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:
lava-test-case
lava-test-case-attach
lava-test-run-attach
lava-background-process-start
lava-background-process-stop
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
Caution
lava-test-case-attach
is retained in the pipeline
dispatcher (V2) but the effect of the script needs consideration by
the test writer. See Handling test attachments.
This attaches a file to a test result with a particular ID, for example:
steps:
- "echo content > file.txt"
- "lava-test-case test-attach --result pass"
- "lava-test-case-attach test-attach file.txt text/plain"
The arguments are:
- The test case id
- The file to attach
- (optional) The MIME type of the file (if no MIME type is passed, a guess is made based on the filename)
Caution
lava-test-run-attach
is retained in the pipeline
dispatcher (V2) but the effect of the script needs consideration by
the test writer. See Handling test attachments.
This attaches a file to the overall test run that lava-test-shell is currently executing, for example:
steps:
- "echo content > file.txt"
- "lava-test-run-attach file.txt text/plain"
The arguments are:
- The file to attach
- (optional) The MIME type of the file (if no MIME type is passed, a guess is made based on the filename)
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:
- The name that is used to identify the process later in lava-background-process-stop
- The command line for the process to be run in the background
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:
- The name that was specified in lava-background-process-start
- (optional) An indication that you want to attach file(s) to the test run with specified mime type. See Handling test attachments.
The V1 dispatcher support for test attachments depends on the deprecated bundle and bundle stream support. The scripts available in lava-test shell do not actually attach the requested files, just copy the files to a hard-coded directory where the bundle processing code expects to find data to put into the bundle. This relies on the device being booted into an environment with a working network connection - what was called the master image.
In the V2 pipeline dispatcher, master images and bundles have been removed. This puts the handling of attachments into the control of the test writer. An equivalent method would be to simply add another deploy and boot action to get the test device into an environment where the network connection is known to work, however the eventual location of the file needs to be managed by the test writer. An alternative method for text based data is simply to output the contents into the log file.
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.linaro.org/lava/lava-dispatcher.git
skip_by_default: False
- url: https://git.linaro.org/lava/lava-dispatcher.git
destination: lava-d-r
branch: release
- url: https://git.linaro.org/lava/lava-dispatcher.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
If your test depends on other tests to be executed before you run the current test, the following definition will help:
test-case-deps:
- git-repo: git://git.linaro.org/qa/test-definitions.git
testdef: common/passfail.yaml
- bzr-repo: lp:~stylesen/lava-dispatcher/sampletestdefs-bzr
testdef: testdef.yaml
- url: https://people.linaro.org/~senthil.kumaran/deps_sample.yaml
The test cases specified within the ‘test-case-deps’ section will be fetched from the given repositories/URLs and then executed in the same specified order. The valid possible repository or URL source keys that may be specified inside the ‘test-case-deps’ section are:
1. git-repo
2. bzr-repo
3. tar-repo
4. url
Caution
lava-test-shell does not take care of circular dependencies within test definitions.
As an example, if testA.yaml
lists a dependency on testB.yaml
in its test-case-deps
section then that will cause testB.yaml
to be loaded and run first. However, if testB.yaml
also points
to testA.yaml
in its test-case-deps
section, that will cause
testA.yaml
to be loaded and run. This is an obvious circular
dependency; real loops may be much more subtle, running through
multiple test definitions in a complex setup with many defined
dependencies. Be careful to avoid this! The log for a case like this
would show many attempts at loading test definition...
until the
job is failed due to timeout.