progress with the broader community and received great feedback and more visibility, along with people interested in contributing to the team. Our first tasks was to migrate existing Kubernetes-related tools such as kubectx, kubernetes-split-yaml and kubetail into a dedicated namespace on Salsa, Debian s GitLab instance. Many of these tools were stored across different teams (like the Go team), and consolidating them helped us organize development and focus our efforts.

De-vendorizing Kubernetes Our main goal was to un-vendorize Kubernetes and bring it up-to-date with upstream releases. This meant:
  • Removing the vendor directory and all embedded third-party code.
  • Trimming the build scope to focus solely on building kubectl, Kubernetes CLI.
  • Using Files-Excluded in debian/copyright to cleanly drop unneeded files during source imports.
  • Rebuilding the dependency tree, ensuring all Go modules were separately packaged in Debian.
We used uscan, a standard Debian packaging tool that fetches upstream tarballs and prepares them accordingly. The Files-Excluded directive in our debian/copyright file instructed uscan to automatically remove unnecessary files during the repackaging process:
$ uscan
Newest version of kubernetes on remote site is 1.32.3, specified download version is 1.32.3
Successfully repacked ../v1.32.3 as ../kubernetes_1.32.3+ds.orig.tar.gz, deleting 30616 files from it.
The results were dramatic. By comparing the original upstream tarball with our repackaged version, we can see that our approach reduced the tarball size by over 75%:
$ du -h upstream-v1.32.3.tar.gz kubernetes_1.32.3+ds.orig.tar.gz
14M	upstream-v1.32.3.tar.gz
3.2M	kubernetes_1.32.3+ds.orig.tar.gz
This significant reduction wasn t just about saving space. By removing over 30,000 files, we simplified the package, making it more maintainable. Each dependency could now be properly tracked, updated, and patched independently, resolving the security concerns that had plagued the previous packaging approach.

Dependency Graph To give you an idea of the complexity involved in packaging Kubernetes for Debian, the image below is a dependency graph generated with debtree, visualizing all the Go modules and other dependencies required to build the kubectl binary. kubectl-depgraph This web of nodes and edges represents every module and its relationship during the compilation process of kubectl. Each box is a Debian package, and the lines connecting them show how deeply intertwined the ecosystem is. What might look like a mess of blue spaghetti is actually a clear demonstration of the vast and interconnected upstream world that tools like kubectl rely on. But more importantly, this graph is a testament to the effort that went into making kubectl build entirely using Debian-packaged dependencies only, no vendoring, no downloading from the internet, no proprietary blobs.

Upstream Version 1.32.3 and Beyond After nearly two years of work, we successfully uploaded version 1.32.3+ds of kubectl to Debian unstable. kubernetes/-/merge_requests/1 The new package also includes:
  • Zsh, Fish, and Bash completions installed automatically
  • Man pages and metadata for improved discoverability
  • Full integration with kind and docker for testing purposes

Integration Testing with Autopkgtest To ensure the reliability of kubectl in real-world scenarios, we developed a new autopkgtest suite that runs integration tests using real Kubernetes clusters created via Kind. Autopkgtest is a Debian tool used to run automated tests on binary packages. These tests are executed after the package is built but before it s accepted into the Debian archive, helping catch regressions and integration issues early in the packaging pipeline. Our test workflow validates kubectl by performing the following steps:
  • Installing Kind and Docker as test dependencies.
  • Spinning up two local Kubernetes clusters.
  • Switching between cluster contexts to ensure multi-cluster support.
  • Deploying and scaling a sample nginx application using kubectl.
  • Cleaning up the entire test environment to avoid side effects.
  • debian/tests/kubectl.sh

Popcon: Measuring Adoption To measure real-world usage, we rely on data from Debian s popularity contest (popcon), which gives insight into how many users have each binary installed. popcon-graph popcon-table Here s what the data tells us:
  • kubectl (new binary): Already installed on 2,124 systems.
  • golang-k8s-kubectl-dev: This is the Go development package (a library), useful for other packages and developers who want to interact with Kubernetes programmatically.
  • kubernetes-client: The legacy package that kubectl is replacing. We expect this number to decrease in future releases as more systems transition to the new package.
Although the popcon data shows activity for kubectl before the official Debian upload date, it s important to note that those numbers represent users who had it installed from upstream source-lists, not from the Debian repositories. This distinction underscores a demand that existed even before the package was available in Debian proper, and it validates the importance of bringing it into the archive.
Also worth mentioning: this number is not the real total number of installations, since users can choose not to participate in the popularity contest. So the actual adoption is likely higher than what popcon reflects.

