GUADEC 2022 has been happening in person for the first time in two years, in Guadalajara. Twenty of us in Europe met up in Berlin for a mini-GUADEC, to attend the main conference remotely. There have been several talks given from here using the nice A/V setup in c-base, who are hosting us.
I gave my talk this afternoon, on the threading rework which is ongoing in gnome-software. The slides are here, the notes are here (source is here), and the recording should be available soon on the GUADEC YouTube channel.
As part of the question and answer session afterwards, it was suggested that it might be helpful to write a blog post about strategies for making async code in C more readable. I’ll try and write something about that soon.
Thanks to the GUADEC organising team for hosting the conference and integrating remote participation so well, to Sonny and Tobias for organising the Berlin mini-GUADEC, and to the c-base technical people for setting things up nicely here.
The appstream specification, used for appdata files for apps on Linux, supports specifying what input devices and display sizes an app requires or supports. GNOME Software 41 will hopefully be able to use that information to show whether an app supports your computer. Currently, though, almost no apps include this metadata in their appdata.xml file.
Please consider taking 5 minutes to add the information to the appdata.xml files you care about. Thanks!
If your app supports (and is tested with) touch devices, plus keyboard and mouse, add:
There have been a few instances recently where people have pointed out that GNOME Software marks some apps as not free software when they are. This is a bug in the appdata files provided by those applications, which typically includes something like
<project_license>GPL-3.0+ and CC-BY-SA-3.0</project_license>
This is generally an attempt to list the license of the code and of the documentation. However, the resulting SPDX expression means to apply the most restrictive interpretation of both licenses. As a result, the expression as a whole is considered not free software (CC-BY-SA-3.0 is not a free software license as per the FSF or OSI lists).
Instead, those apps should probably just list the ‘main’ license for the project:
and document the license for their documentation separately. As far as I know, the appdata format doesn’t currently have a way of listing the documentation license in a machine readable way.
If you maintain an app, or want to help out, please check the licensing is correctly listed in your app’s appdata.
tl;dr: Use callgrind to profile CPU-heavy workloads. In some cases, movingheap allocations to the stack helps a lot. GNOME Software startup time has decreased from 25 seconds to 12 seconds (-52%) over the GNOME 40 cycle.
To wrap up the sporadic blog series on the progress made with GNOME Software for GNOME 40, I’d like to look at some further startup time profiling which has happened this cycle.
This profiling has focused on libxmlb, which gnome-software uses extensively to query the appstream data which provides all the information about apps which it shows in the UI. The basic idea behind libxmlb is that it pre-compiles a ‘silo’ of information about an XML file, which is cached until the XML file next changes. The silo format encodes the tree structure of the XML, deduplicating strings and allowing fast traversal without string comparisons or parsing. It is memory mappable, so can be loaded quickly and shared (read-only) between multiple processes. It allows XPath queries to be run against the XML tree, and returns the results.
gnome-software executes a lot of XML queries on startup, as it loads all the information needed to show many apps to the user. It may be possible to eliminate some of these queries – and some earlierwork did reduce the number by binding query parameters at runtime to pre-prepared queries – but it seems unlikely that we’ll be able to significantly reduce their number further, so better speed them up instead.
Profiling work which happens on a CPU
The work done in executing an XPath query in libxmlb is largely on the CPU — there isn’t much I/O to do as the compiled XML file is only around 7MB in size (see ~/.cache/gnome-software/appstream), so this time the most appropriate tool to profile it is callgrind. I ruled out using callgrind previously for profiling the startup time of gnome-software because it produces too much data, risks hiding the bigger picture of which parts of application startup were taking the most time, and doesn’t show time spent on I/O. However, when looking at a specific part of startup (XML queries) which are largely CPU-bound, callgrind is ideal.
It takes about 10 minutes for gnome-software to start up and finish loading the main window when running under callgrind, but eventually it’s shown, the process can be interrupted, and the callgrind log loaded in kcachegrind:
Here I’ve selected the main() function and the callee map for it, which shows a 2D map of all the functions called beneath main(), with the area of each function proportional to the cumulative time spent in that function.
The big yellow boxes are all memset(), which is being called on heap-allocated memory to set it to zero before use. That’s a low hanging fruit to optimise.
In particular, it turns out that the XbStack and XbOperands which libxmlb creates for evaluating each XPath query were being allocated on the heap. With a few changes, they can be allocated on the stack instead, and don’t need to be zero-filled when set up, which saves a lot of time — stack allocation is a simple increment of the stack pointer, whereas heap allocation can involve page mapping, locking, and updates to various metadata structures.
The changes are here, and should benefit every user of libxmlb without further action needed on their part. With those changes in place, the callgrind callee map is a lot less dominated by one function:
There’s still plenty left to go at, though. Contributions are welcome, and we can help you through the process if you’re new to it.
