The last two weeks of the year, and also the first one of the new year, are vacation season in many parts of the world and YaSTland is not an exception. But before we enter hibernation mode, let’s take a look to the most important features and bugfixes we implemented in the last sprint of 2019. That includes:

• bringing back to life some sections of the Software Manager,
• implementing system upgrade with the new SLE media types,
• making the installation in Raspberry Pi and IBM Z System even better,
• improving usability of encryption,
• reducing the footprint of the Snapper plugin for ZYpp,
• as always, many other small improvements and fixes.

Restored Some Package Views: Recommended, Suggested, etc.

Let’s start with a redemption story. Some time ago we implemented feature fate#326485 which requested dropping the “Package Groups” view from the package manager UI. That was quite an easy task.

However, a few weeks later we got a bug report that the lists of recommended, suggested, etc… packages couldn’t be displayed anymore. It turned out that, in the Qt package manager front-end, the removed “Package Groups” view not only used to display the static group data from the packages but it also contained some special computed package lists like orphaned, suggested or recommended packages. So these lists were lost as a collateral damage of removing the “Package Groups” view.

The ncurses package manager was not affected by the same problem because, in that front-end, those views are grouped in a separate “Package Classification” section. So the task for this sprint was to somehow revive the lists in Qt and make them again available to the users.

We partly reverted the Package Groups removal and restored displaying those special package groups. To make it consistent we also use the “Package Classification” name for the view, like in the ncurses package manager.

On the other hand, the ncurses front-end was missing some lists like the “Multiversion Packages” and “All Packages”. To take consistency another step further, we added these missing lists and did some small cleanup and fixes so now both the Qt and the ncurses package managers should offer the same functionality and should look similar.

User-friendly Encryption Device Names

And talking about bug reports that trigger some usability revamp, some users had pointed that, when the system is booting and prompts for the password of an encrypted device, it’s not always that easy to identify which exact device it is referring to:

The root of the problem is that when YaST creates an encryption device (during the installation by means of the storage proposal, or manually with the Expert Partitioner), the device mapper name for the new encrypted device is generated from the udev id of the underlying device (e.g., cr_ccw-0XAF5E-part2).

We decided to improve the encryption naming generation in YaST for Tumbleweed and future releases of Leap and SLE. From now on, the name will be based on the mount point of the device. For example, if an encrypted device is going to be mounted at root, its device mapper name would be cr_root. In general, when the encrypted device is mounted, the device mapper name would be cr_mount_point (e.g., cr_home_linux for an encrypted device mounted at /home/linux).

Note that udev-based names might still be used for some scenarios. For example, when the device is not mounted or for an indirectly used encrypted device (e.g., an encrypted LVM Physical Volume).

And related to the identification of encryption devices, we have also added more information about the device when the encryption devices are activated during the installation process. Providing the password for the correct device was very difficult because the user needed to know the UUID of the encryption device. Now on, the activation popup also informs about the kernel name of the underlying device, making much easier to identify it.

Because names matter… which leads us to the next topic.

How does it Feel to Run a Mainframe?

As you may know, (open)SUSE runs in a vast range of hardware, including powerful mainframes like the IBM Z family. One of the strengths of our beloved distributions is that, despite the differences in hardware and scope, the installation and usage experience is very similar in all the supported systems.

Consistency and ease of use are good, but when you drive a luxury car you want to see the brand’s badge on top of the hood. So in future versions of the installer, the model of the machine will be displayed when installing in an IBM Z system. See the right-top corner of the following screenshot.

The text-based installer also has been modified to include the same banner in a similar place.

But in the same way that (open)SUSE enables you to install and use Linux in a mainframe “just like in any other computer”, we also target to do the same in the other extreme of the hardware spectrum.

Better Support for Raspberry Pi in the Partitioning Proposal

One year ago we announced that openSUSE Leap 15.1 and SLE-15-SP1 would be the first Linux distributions that could be installed in Raspberry Pi devices following the standard installation procedure, instead of deploying a Raspberry-specific pre-built image. The only prerequisite was the existence in the target SD card (or disk) of a partition containing the Raspberry Pi boot code.

But we are now able to go one step further for SLE-15-SP2 (and Leap 15.2). Thanks to the technologies included in those upcoming releases, (open)SUSE will not longer need a separate partition with the boot code in all cases. Now the installer can make a reasonable installation proposal in all situations, even if the target storage device doesn’t contain a booting partition in advance. See, for example, what the installer suggests by default for installing a fully standard SLE-15-SP2 Beta1 in a 32 GiB SD card that contained initially a GPT partition table (tip: GPT partition tables cannot be used to boot in a Raspberry Pi device… and the installer knows it).