Community and Documentation The team also maintains a dedicated wiki page which documents:
  • Maintained tools and packages
  • Contribution guidelines
  • Our roadmap for the upcoming Debian releases
https://debian-kubernetes.org

Looking Ahead to Debian 13 (Trixie) The next stable release of Debian will ship with kubectl version 1.32.3, built from a clean, de-vendorized source. This version includes nearly all the latest upstream features, and will be the first time in years that Debian users can rely on an up-to-date, policy-compliant kubectl directly from the archive. By comparing with upstream, our Debian package even delivers more out of the box, including shell completions, which the upstream still requires users to generate manually. In 2025, the Debian Kubernetes team will continue expanding our packaging efforts for the Kubernetes ecosystem. Our roadmap includes:
  • kubelet: The primary node agent that runs on each node. This will enable Debian users to create fully functional Kubernetes nodes without relying on external packages.
  • kubeadm: A tool for creating Kubernetes clusters. With kubeadm in Debian, users will then be able to bootstrap minimum viable clusters directly from the official repositories.
  • helm: The package manager for Kubernetes that helps manage applications through Kubernetes YAML files defined as charts.
  • kompose: A conversion tool that helps users familiar with docker-compose move to Kubernetes by translating Docker Compose files into Kubernetes resources.

Final Thoughts This journey was only possible thanks to the amazing support of the debian-devel-br community and the collective effort of contributors who stepped up to package missing dependencies, fix bugs, and test new versions. Special thanks to:
  • Carlos Henrique Melara (@charles)
  • Guilherme Puida (@puida)
  • Jo o Pedro Nobrega (@jnpf)
  • Lucas Kanashiro (@kanashiro)
  • Matheus Polkorny (@polkorny)
  • Samuel Henrique (@samueloph)
  • Sergio Cipriano (@cipriano)
  • Sergio Durigan Junior (@sergiodj)
I look forward to continuing this work, bringing more Kubernetes tools into Debian and improving the developer experience for everyone.