What’s this mean for gnome-software in GNOME 40?
Overall, after all the performance work in the GNOME 40 cycle, startup time has decreased from 25 seconds to 12 seconds (-52%) when starting for the first time since the silo changed. This is the situation in which gnome-software normally starts, as it sits as a background process after that, and the silo is likely to change every day or two.
Aside from performance work, there’s a number of other improvements to gnome-software in GNOME 40, including a new icon, some improvements to parts of the interface, and a lot of bug fixes. But perhaps they should be explored in a separate blog post.
Many thanks to my fellow gnome-software developers – Milan, Phaedrus and Richard – for their efforts this cycle, and my employer the Endless OS Foundation for prioritising working on this.
Following on from the heap profiling I did on gnome-software to try and speed it up for Endless, the next step was to try profiling the computation done when starting up gnome-software — which bits of code are taking time to run?
tl;dr: There is new tooling in sysprof and GLib from git which makes profiling the performance of high-level tasks simpler. Some fixes have landed in gnome-software as a result.
Approaches which don’t work
The two traditional tools for this – callgrind, and print statements – aren’t entirely suitable for gnome-software.
I tried running valgrind --tool=callgrind gnome-software, and then viewing the results in KCachegrind, but it slowed gnome-software down so much that it was unusable, and the test/retry cycle of building and testing changes would have been soul destroyingly slow.
callgrind works by simulating the CPU’s cache and looking at cache reads/writes/hits/misses, and then attributing costs for those back up the call graph. This makes it really good at looking at the cost of a certain function, or the total cost of all the calls to a utility function; but it’s not good at attributing the costs of higher-level dynamic tasks. gnome-software uses a lot of tasks like this (GsPluginJob), where the task to be executed is decided at runtime with some function arguments, rather than at compile time by the function name/call. For example “get all the software categories” or “look up and refine the details of these three GsApp instances”.
That said, it was possible to find and fix a couple of bits of low-hanging optimisation fruit using callgrind.
Print statements are the traditional approach to profiling higher-level dynamic tasks: print one line at the start of a high-level task with the task details and a timestamp, and print another line at the end with another timestamp. The problem comes from the fact that gnome-software runs so many high-level tasks (there are a lot of apps to query, categorise, and display, using tens of plugins) that reading the output is quite hard. And it’s even harder to compare the timings and output between two runs to see if a code change is effective.
Having looked at sysprof briefly for the heap profiling work, and discounted it, I thought it might make sense to come back to it for this speed profiling work. Christian had mentioned at GUADEC in Thessaloniki that the design of sysprof means apps and libraries can send their own profiling events down a socket, and those events will end up in the sysprof capture.
So I played around with this newly-instrumented version of gnome-software for a bit, but found that there were still empty regions in the profiling trace, where time passed and computation was happening, but nothing useful was logged in the sysprof capture. More instrumentation was needed.
sysprof + GLib
gnome-software does a lot of its computation in threads, bringing the results back into the main thread to be rendered in the UI using idle callbacks.
For example, the task to list the apps in a particular category in gnome-software will run in a thread, and then schedule an idle callback in the main thread with the list of apps. The idle callback will then iterate over those apps and add them to (for example) a GtkFlowBox to be displayed.
Adding items to a GtkFlowBox takes some time, and if there are a couple of hundred of apps to be added in a single idle callback, that can take several hundred milliseconds — a long enough time to block the main UI from being redrawn that the user will notice.
How do you find out which idle callback is taking too long? sysprof again! I added sysprof support to GLib so that GSource.dispatch events are logged (along with a few others), and now the long-running idle callbacks are displayed in the sysprof graphs. Thanks to Christian and Richard for their reviews and contributions to those changes.
This capture file was generated using sysprof-cli --gtk --use-trace-fd -- gnome-software, and the ‘gnome-software’ and ‘GLib’ lines in the ‘Timings’ row need to be made visible using the drop-down menu in the ‘Timings’ row.
The last week has been a fun process of starting to profile gnome-software with the aim of lowering its resource consumption and improving its startup speed. gnome-software is an important part of the desktop, so having it work speedily, especially on resource constrained computers, is important. This work is important for Endless OS, and is happening upstream.
To start with, I’ve looked at gnome-software’s use of heap memory, particularly during startup. While allocating lots of memory on the heap isn’t always a bad thing (caches are a good example of heap allocations being used to speed up a program overall), it’s often a sign of unnecessary work being done. Large heap allocations do take a few tens of milliseconds to be mapped through the allocator too. To do this profiling, I’ve been using valgrind’s massif tool, and massif-visualizer to explore the heap allocations. I could also have used heaptrack, or gobject-list, but they’re tools to explore another time.