With that, the installation of the standard SLE-15-SP2 Beta1 (the aarch64 version, of course) in a Raspberry Pi 3 or 4 is as easy as “next”, “next”, “next”… with the only exception of a couple of packages that must be manually selected for installation (raspberrypi-* and u-boot-rpi3). Hopefully, future beta images of both SLE and openSUSE Leap 15.2 will select those packages automatically when installing in a Pi, which will make the (open)SUSE experience in those devices basically identical to any other computer.

SLE Upgrade with the New Media Types

And talking about the standard installation images of the upcoming SLE-15-SP2, we explained in our previous blog post that those versions of SUSE Linux Enterprise (SLE) and all its associated products will be distributed in two new kinds of media – Full and Online. The Full Media contains many repositories and the system can be installed without network connectivity. The Online Media is similar to the openSUSE’s net installer, it contains no repository and it must download everything from the network. The big difference with openSUSE is that SLE systems need to be registered in order to have access to remote repositories.

But apart from installation, those two new media types can also be used to upgrade an existing system… at least after all the improvements implemented during the latest sprint.

In the case of the Online Media, if the system is registered the upgrade process will switch all repositories to point to their corresponding versions at the SUSE Customer Center (SCC) and will get the new software from there. If the system is not registered, the upgrade process is cancelled and the user is advised to either register the system or use the Full Media.

The Full Media can be used to upgrade any system, registered or not, but the process is different in each case. For a non-registered system, the repositories will be switched to the ones included in the media and the system will be upgraded from there. For registered systems the process is the same that with the Online Media, so the software will be fetched from the remote repositories at the SUSE Customer Center.

Last but not least, we also made sure the process with both medias works with an AutoYaST upgrade (yes, you can also use AutoYaST to perform an unattended upgrade, in addition to the better known unattended installation). For a registered system, we simplified the procedure as much as possible and it only needs access to SCC and an empty AutoYaST profile. For non-registered systems it is a little bit more complex because the profile must specify which repositories from the media should be used for the upgrade. But other than that, the process works quite smooth.

And, of course, we used the opportunity to improve the unit test coverage of the code and to improve the documentation, including the profiles we used for testing.

The Snapper Plugin for ZYpp Becomes More Compact and Future-proof

Snapper lets you make filesystem snapshots. It has a companion, snapper-zypp-plugin, a plugin for ZYpp that makes snapshots automatically during commits. See the “zypp” descriptions in this listing:

# snapper list

  # | Type   |Pre # | Date                     | User | Used Space | Cleanup  | Description  | Userdata     
----+--------+------+--------------------------+------+------------+----------+--------------+-------------
  0 | single |      |                          | root |            |          | current      |              
[...]
824 | pre    |      | Tue Dec 17 10:00:27 2019 | root |  16.00 KiB | number   | zypp(zypper) | important=no
826 | post   |  824 | Tue Dec 17 10:02:19 2019 | root |  16.00 KiB | number   |              | important=no
827 | single |      | Tue Dec 17 11:00:01 2019 | root |  16.00 KiB | timeline | timeline     |             
828 | single |      | Tue Dec 17 11:00:01 2019 | root |  16.00 KiB | timeline | timeline     |             

To make our enterprise products supportable for a looong time, we have rewritten this plugin to C++, starting with snapper-0.8.7. (The original Python implementation is not dead, it is resting in old Git commits.)

As a result, Python regular expressions are no longer supported in the /etc/snapper/zypp.conf file. POSIX extended regular expressions work instead, which should work sufficiently well for the purpose of package name matching. Shell patterns continue working unchanged.

Happy new year!

During the following three weeks, the YaST team will interrupt the usual sprint-based development pace. That also means, almost for sure, that we will not publish any blog post about the development of YaST until mid January of 2020. So we want to take this opportunity to wish you a happy new year full of joy and Free Software.

See you soon and make sure to start the year with a lot of fun!

Time goes by and the YaST wheel keeps rolling. So let’s take a look to what have moved since our previous development report.

More flexible NET installation ISOs

Network installation media for Tumbleweed or Leap only work properly with the exact repository they have been built for – which for Tumbleweed may mean they could be outdated after just one day.

You would then run into this message:

To improve the situation the installer can now offer to download matching boot files (kernel and initrd, to be precise) from the repository if it detects this situation:

Of course, you can say ‘No’ here – but then you’re back to the red dialog. 😀

Technically, what’s done is to download a new kernel/initrd pair from the repository and restart the installation process with them (using kexec). So be prepared for a slight déjà vu.