27 May 2025

Russell Coker: Leaf ZE1

I ve just got a second hand Nissan LEAF. It s not nearly as luxurious as the Genesis EV that I test drove [1]. It s also just over 5 years old so it s not as slick as the MG4 I test drove [2]. But the going rate for a LEAF of that age is $17,000 vs $35,000 or more for a new MG4 or $130,000+ for a Genesis. At this time the LEAF is the only EV in Australia that s available on the second hand market in quantity. Apparently the cheapest new EV in Australia is a Great Wall one which is $32,000 and which had a wait list last time I checked, so $17,000 is a decent price if you want an electric car and aren t interested in paying the price of a new car. Starting the Car One thing I don t like about most recent cars (petrol as well as electric) is that they needlessly break traditions of car design. Inserting a key and turning it clockwise to start a car is a long standing tradition that shouldn t be broken without a good reason. With the use of traditional keys you know that when a car has the key removed it can t be operated, there s no situation of the person with the key walking away and leaving the car driveable and there s no possibility of the owner driving somewhere without the key and then being unable to start it. To start a LEAF you have to have the key fob device in range, hold down the brake pedal, and then press the power button. To turn on accessories you do the same but without holding down the brake pedal. They also have patterns of pushes, push twice to turn it on, push three times to turn it off. This is all a lot easier with a key where you can just rotate it as many clicks as needed. The change of car design for the key means that no physical contact is needed to unlock the car. If someone stands by a car fiddling with the door lock it will get noticed which deters certain types of crime. If a potential thief can sit in a nearby car to try attack methods and only walk to the target vehicle once it s unlocked it makes the crime a lot easier. Even if the electronic key is as secure as a physical key allowing attempts to unlock remotely weakens security. Reports on forums suggest that the electronic key is vulnerable to replay attacks. I guess I just have to hope that as car thieves typically get less than 10% of the value of a car it s just not worth their effort to steal a $17,000 car. Unlocking doors remotely is a common feature that s been around for a while but starting a car without a key being physically inserted is a new thing. Other Features The headlights turn on automatically when the car thinks that the level of ambient light warrants it. There is an option to override this to turn on lights but no option to force the lights to be off. So if you have your car in the on state while parked the headlights will be on even if you are parked and listening to the radio. The LEAF has a bunch of luxury features which seem a bit ridiculous like seat warmers. It also has a heated steering wheel which has turned out to be a good option for me as I have problems with my hands getting cold. According to the My Nissan LEAF Forum the seat warmer uses a maximum of 50W per seat while the car heater uses a minimum of 250W [3]. So if there are one or two people in the car then significantly less power is used by just heating the seats and also keeping the car air cool reduces window fog. The Bluetooth audio support works well. I ve done hands free calls and used it for playing music from my phone. This is the first car I ve owned with Bluetooth support. It also has line-in which might have had some use in 2019 but is becoming increasingly useless as phones with Bluetooth become more popular. It has support for two devices connecting via Bluetooth at the same time which could be handy if you wanted to watch movies on a laptop or tablet while waiting for someone. The LEAF has some of the newer safety features, it tracks lane markers and notifies the driver via beeps and vibration if they stray from their lane. It also tries to read speed limit signs and display the last observed speed limit on the dash display. It also has a skid alert which in my experience goes off under hard acceleration when it s not skidding but doesn t go off if you lose grip when cornering. The features for detecting changing lanes when close to other cars and for emergency braking when another car is partly in the lane (even if moving out of the lane) don t seem well tuned for Australian driving, the common trend on Australian roads is lawful-evil to use DND terminology. Range My most recent driving was just over 2 hours driving with a distance of a bit over 100Km which took the battery from 62% to 14%. So it looks like I can drive a bit over 200Km at an average speed of 50Km/h. I have been unable to find out the battery size for my car, my model will have either a 40KWh or 62KWh battery. Google results say it should be printed on the B pillar (it s not) and that it can be deduced from the VIN (it can t). I m guessing that my car is the cheaper option which is supposed to do 240Km when new which means that a bit over 200Km at an average speed of 50Km/h when 6yo is about what s expected. If it has the larger battery designed to do 340Km then doing 200Km in real use would be rather disappointing. Assuming the battery is 40KWh that means it s 5Km/KWh or 10KW average for the duration. That means that the 250W or so used by the car heater should only make a about 2% difference to range which is something that a human won t usually notice. If I was to drive to another state I d definitely avoid using the heater or airconditioner as an extra 4km could really matter when trying to find a place to charge when you aren t familiar with the area. It s also widely reported that the LEAF is less efficient at highway speeds which is an extra difficulty for that. It seems that the LEAF just isn t designed for interstate driving in Australia, it would be fine for driving between provinces of the Netherlands as it s difficult to drive for 200km without leaving that country. Driving 700km to another city in a car with 200km range would mean charging 3 times along the way, that s 2 hours of charging time when using fast chargers. This isn t a problem at all as the