Profile your app
Before diving into the process of optimising, the summary is that this work dropped gnome-software’s pixbuf heap usage by 24MB, and its non-pixbuf heap usage after initialisation (i.e. at the point when the main window is visible and ready to use) by 12%, from 15.7MB to 13.7MB (on my set of flatpak repositories on Fedora 32). I’ve been doing this work upstream, and it’ll trickle down to the downstream copy of gnome-software in Endless OS.
There is more low-hanging fruit to explore, and plenty of opportunities to dive in and trim more memory usage from gnome-software, or other apps. If you’re interested, please dive in! Get in touch if you have questions, or post them on GNOME’s Discourse instance and tag me. I’ll be happy to help!
How to profile heap usage
Profiling heap usage using massif is an iterative process: run your program under massif, do some actions in the program, quit, then open the resulting massif.out.pid file in massif-visualizer and see where allocations are coming from. Pick an allocation which looks large or unnecessary, find it in the code, optimise the code (if possible), and then repeat the process.
When running it, I wait for gnome-software to finish loading its main window, then I exit; so all this profiling work is for allocations during startup.
I run massif using this script, which I’ve put in ~/.local/bin/massif:
All the --alloc-fn arguments hide internal GLib functions so that the output is a little easier to interpret directly. There currently isn’t a way to store them in a config file or suppression file.
Some typical output from massif-visualizer before any code improvements:
The window shows heap allocations against time in instructions executed. The breakdown of where each allocation came from is known in detail at key snapshots (which are expandable in the side pane), and the total heap usage is known in summary for the other snapshots, which allows the graph to be drawn. Allocations coming from different functions are coloured differently in the graph.
There are two sets of allocations to focus on: the red plateau between time 1e+10 and 1.6e+10, and the orange step from time 1e+10 onwards.
Selecting the red plateau shows the backtrace which led to its allocation in the side pane, and (despite some missing debug symbols, leading to the ‘???’ entries), it seems to have come from within libjpeg, as part of loading a JPEG pixbuf. gnome-software has various large JPEG images which are displayed in the featured app banners. It seems that libjpeg makes some big temporary allocations when loading a JPEG.
The orange step from 1e+10 onwards is another target. Looking at the backtraces, it seems it’s a series of similar allocations for pixbuf pixel storage for the featured app banners and for app icons. Some quick calculations show that each 1024×400 pixel banner will take around 6.5MB of memory to store its uncompressed pixels (at 16B per pixel).
From the graph and the backtraces, it seems that almost 100MB is used for pixbuf data for featured app banners. At 6.5MB per banner, that’s 15 banners, which seems reasonable. But actually gnome-software limits itself to 5 banners, so something’s amiss.
Style providers aren’t cheap
After adding some debug prints in GTK where it loads the pixbufs for CSS background properties, it became evident that the same few images were being loaded multiple times. CSS is used to style each featured app tile, including setting the background, since that allows a lot of artistic freedom quite easily. However, the CSS was being refreshed and set a few times for each tile, with a new GtkCssProvider each time. The old provider was staying in place, but with its properties overridden. This included the previously-loaded background image, which remained loaded but unused (essentially, leaked!). With that chased down, it was possible to fix the problem.
Back to profiling
With one issue investigated and fixed, the next step is to do another profiling run, find another target for reducing heap allocations, and repeat.
While we might have fixed one pixbuf bug in gnome-software, it does still use a lot of memory for pixbufs, since it displays a lot of high-resolution app icons. Those pixbuf allocations occupy a lot of space in the massif-visualizer view, and take up a large percentage of the ‘threshold’ of heap allocations which massif includes in its traces.
massif provides the --ignore-fn argument to allow certain allocations to be ignored, so that you can more easily profile others. So I did further profiling runs with a series of --ignore-fn arguments to ignore pixbuf allocations.
With the --ignore-fn arguments, and increasing the ‘Stacked diagrams’ level in the toolbar to show more individual areas on the graph, it’s now possible to see more detail on the largest non-pixbuf allocations, and hence easier to choose where to focus next.
From this screenshot, perhaps the next place to focus on would be GHashTable creation and insertions, since that totals around 1MB of the heap usage (once pixbufs are ignored).
I have iterated through the gnome-software massif profiles a few times, and have submitted various other fixes to gnome-software and libxmlb which are in the process of being reviewed and merged, but I won’t walk through each of them. There are still improvements to be made in future: gnome-software is quite complex!
In total, the changes reduced gnome-software’s heap usage at startup by 26MB, though the actual numbers will vary on other systems depending on how often feature tiles get refreshed, and how many apps and repositories you have configured.
These changes have not made a significant improvement to the startup time of gnome-software, which is more significantly influenced by network activity and file parsing (and the subject of some future work).
Hopefully this post gives a workable introduction to how to use massif on your own software. Please speak up if you have any questions. If you do profiling work on your software, please blog about it — it would be interesting to see what improvements are possible.