Created 4 years ago
Maintained by nkondras
A mirror of "Sonic Kernel Testing" tool  |
Members 3

skt - sonic kernel testing

Travis CI Build Status Test Coverage Status

Skt is a tool for automatically fetching, building, and testing kernel patches published on Patchwork instances.


Install dependencies needed for running skt like this:

sudo dnf install -y python2 python2-junit_xml beaker-client

Dependencies needed to build kernels:

sudo dnf install -y bison flex dnf-plugins-core
sudo dnf builddep -y kernel-`uname -r`

Extra dependencies needed for running the testsuite:

sudo dnf install -y python2-mock PyYAML

Run tests

To run all tests execute:

python -m unittest discover tests

To run some specific tests, you can execute a specific test like this:

python -m unittest tests.test_publisher


Install skt directly from git:

pip install git+

If support for beaker is required, install skt with the beaker extras:

pip install git+[beaker]

Test the skt executable by printing the help text:

skt -h


The skt tool implements several "commands", and each of those accepts its own command-line options and arguments. However there are also several "global" command-line options, shared by all the commands. To get a summary of the global options and available commands, run skt --help. To get a summary of particular command's options and arguments, run skt <COMMAND> --help, where <COMMAND> would be the command of interest.

Most of command-line options can also be read by skt from its configuration file, which is ~/.sktrc by default, but can also be specified using the global --rc command-line option. However, there are some command-line options which cannot be stored in the configuration file, and there are some options read from the configuration file by some skt commands, which cannot be passed via the command line. Some of the latter are required for operation.

Most skt commands can write their state to the configuration file as they work, so that the other commands can take the workflow task over from them. Some commands can receive that state from the command line, via options, but some require some information stored in the configuration file. For this reason, to support a complete workflow, it is necessary to always make the commands transfer their state via the configuration file. That can be done by passing the global --state option with every command.

To separate the actual configuration from the specific workflow's state, and to prevent separate tasks from interfering with each other, you can store your configuration in a separate (e.g. read-only) file, copy it to a new file each time you want to do something, then discard the file after the task is complete. Note that reusing a configuration file with state added can break some commands in unexpected ways. That includes repeating a previous command after the next command in the workflow has already ran.

The following commands are supported by skt:

  • merge
    • Fetch a kernel repository, checkout particular references, and optionally apply patches from patchwork instances.
  • build
    • Build the kernel with specified configuration and put it into a tarball. This command expects merge command to have completed succesfully.
  • publish
    • Publish (copy) the kernel tarball, configuration, and build information to the specified location, generating their resulting URLs, using the specified "publisher". Only "cp", "scp", and "sftp" publishers are supported at the moment. This command expects build command to have completed succesfully.
  • run
    • Run tests on a built kernel using the specified "runner". Only "Beaker" runner is currently supported. This command expects publish command to have completed succesfully.
  • report
    • Report build and/or test results using the specified "reporter". Currently results can be reported by e-mail or printed to stdout. This command expects run command to have completed.
  • console-check
    • Check the specified console logs for any traces, and report the first trace found for each console log.
  • cleanup
    • Remove the build information file, kernel tarball. Remove state information from the configuration file, if saving state was enabled with the global --state option, and remove the whole working directory, if the global --wipe option was specified.
  • all
    • Run the following commands in order: merge, build, publish, run, report (if --wait option was specified), and cleanup.

The following is a walk through the process of checking out a kernel commit, applying a patch from Patchwork, building the kernel, running the tests, reporting the results, and cleaning up.

All the following commands use the -vv option to increase verbosity of the command's output, so it's easier to debug problems. Remove the option for quieter, shorter output.

You can make skt output junit-compatible results by adding a --junit <JUNIT_DIR> option to any of the following commands. The results will be written to the <JUNIT_DIR> directory.


To checkout a kernel tree run:

skt --rc <SKTRC> --state --workdir <WORKDIR> --output-dir <OUTDIR> -vv \
    merge --baserepo <REPO_URL> --ref <REPO_REF>

Here <SKTRC> would be the configuration file to retrieve the configuration and the state from, and store the updated state in. <WORKDIR> would be the directory to clone and checkout the kernel repo to, <OUTDIR> would be directory to store output files, <REPO_URL> would be the source kernel Git repo URL, and <REPO_REF> would be the reference to checkout.