This feature is controlled by the kexec boot option.

Storage-ng lands into Tumbleweed: handle with care

But that’s not the only news we have about openSUSE Tumbleweed. Our usual readers already know about Storage-ng, our effort to rewrite the whole YaST storage stack from scratch. And they also know it’s still a work in progress. But since there were too many valuable changes blocked by the adoption of Storage-ng, it was decided it was time to push the red button. So we are glad to announce the Storage-ng era has started with its inclusion in the first official (open)SUSE product – starting with snapshot 20180117, libstorage-ng has replaced libstorage and, thus, yast2-storage-ng has replaced yast2-storage.

They say forewarned is forearmed, so an article was published in advance in news.opensuse.org to set the expectations and to provide and overview of the current status. We would like to encourage all openSUSE Tumbleweed users to (re)visit the article to get a better picture of the situation.

Alignment of partitions in the expert partitioner

An important part of that work in progress is the re-implementation of the Expert Partitioner with Storage-ng technologies. As mentioned many times in previous posts, this is mainly a 1:1 clone, with the same functionality presented in exactly the same way than the classic YaST partitioner. But some times we take the opportunity to introduce some improvement here and there, as we did this week with a topic that can have a very noticeable impact in the system performance: partitions alignment.

Although many people is not aware of it, the partitions in a system must be properly aligned to avoid the performance drop caused by excessive read-modify-write cycles. For details please refer to the great article at Wikipedia explaining the topic, especially the sections titled “4 KB sector alignment” and “SSD page partition alignment”. Moreover, leaving performance considerations aside, some partition tables require alignment to simply work, like DASD partition tables which need alignment to tracks (usually 12 sectors).

The new expert partitioner takes all that into consideration when creating and resizing partitions, ensuring always the required alignment (like the DASD tracks) and encouraging the optional performance-related one, avoiding undesired gaps between partitions in the process.

Above you can see the dialog for choosing the size for a new partition that, unsurprisingly, looks very much like the same dialog in the pre-storage-ng Expert Partitioner. If a size is specified by the user in that dialog (any of the two first options in the form), the start and end of the partition will be aligned to ensure optimal performance and to minimize gaps. That may result in a slightly smaller partition (with the difference being usually less than 1MiB). If a custom region is specified, the start and end will be honored as closely as possible, with no performance optimizations (although mandatory alignment, like DASD tracks, still will take place). This third option is the best to create very small partitions.

The same considerations for optimal alignment will also be taken into account while resizing an existing partition and calculating the minimal and maximal sizes suggested by the partitioner during that process.

Sanity checks for the storage setup

The possibility of bypassing the performance optimizations in the Expert Partitioner is just one example of the (potentially unleashed) power that tool provides. As a consequence of that flexibility, sometimes the user can overlook some important setup configurations or even make mistakes. To help with that, the Expert Partitioner recovered this week its ability to check the entered storage setup.

Once the user has set partitions, LVM volumes, file systems, mount points, etc. and decides to proceed, the Partitioner will validate that setup to ensure it fulfills all necessary requirements for booting and running the system. When some issue is detected, a popup message is presented to show what the problem is, offering the option to ignore the warning and move forward.

Two kind of checks are carried out to ensure the partitioning setup validity. First, the presence of needed partitions for booting is checked. Booting requirements depends on the current architecture (x86, PowerPC, AArch, etc.) and other technical details like the partition table type (GPT vs MS-DOS). Then, the mandatory volumes for the current product are checked. The mandatory volumes are defined in the revamped partitioning section of the control file. Typically, only a volume for root and another for swap used to be mandatory, but now this is totally configurable by anyone defining the product (SLE, Leap, Tumbleweed, your own custom openSUSE derivative…).

As a bonus, all the sanity checks are now centralized (they used to be scattered around the YaST source code) and it’s easier to add new ones (you will miss some old checks at this moment) and to use them from other parts of YaST (like the bootloader module or AutoYaST).

More improvements in the Expert Partitioner

The new warnings and the alignment improvements commented above are not the only news on the evolution of the Expert Partitioner clone this week. Resizing of LVM devices has also been brought back to life, both for volume groups and logical volumes. In the case of logical volumes, the functionality is not much different, at least in the surface, from the partition resizing that was already present and that you can see in the screenshot of the alignment section.

On the other hand, in the context of the Partitioner, resizing a volume group actually means adding or removing physical volumes. Actions that are now possible again, including the corresponding checks. For example, a physical volume cannot be removed if it already exists on disk (that could destroy your data) or if the resulting size of the volume group is not enough to cover all its logical volumes.

