Beaker tasks used with skt runner
THIS PROJECT HAS MOVED TO https://gitlab.com/cki-project/kernel-tests/.
Here is a list of common prerequisites for all beaker tests. Test-specific dependencies and steps can be found in the README.md within each test's directory.
$ sudo wget -O /etc/yum.repos.d/beaker-client.repo https://beaker-project.org/yum/beaker-client-Fedora.repo
$ sudo wget -O /etc/yum.repos.d/beaker-harness.repo https://beaker-project.org/yum/beaker-harness-Fedora.repo
$ sudo dnf install -y beaker-client beakerlib restraint-rhts
Currently, all onboarded tests must use the following combinations of result/status fields:
- SKIP/COMPLETED if the test requirements aren't fulfilled (eg. test is running on incompatible architecture/hardware)
- PASS/COMPLETED if the test finished successfully
- WARN/ABORTED in case of infrastructure issues or other errors (eg. the test checks out a git repo and the git server is unavailable)
- WARN/COMPLETED or FAIL/COMPLETED in case of any test failures, based on how serious they are (left to decide by test authors)
See examples below to properly abort or skip in beaker:
if [ $? -ne 0 ]; then
rlLog "Aborting test because $reason"
rstrnt-abort --server $RSTRNT_RECIPE_URL/tasks/$TASKID/status
fi
if [ $? -ne 0 ]; then
rlLog "Aborting recipe because $reason"
rstrnt-abort recipe
fi
if [ $? -ne 0 ]; then
rlLog "Skipping test because $reason"
rstrnt-report-result $TEST SKIP
exit 0
fi
When onboarding a test, please check especially the point about infrastructure issues: a lot of tests simply report a warning if eg. external server can’t be reached and then continue. This kind of situation falls under infrastructure issues and the test must use the WARN/ABORTED combination, otherwise the infrastructure problem is reported to people as a bug in their code!
The order of Beaker tasks in the XML determines if the task is a preparation for the testing or a test. Anything after kpkginstall task is treated as a test and must follow the rules above. Anything except PASS before kpkginstall is treated as infrastructure issue. Machineinfo (to get HW specification) is ran before kpkginstall, so the same logic applies there. PANIC during kpkginstall means the kernel is bad (can't boot).