E.g. to checkout "master" branch of the "net-next" repo:

skt --rc skt-rc --state --workdir skt-workdir --output-dir skt-outdir -vv \
    merge --baserepo git:// \
          --ref master

To apply a patch from Patchwork run:

skt --rc <SKTRC> --state --workdir <WORKDIR> -vv \
    merge --baserepo <REPO_URL> \
          --ref <REPO_REF> \
          --pw <PATCHWORK_PATCH_URL>

Here, <REPO_REF> would be the reference to checkout, and to apply the patch on top of, and <PATCHWORK_PATCH_URL> would be the URL pointing to a patch on a Patchwork instance.

E.g. to apply a particular patch to a particular, known-good commit from the "net-next" repo, run:

skt --rc skt-rc --state --workdir skt-workdir -vv \
    merge --baserepo git:// \
          --ref a870a02cc963de35452bbed932560ed69725c4f2 \

To apply a local patch run:

skt --rc <SKTRC> --state --workdir <WORKDIR> -vv \
    merge --baserepo <REPO_URL> \
          --ref <REPO_REF> \
          --patch <PATH_TO_PATCH>

Here, <PATH_TO_PATCH> would be the patch file. E.g. to apply a particular patch to a particular, known-good commit from the "net-next" repo, run:

skt --rc skt-rc --state --workdir skt-workdir -vv \
    merge --baserepo git:// \
          --ref a870a02cc963de35452bbed932560ed69725c4f2 \
          --patch net-next-cxgb4-notify-fatal-error-to-uld-drivers.patch

Faster clones

In some instances, a full git history is not needed. Shallow clones are git repositories that have a truncated history and they can be cloned much faster than a full copy of the repository. Use the --fetch-depth option with skt merge:

skt ... merge ... --fetch-depth 5

In the example above, the repository will be cloned with a git history of five commits. This speeds up the cloning process, but it also prevents the use of any git references (tags, branches, etc) that were made before the last five commits.


And to build the kernel run:

skt --rc <SKTRC> --state --workdir <WORKDIR> -vv \
    build -c `<CONFIG_FILE>`

Where <CONFIG_FILE> would be the kernel configuration file to build the kernel with. The configuration will be applied with make olddefconfig, by default.

E.g. to build with the current system's config file run:

skt --rc skt-rc --state --workdir skt-workdir -vv \
    build -c /boot/config-`uname -r`

NOTE: Kernels are built without debuginfo by default to save disk space and improve build times. In some cases, deep troubleshooting may require debug symbols. Use the --enable-debuginfo argument to build a kernel with debug symbols included.

Provide additional arguments and options to make by using --makeopts.

Kernel configuration file options

Three kernel configuration file options are supported by skt:

  • Provide a kernel configuration file directly: --config CONFIG_FILE_PATH
  • Use a minimal configuration for a very small kernel: --cfgtype tinyconfig
  • Build kernel configuration files for Red Hat kernels: --cfgtype rh-configs

Users must specify a filename glob using --rh-configs-glob with --cfgtype rh-configs. This allows skt to choose the correct kernel configuration file. The path should be relative to the path of the kernel source.

The following example works for x86_64:

--rh-configs-glob "redhat/configs/kernel-*-x86_64.config"

The following example works for POWER systems (little endian):

--rh-configs-glob "redhat/configs/kernel-*-ppc64le.config"


To "publish" the resulting build using the simple "cp" (copy) publisher run:

skt --rc <SKTRC> --state --workdir <WORKDIR> -vv \
    publish -p cp <DIRECTORY> <URL_PREFIX>

Here <DIRECTORY> would be the location for the copied build artifacts, and URL_PREFIX would be the string to add to prepend the filenames with (together with a slash /) to construct the URLs the files will be reachable at. The resulting URLs will be passed to other commands, such as run, via the saved state in the configuration file.

E.g. to publish to the /srv/builds directory available at http://skt-server run:

skt --rc skt-rc --state --workdir skt-workdir -vv \
    publish -p cp /srv/builds http://skt-server


To run the tests you will need access to a Beaker instance configured to the point where bkr whoami completes successfully. You will also need a template file for generating a Beaker job XML, which runs the tests. The template file can contain the following placeholder strings replaced by skt before submitting the job XML to Beaker:

  • ##KVER##
    • The kernel release version output by make -s kernelrelease.
  • ##KPKG_URL##
    • The URL of the kernel tarball, generated and published to with publish.

Below is an example of a superficial template. Note that it won't work as is.