Apart from the mentioned functionality, there has also been improvements in how the Expert Partitioner presents the information. For example, now the “type” column shows the correct label and icon for each device instead of that useless TODO label. Moreover, similar TODO marks were replaced by proper data in the device overview tab.

Minimize changes between the SLE15 “Installer” and “Packages” DVDs

The SUSE Enterprise Server 15 (SLES15) product can be installed from a bootable “Installer” DVD medium which contains the installer and a subset of packages needed for a very minimal system. The other packages are available either from a registration server (after registering the SLES product) or via a separate “Packages” DVD medium.

Due to the structure of those DVDs (with some packages being in present in both) the SLES installer was asking the user to change the medium several times during the installation process. Ideally the installer should use all packages from the “Packages” medium without changing the media.

In addition, there is yet another requirement for preferring the packages from the installation DVD to the packages available via a remote repository. Downloading a package from the internet is usually much slower than the DVD and can be problematic in network connections with a download limit or with a price based on the bandwidth usage.

Now the installer properly adjust the priority of all the repositories to achieve the desired behavior. To avoid possible side effects we decided to change the repository priority only when more than one repository is used and all repositories are local (e.g. DVD, hard disk, USB flash disk…). That means in some less common cases (2 DVDs + a remote repository) you will still need to change the medium but this is a safer solution.

Add On products in AutoYaST

For those using SLE Add On products, we have improved the error message if an Add On Product cannot be added during an AutoYaST installation. The user can see now which wrongly configured Add On Product has produced the error.

This will be specially useful with the upcoming SLE15, in which the concepts of Add Ons and Modules will become more relevant than ever.

Fixed a crash when shutting down the YaST user interface

And now it’s time for the corresponding dose of technical insights for those who enjoy that part of our reports.

When UI::OpenDialog() and UI::CloseDialog() calls didn’t match when shutting down the UI (user interface YaST component), you’d get a segmentation fault with a core dump. Well, you did want to shut down YaST, but probably not like that. This is now fixed.

After tracking this down, it was surprisingly simple to reproduce: Just use the YaST version of the trivial “Hello, World” program and comment out the UI::CloseDialog() call.

This was a case of providing additional error reporting causing more problems than the original error: leaving dialogs open while terminating the program is an error, of course. But fixing this little problem by cleaning up the remaining dialogs lead to handling widgets after some of the underlying infrastructure (in this case the QApplication) was already destroyed, so all the QWidgets were also destroyed (because the QApplication takes care of that), but YaST’s generic UI layer was still unaware of that fact and tried to destroy them again.

This is now fixed by properly cleaning up the widget tree in YaST’s generic UI layer first which will also clean up the associated QWidgets so there is nothing left to clean up for the QApplication.

This might also fix a number of similar segfaults in other situations where the YaST Ruby engine would need to shut down because of other problems, e.g. when there is an unhandled Ruby exception.

Surprisingly enough, this must have been a very old (10+ years?) bug, but it never became quite obvious, or at least nobody was ever annoyed enough to try to track it down.

If you want even more details, check the conversation in the bug report.

More to come

The end of this sprint caught up with a lot of almost finished stuff. But following the Scrum principle of “nothing is done until it fits the Definition of Done”, we don’t blog about such stuff. Fortunately, that means the next report will likely be quite juicy. So, see you again in a couple of weeks!

Since we announced in the latest report, we decided to shorten our development sprints from three to only two weeks. As a natural consequence, this is the first report of a series of shorter ones. But shorter doesn’t have to mean less juicy! Keep reading and enjoy.

Displaying very long changelog

We got a bug report about YaST not responding when a very long package changelog was displaying in the package manager. It turned out that some packages have a huge change log history with several thousands entries (almost 5000 for the kernel-default package). That produces a very long table which takes long time to parse and display in the UI.

The solution is to limit the maximum number of displayed items in the UI. You cannot easily read that very long text anyway, for such a long text you would need some search functionality which the YaST UI does not provide.

Finding the limit, that magic number, was not easy as we want to be backward compatible and display as much as possible but still avoid that pathological cases with a huge list.

In the end we decided to go for a limit of 512 change entries. The vast majority of packages have way fewer entries, so you should not notice this clipping functionality. When the limit is reached a command to display the full log is displayed at the end so you can easily see the missing part when needed. (Hint: the widget supports the usual copy&paste functionality, you can copy (with Ctrl+C) the displayed command and paste it into a terminal window directly.)

