The lava_test_shell
action provides a way to employ a black-box style
testing appoach with the target device. The test definition format is designed
to be flexible, allowing many options on how to do things.
A minimal test definition looks like this:
metadata:
name: passfail
format: "Lava-Test-Shell Test Definition 1.0"
description: "A simple passfail test for demo."
os:
- ubuntu
- openembedded
devices:
- origen
- panda
environment:
- lava-test-shell
params:
TEST_1: pass
run:
steps:
- echo "test-1: $TEST_1"
- echo "test-2: fail"
parse:
pattern: "(?P<test_case_id>.*-*):\\s+(?P<result>(pass|fail))"
Note
The parse pattern has similar quoting rules as Python, so \s must be escaped as \\s and similar.
Some of the parameters here (os, devices, environment) are optional in the metadata section. Others are mandatory (name, format, description).
If your test definition is not part of a git or bzr repository then it is must include a version parameter in the metadata section like in the following example.
metadata:
name: passfail
format: "Lava-Test-Shell Test Definition 1.0"
version: "1.0"
description: "A simple passfail test for demo."
os:
- ubuntu
- openembedded
devices:
- origen
- panda
environment:
- lava-test-shell
A lava-test-shell is run by:
building the test definition into a shell script.
Note
This shell script will include set -e
, so a failing step will
abort the entire test run. If you need to specify a step that might fail,
finish the command with || true
to make that failure not abort
the test run.
copying this script onto the device and arranging for it to be run when the device boots
booting the device and letting the test run
retrieving the output from the device and turning it into a test result
run subsequent test definitions, if any.
For the majority of cases, the above approach is the easiest thing to do: write shell code that outputs “test-case-id: result” for each test case you are interested in. See the Test Developer Guide:
A possible advantage of the parsing approach is that it means your test is easy to work on independently from LAVA: simply write a script that produces the right sort of output, and then provide a very small amount of glue to wire it up in LAVA. However, using the parsing option will mean writing potentially complicated regular expressions.
When you need it, there is also a more powerful, LAVA-specific, way of writing
tests. When a test runs, $PATH
is arranged so that some LAVA-specific
utilities are available:
See also
lava-test-case records the results of a single test case. For example:
steps:
- "lava-test-case simpletestcase --result pass"
- "lava-test-case fail-test --shell false"
It has two forms. One takes arguments to describe the outcome of the test case. The other takes the shell command to run, and the exit code of this shell command is used to produce the test result.
Both forms take the name of the testcase as the first argument.
The first form takes these additional arguments:
--result $RESULT
: $RESULT should be one of pass/fail/skip/unknown--measurement $MEASUREMENT
: A numerical measurement associated with the test result--units $UNITS
: The units of $MEASUREMENT--result
must always be specified. For example:
run:
steps:
- "lava-test-case simpletestcase --result pass"
- "lava-test-case bottle-count --result pass --measurement 99 --units bottles"
If --measurement
is used, --units
must also be specified, even
if the unit is just a count.
The most useful way to produce output for lava-test-case result
is
Writing custom scripts to support tests which allow preparation of LAVA results from other
sources, complete with measurements. This involves calling lava-test-case
from scripts executed by the YAML file:
#!/usr/bin/env python
from subprocess import call
def test_case():
"""
Calculate something based on a test
and return the data
"""
return {"name": "test-rate", "result": "pass",
"units": "Mb/s", "measurement": 4.23}
def main():
data = test_case()
call(
['lava-test-case',
data['name'],
'--result', data['result'],
'--measurement', data['measurement'],
'--units', data['units']])
return 0
if __name__ == '__main__':
main()
The custom scripts themselves can be called from a lava-test-case
using the
--shell
command to test whether failures from the tests caused a subsequent
failure in the custom script.
