LAVA is written in Python, so you will need to know (or be willing to learn) the language. Likewise, the web interface is a Django application so you will need to use and debug Django if you need to modify the web interface. The pipeline model uses YAML (so you’ll need the YAML Parser) and Jinja2. All LAVA software is maintained in git, as are many of the support scripts, test definitions and job submissions. Some familiarity with Debian is going to be useful, helper scripts are available when preparing packages based on your modifications.
LAVA is complex and works to solve complex problems. This has implications for how LAVA is developed, tested, deployed and used.
Development occurs on the master
branch, testing of each release happens on
the staging
branch and releases are made from the release
branch. Git
tags are also based on the release
branch. All reviews need to be made
against the master
branch.
Note
It is worth checking if someone already has a review which relates to
your proposed changes. Check for open reviews on gerrit, e.g.
https://review.linaro.org/#/q/status:open+lava-dispatcher
The Django backend used with LAVA is PostgreSQL and some postgres-specific support is used. The LAVA UI has some use of Javascript and CSS. LAVA also uses ZMQ and XML-RPC and the LAVA documentation is written with RST.
In addition, test jobs and device support can involve use of U-Boot, GuestFS, ADB, QEMU, Grub, SSH and a variety of other systems and tools to access devices and debug test jobs.
If you are an administrator, you may think the previous link sent you to the wrong section. However, administrators need to understand how device-type templates operate and how the template engine will use the template to be able to make changes.
Important
Device type templates are more than configuration files - the
templates are processed as source code at runtime. Anyone making changes to
a device-type .jinja2
template file must understand the basics of
how to test templates using the same tools as developers.
Jinja2 provides a powerful templating engine. Templates in LAVA use several standard programming concepts:
In addition, LAVA templates need to always render to valid YAML. It is this
YAML which is sent to the worker as device.yaml
. The worker does not handle
the templates. All operations are done on the master.
The simplest check is to render the new template to YAML and check that it contains the expected commands. As with test job files, there are common YAML errors which can block the use of new templates.
See also
lava-server manage device-dictionary --hostname <HOSTNAME> --review
A more rigorous test is to use the dedicated unit test which does not
require lava-server
to be installed, i.e. it does not require a database to
be configured. This test can be run directly from a git checkout of
lava-server
with a few basic python packages installed (including
python-jinja2
).
$ python -m unittest -vcf lava_scheduler_app.tests.test_templates.TestTemplates.test_all_templates
Individual templates have their own unit tests to test for specific elements of the rendered device configuration.
Most changes to device-type templates take effect immediately - as soon as
the file is changed in /etc/lava-server/dispatcher-config/device-types/
the
next testjob for that device-type will use the output of that template. Always
test your templates locally before deploying the template to the master.
(Test jobs which have already started are not affected by template changes.)
This cannot be stressed enough. ALL admins need to keep device-type templates in some form of version control. The template files are code and admins will need to be able to upgrade templates when packages are upgraded and when devices need to implement new support.
As code, device-type templates need to develop alongside the rest of the codebase. The best way to maintain support is to Contributing Upstream so that new features can be tested against your templates and new releases can automatically include updates to your templates.
Some individual device files exist in the codebase in
lava_scheduler_app/tests/devices
but these are only for use in the
existing unit tests. There is no need to contribute individual device dictionaries
unless there are new unit tests which use those device dictionaries.
Note
LAVA is developed using Debian packaging to ensure that daemons and
system-wide configuration is correctly updated with changes in the codebase.
There is no support for pypi or python virtual environments or installing
directly from a git directory. python-setuptools
is used but only
with sdist
to create the tarballs to be used for the Debian packaging,
not for install
. Some dependencies of LAVA are not available with pypi,
for example python-guestfs
.
See also
Developing LAVA on Debian and a summary of the Debian LAVA team activity
Developers can update the installed code on their own systems manually (by copying files into the system paths) and/or use symlinks where appropriate but changes need to be tested in a system which is deployed using the Developer package build before being proposed for review. All changes must also pass all the unit tests, unless those tests are already allowed to be skipped using unittest decorators.
Mixing the use of python code in /usr/local/lib
and /usr/lib
on a
single system is known to cause spurious errors and will only waste your
development time. Be very careful when copying files and when using symlinks.
If in doubt, remove /usr/local/lib/python*
and ~/.local/lib/python*
then build a local developer package and install it.
If your change introduces a dependency on a new python module, always ensure
that this module is available in Debian by searching the Debian package lists. If the module
exists but is not in the current stable release of Debian, it can be
backported but be aware that this will delay testing and acceptance of your
change. It is expressly not acceptable to add a dependency on a python
module which is only available using pypi or pip install
. Introducing such
a module to Debian can involve a large amount of work - talk to us before spending time on code which relies on such modules or
which relies on newer versions of the modules than are currently available in
Debian testing.