Storage reimplementation: MD RAID support in the rewritten partitioner

In the previous report we told you we are rewriting the YaST partitioner to use the new storage stack and modern reusable CWM widgets under the hood while retaining exactly the same behavior and look&feel on the surface. We are reimplementing one feature at a time, and this sprint it was the MD RAID’s turn.

Now the partitioner displays current RAIDs, with details about them and the devices used to construct each RAID. If a picture is worth a thousand words, here you have 3,000 words:

No, you can’t have a separate partition for /var/lib/docker for CaaSP – so here you have it

Our brand new containers-oriented solutions, SUSE CaaSP and its openSUSE counterpart Kubic, need to have /var/lib/docker as a separate partition for several reasons. It was a separate Btrfs subvolume already, but that wasn't quite separate enough. 🙂

The problem was just that the automated storage proposal in the old YaST storage subsystem is not prepared for anything like this; it can deal with root, swap and optionally a separate home partition (or volume if using LVM). That's it. Extending the current syntax of the control.xml file to be able to specify arbitrary partitions and adapting the code of the old proposal to understand and handle this new specification was unfortunately almost impossible. Sadly, we have to admit the old code is hardly maintainable these days and not flexible enough to accommodate this kind of changes in a safe way.

This is one (out of many) reasons why we are currently in the process of rewriting that entire YaST storage subsystem, as you already know from many previous posts in this blog. But, as you also know, the new system will debut in SLES15 and openSUSE Leap 15.0, too late for the current SUSE CaaSP and openSUSE Kubic.

We decided we'd introduce a hack based in the old proposal, well knowing that hacks accumulated in code can develop their own life just like Dust Puppy. But we plan to kill that hack immediately once StorageNG arrives.

So the hack was to simply use the logic for the separate home partition and repurpose it, keeping all the respective parameters in control.xml and introducing yet another one called home_path where the actual mount point for that partition can be specified — in this case /var/lib/docker. The tradeoff is that there can be no separate /home anymore in parallel to that.

That "feature" should not be used outside of that very specific use case in CaaSP and we even considered the possibility of not documenting it too publicly to avoid misuse. But we are living the open source spirit and whatever we do, we do in public. Even if it's the (quite embarrassing) fact that we consider changes to certain parts of our own code too risky. But again, that's why we are pushing the storage layer reimplementation (Storage-NG) so hard: we want to regain control over that part.

Storage reimplementation: custom partitioning with AutoYaST

And talking about the new storage layer and our previous posts, you already know we are working hard to integrate it with AutoYaST. For the time being, custom partitioning layouts are supported, but only using plain partitions and LVM. Other features, like RAID or Btrfs subvolumes support, are still missing.

A nice thing about the new code is that it relies as much as possible on the new storage layer. On older versions, AutoYaST implemented some logic on its own and that caused some unexpected troubles. Fortunately, that is not the case anymore, and the new code looks way easier to extend and maintain.

YaST does not write directly in /etc/vconsole.conf anymore

When configuring the system keyboard, YaST used to write the keyboard map configuration directly in the /etc/vconsole.conf file. However, this approach is no longer appropriated since it may cause undesired effects to other tools. Now YaST uses the Systemd tool localectl to set the keyboard map in /etc/vconsole.conf, instead of writing in it directly. Another step to make YaST a good citizen of the Systemd world.

Storage reimplementation: named RAID arrays

If you use MD RAID arrays you probably know there are two ways of identifying them – by number or assigning a meaningful name to them. Named arrays use device names like /dev/md/<name> instead of /dev/md<number>.

During this sprint, we taught the new libstorage how to create and manage named RAID array, in addition to the already supported numbered ones. If the user decides so, the MD RAID name is used in fstab instead of the UUID (which was the only option with the old libstorage). We also made sure to improve the behavior in several scenarios, so we consider some bugs of the old storage to be fixed (or obsoleted, if you prefer) with storage-ng.

Linux also supports names of the form /dev/md_<name>. The new library is also able to handle this format, but the feature is intentionally disabled and documented to be unsupported because other parts of the system could not be 100% verified to work in that scenario. And we take quality assurance very seriously before labeling a feature as “supported”.

If you are not happy enough with the screenshots showing the RAID support in the partitioner, here you have more pictures. But, as usual with stuff implemented in the library, here “pictures” doesn’t mean screenshots, but fancy automatically generated graphs.

Stay tuned

Of course there are a lot of other things we did during the last two weeks, although some of them were considered not interesting enough for this report or were not finished on time. We are already working in finishing the unfinished stuff and implementing new exciting improvements. And the bright side is that you only have to wait two weeks to know more… so stay tuned.