  <whiteboard>skt ##KVER##</whiteboard>
    <recipe whiteboard="##KVER##">
          <distro_family op="=" value="Fedora26"/>
          <distro_tag op="=" value="RELEASED"/>
          <distro_variant op="=" value="Server"/>
          <distro_arch op="=" value="x86_64"/>
          <arch op="=" value="x86_64"/>
      <task name="/distribution/install" role="STANDALONE"/>
      <task name="/distribution/kpkginstall" role="STANDALONE">
          <param name="KPKG_URL" value="##KPKG_URL##"/>
          <param name="KVER" value="##KVER##"/>

Provided you have both Beaker access and a suitable job XML template, you can run the tests with the built kernel as such:

skt --rc <SKTRC> --state --workdir <WORKDIR> -vv run \
    --runner beaker '{"jobtemplate": "<JOBTEMPLATE>"}' \

Here, <JOBTEMPLATE> would be the name of the file with the Beaker job XML template. If you remove the --wait option, the command will return once the job was submitted. Otherwise it will wait for its completion and report the result.

In case running on specific hosts is not desired, one can use a simple text file containing one hostname per line, and pass the file via blacklist parameter. Tests will not attempt to run on machines which names are specified in the file. This is useful for example as a temporary fix in case the hardware is buggy and the maintainer of the pool doesn't have time to exclude it from the pool.

E.g. to run the tests from a job XML template named beakerjob.xml and exclude machines in blacklist.txt file execute:

skt --rc skt-rc --state --workdir skt-workdir -vv run \
    --runner beaker '{"jobtemplate": "beakerjob.xml", \
                  "blacklist": "blacklist.txt"}, \


There are two "reporters" supported at the moment: "stdio" and "mail". The former prints the report on stdout and the latter sends it by mail to specified addresses, with specified "From" address.

This command requires the runner parameters from the "run" command to be present in the configuration file. It needs this minimum "runner" section:

type = beaker
jobtemplate = <JOBTEMPLATE>

Here, <JOBTEMPLATE> is the same Beaker job template file name you used for the "run" command. E.g., continuing from the example above, it can be:

type = beaker
jobtemplate = beakerjob.xml

stdio Reporter

The stdio reporter prints the report to stdout and requires no additional options:

skt --rc <SKTRC> --state --workdir <WORKDIR> -vv \
    report --reporter stdio

mail Reporter

The mail reporter sends an email to one or more recipients with the results of the testing. Multiple options are available:

  • --mail-to: one or more recipient email addresses (required)
  • --mail-from: the email address of the sender (required)
  • --mail-subject: the email subject (optional)
  • --mail-header: one or more email headers to add to the email (optional)

The most basic email report can be sent using these arguments:

skt --rc skt-rc --state --workdir skt-workdir -vv \
    report --reporter mail \
    --mail-to \

Multiple recipient addresses can be provided:

skt --rc skt-rc --state --workdir skt-workdir -vv \
    report --reporter mail \
    --mail-to --mail-to \

Here is a full example with multiple email headers:

skt --rc skt-rc --state --workdir skt-workdir -vv \
    report --reporter mail \
    --mail-to \
    --mail-from \
    --mail-subject "Important tests results" \
    --mail-header "X-Build-ID: 225" \
    --mail-header "In-Reply-To: <>"

The reporter command is able to send a single report for multiple test runs. For now, only runs with same patch series and base are supported. This is especially useful if the same patch series are being tested on multiple architectures. To run reporter in multireport mode, add the --result option with an skt state file created by the run command you wish to report. The --result option can be repeated multiple times for multiple state files.

Using the previous example of mail report command and state files ./state_x86 and ./state_s390x, the command will be:

skt --rc skt-rc --state --workdir skt-workdir -vv \
    report --reporter mail \
    --mail-to --mail-to \
    --mail-from \
    --result ./state_x86 --result ./state_s390x

Console check

The console checker is not a part of the default flow, but allows parsing console logs to detect any call traces. A kernel release string is needed (can be grabbed from state file), and URL or file path to one or more console logs can be specified. Local files can be gzipped.

Example of the command to check console log present at given URL, when the kernel release string is parsed out of state file:

skt --rc skt-rc --state --workdir skt-workdir -vv \
    console-check --console


The cleanup command doesn't have its own options, but recognizes the global --state and --wipe options. It will remove the state section from the configuration file, if --state is specified, and it will remove the working directory, if --wipe is specified. Otherwise it will just remove the built tarball and the build information file.

Developer Guide

Developers can test changes to skt by using "development mode" from python's setuptools package. First, cd to the directory where skt is cloned and run:

pip install --user -e .

This installs skt in a mode where any changes within the repo are immediately available simply by running skt. There is no need to repeatedly run pip install . after each change.

Using a virtual environment is highly recommended. This keeps skt and all its dependencies in a separate Python environment. Developers can build a virtual environment for skt quickly:

virtualenv ~/skt-venv/
source ~/skt-venv/bin/activate
pip install -e .

To deactivate the virtual environment, simply run deactivate.


skt is distributed under GPLv2 license.

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 2 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see