average household in Australia has 1.8 cars and the battery electric vehicles only comprise 6.3% of the market. So if a household had a LEAF and a Prius they could just use the Prius for interstate driving. A recent Prius could drive from Melbourne to Canberra or Adelaide without refuelling on the way. If I was driving to another state a couple of times a year I could rent an old fashioned car to do that and still be saving money when compared to buying petrol all the time. Running Cost Currently I m paying about $0.28 per KWh for electricity, it s reported that the efficiency of charging a LEAF is as low as 83% with the best efficiency when fast charging. I don t own the fast charge hardware and don t plan to install it as that would require getting a replacement of the connection to my home from the street, a new switchboard, and other expenses. So I expect I ll be getting 83% efficiency when charging which means 48KWh for 200KM or 96KWH for the equivalent of a $110 tank of petrol. At $0.28/KWh it will cost $26 for the same amount of driving as $110 of petrol. I also anticipate saving money on service as there s no need for engine oil changes and all the other maintenance of a petrol engine and regenerative braking will reduce the incidence of brake pad replacement. I expect to save over $1100 per annum on using electricity instead of petrol even if I pay the full rate. But if I charge my car in the middle of the day when there is over supply and I don t get paid for feeding electricity from my solar panels into the grid (as is common nowadays) it could be almost free to charge the car and I could save about $1500 on fuel. Comfort Electric cars are much quieter than cars with petrol or Diesel engines which is a major luxury feature. This car is also significantly newer than any other car I ve driven much so it has features like Bluetooth audio which weren t in other cars I ve driven. When doing 100Km/h I can hear a lot of noise from the airflow, part of that would be due to the LEAF not having the extreme streamlining features that are associated with Teslas (such as retracting door handles) and part of that would be due to the car being older and the door seals not being as good as they were when new. It s still a very quiet car with a very smooth ride. It would be nice if they used the quality of seals and soundproofing that VW uses in the Passat but I guess the car would be heavier and have a shorter range if they did that. This car has less space for the driver than any other car I ve driven (with the possible exception of a 1989 Ford Laser AKA Mazda 323). The front seats have less space than the Prius. Also the batteries seem to be under the front seats so there s a bulge in the floor going slightly in front of the front seats when they are moved back which gives less space for the front passenger to move their legs and less space for the driver when sitting in a parked car. There are a selection of electric cars from MG, BYD, and Great Wall that have more space in the front seats, if those cars were on the second hand market I might have made a different choice but a second hand LEAF is the only option for a cheap electric car in Australia now. The heated steering wheel and heated seats took a bit of getting used to but I have come to appreciate the steering wheel and the heated seats are a good way of extending the range of the car. Misc Notes The LEAF is a fun car to drive and being quiet is a luxury feature, it s no different to other EVs in this regard. It isn t nearly as fast as a Tesla, but is faster than most cars actually drive on the road. When I was looking into buying a LEAF from one of the car sales sites I was looking at models less than 5 years old. But the ZR1 series went from 2017 to 2023 so there s probably not much difference between a 2019 model and a 2021 model but there is a significant price difference. I didn t deliberately choose a 2019 car, it was what a relative was selling at a time when I needed a new car. But knowing what I know now I d probably look at that age of LEAF if choosing from the car sales sites. Problems When I turn the car off the side mirrors fold in but when I turn it on they usually don t automatically unfold if I have anything connected to the cigarette lighter power port. This is a well known problem and documented on forums. This is something that Nissan really should have tested before release because phone chargers that connect to the car cigarette lighter port have been common for at least 6 years before my car was manufactured and at least 4 years before the ZE1 model was released. The built in USB port doesn t supply enough power to match the power use of a Galaxy Note 9 running Google maps and playing music through Bluetooth. On it s own this isn t a big deal but combined with the mirror issue of using a charger in the cigarette lighter port it s a problem. The cover over the charging ports doesn t seem to lock easily enough, I had it come open when doing 100Km/h on a freeway. This wasn t a big deal but as the cover opens in a suicide-door manner at a higher speed it could have broken off. The word is that LEAF service in Australia is not done well. Why do you need regular service of an electric car anyway? For petrol and Diesel cars it s engine oil replacement that makes it necessary to have regular service. Surely you can just drive it until either the brakes squeak or the tires seem worn. I have been having problems charging, sometimes it will charge from ~20% to 100% in under 24 hours, sometimes in 14+ hours it only gets to 30%. Conclusion This is a good car and the going price on them is low. I generally recommend them as long as you aren t really big and aren t too worried about the poor security. It s a fun car to drive even with a few annoying things like the mirrors not automatically extending on start. The older ones like this are cheap enough that they should be able to cover the entire purchase cost in 10 years by the savings from not buying petrol even if you don t drive a lot. With a petrol car I use about 13 tanks of petrol a year so my driving is about half the average for Australia. Some people could cover the purchase price of a second hand leaf in under 5 years.

Next.

Previous.