The second form of lava-test-case
is indicated by the --shell
argument, for example:
run:
steps:
- "lava-test-case fail-test --shell false"
- "lava-test-case pass-test --shell true"
The result of a shell
call will only be recorded as a pass or fail,
dependent on the exit code of the command. The output of the command can,
however, be parsed as a separate result if the command produces output suitable
for the parser in the YAML:
run:
steps:
- lava-test-case echo2 --shell echo "test2b:" "fail"
parse:
pattern: "(?P<test_case_id>.*-*):\\s+(?P<result>(pass|fail))"
This example generates two test results to indicate that the shell command executed correctly but that the result of that execution was a failure:
#. **echo2** - pass
#. **test2b** - fail
This starts a process in the background, for example:
steps:
- lava-background-process-start MEM --cmd "free -m | grep Mem | awk '{print $3}' >> /tmp/memusage"
- lava-background-process-start CPU --cmd "grep 'cpu ' /proc/stat"
- uname -a
- lava-background-process-stop CPU
- lava-background-process-stop MEM --attach /tmp/memusage text/plain --attach /proc/meminfo application/octet-stream
The arguments are:
This stops a process previously started in the background using lava-background-process-start. The user can attach files to the test run if there is a need.
For example:
steps:
- lava-background-process-start MEM --cmd "free -m | grep Mem | awk '{print $3}' >> /tmp/memusage"
- lava-background-process-start CPU --cmd "grep 'cpu ' /proc/stat"
- uname -a
- lava-background-process-stop CPU
- lava-background-process-stop MEM --attach /tmp/memusage text/plain --attach /proc/meminfo application/octet-stream
The arguments are:
Handling of attachments is in the control of the test writer. A separate publishing location can be configured or text based data is simply to output the contents into the log file.
See also
If your test requires some packages to be installed before its run it can
express that in the install
section with:
install:
deps:
- linux-libc-dev
- build-essential
If your test needs code from a shared repository, the action can clone this data on your behalf with:
install:
bzr-repos:
- lp:lava-test
git-repos:
- git://git.linaro.org/people/davelong/lt_ti_lava.git
run:
steps:
- cd lt_ti_lava
- echo "now in the git cloned directory"
There are several options for customising git repository handling in the git-repos action, for example:
install:
git-repos:
- url: https://git.lavasoftware.org/lava/lava.git
skip_by_default: False
- url: https://git.lavasoftware.org/lava/lava.git
destination: lava-d-r
branch: release
- url: https://git.lavasoftware.org/lava/lava.git
destination: lava-d-s
branch: staging
Parameters used in the test definition YAML can be controlled from the YAML job file. See the following YAML test definition to get an understanding of how it works.
Before the test shell code is executed, it will optionally do some install work if needed. For example if you needed to build some code from a git repo you could do:
install:
git-repos:
- git://git.linaro.org/people/davelong/lt_ti_lava.git
steps:
- cd lt_ti_lava
- make
Note
The repo steps are done in the dispatcher itself. The install steps are run directly on the target.
Warning
Parse patterns and fixup dictionaries are confusing and hard to
debug. The syntax is Python and the support remains for compatibility with
existing Lava Test Shell Definitions. With LAVA V2, it is recommended to
move parsing into a custom script contained within
the test definition repository. The script can simply call
lava-test-case
directly with the relevant options once the data is
parsed. This has the advantage that the log output from LAVA can be tested
directly as input for the script.
You may need to incorporate an existing test that doesn’t output results in in
the required pass
/fail
/skip
/unknown
format required by LAVA.
The parse section has a fixup mechanism that can help:
parse:
pattern: "(?P<test_case_id>.*-*)\\s+:\\s+(?P<result>(PASS|FAIL))"
fixupdict:
PASS: pass
FAIL: fail
Note
Pattern can be double-quoted or single quoted. If it’s double-quoted, special characters need to be escaped. Otherwise, no escaping is necessary.
Single quote example:
parse:
pattern: '(?P<test_case_id>.*-*)\s+:\s+(?P<result>(PASS|FAIL))'
fixupdict:
PASS: pass
FAIL: fail