See also
Certain terms used in LAVA V2 have specific meanings, please be consistent in the use of the following terms:
In lava-server
, a device is a database object in LAVA which stores
configuration, information and status relating to a single board. The device
information can be represented in export formats like YAML for use when the
database is not accessible.
In lava-dispatcher
, the database is not accessible so the scheduler
prepares a simple dictionary of values derived from the database and the
template to provide the information about the device.
A database object which collates similar devices into a group for purposes of scheduling. Devices of a single type are often the same vendor model but not all boards of the same model will necessarily be of the same device-type.
See also
lava-dispatcher
source package in
git and in Debian. The dispatcher software for LAVA V2 can be installed
without the server or the scheduler and a machine configured in this way is
also called a dispatcher.self.set_common_data
to set the namespace, key and value
and self.get_common_data
to retrieve the value using the namespace and
the key.common_data
primitives of the Action base class to copy parameters and
store the modified values as dynamic data.The name for the design of LAVA V2, based on how the actions to be executed by the dispatcher are arranged in a unidirectional pipe. The contents of the pipe are validated before the job starts and the description of all elements in the pipe is retained for later reference.
See also
lava-dispatcher
to interact with
external systems and daemons when a shell like environment is not supported.
Protocols need to be supported within the python codebase and currently
include multinode, LXC and vland.A singleton process which is solely responsible for assigning a device to a test job. The scheduler is common to LAVA V1 and LAVA V2 and performs checks on submission restrictions, device availability, device tags and schema compliance.
See also
lava-server
source package in git and
in Debian. It includes components from LAVA V1 and LAVA V2 covering the UI
and the scheduler daemon.LAVA online documentation is written with RST format. You can use the command below to generate html format files for LAVA V2:
$ cd lava-server/
$ make -C doc/v2 clean
$ make -C doc/v2 html
$ firefox doc/v2/_build/html/index.html
(or whatever browser you prefer)
We welcome contributions to improve the documentation. If you are considering adding new features to LAVA or changing current behaviour, ensure that the changes include updates for the documentation.
Wherever possible, all new sections of documentation should come with worked examples.
doc/v2/examples/test-jobs
doc/v2/examples/test-definitions
doc/v2/index.rst
file if you are adding new
pages or altering section headings.The ongoing migration complicates some of the workflow when it comes to finding all of the V2 code. When the V1 code is removed, the organisation of the code will be tidied up.
lava-server includes the lava_scheduler_app
, lava_results_app
,
lava_server
, lava
and linaro_django_xmlrpc
components of LAVA V2.
lava-dispatcher includes the lava_dispatcher
and lava_test_shell
components. All LAVA V2 dispatcher code lives in
lava_dispatcher/pipeline
. Some lava_test_shell
scripts remain in the
top level lava_test_shell
directory with overrides in
pipeline/lava_test_shell
.
See also
There are also locations which provide device configurations to support the unit tests. Only the Jinja2 support is used by the installed packages,
See also
The compatibility mechanism allows the dispatcher-master daemon to prevent issues that would arise if the worker is running older software. A job with a lower compatibility may fail much, much later but this allows the job to fail early. In future, support is to be added for re-queuing such jobs.
Developers need to take note that in the code, compatibility should reflect the
removal of support for particular elements, similar to handling a SONAME when
developing in C. When parts of the submission YAML are changed to no longer
support fields previously used, then the compatibility of the associated
strategy class must be raised to one more than the current highest
compatibility in the lava-dispatcher
codebase. Compatibility does not need
to be changed when adding new classes or functionality. It remains a task for
the admins to ensure that the code is updated when new functionality is to be
used on a worker as this typically involves adding devices and other hardware.
Compatibility is calculated for each pipeline during parsing. Only if the pipeline uses classes with the higher compatibility will the master prevent the test job from executing. Therefore, test jobs using code which has not had a compatibility change will continue to execute even if the worker is running older software. Compatibility is not a guarantee that all workers are running latest code, it exists to let jobs fail early when those specific jobs would attempt to execute a code path which has been removed in the updated code.
The Jinja2 templates can be found in lava_scheduler_app/tests/device-types
in the lava-server
codebase. The reason for this is that all template
changes are checked in the unit-tests. When the package is installed, the
device-types
directory is installed into
/etc/lava-server/dispatcher-config/device-types/
. The contents of
lava_scheduler_app/tests/devices
is ignored by the packaging, these files
exist solely to support the unit tests.
See also
Unit-tests and Testing the new design for examples of how to run individual unit tests or all unit tests within a class or module.
Individual instances will each have their own locations for the device
dictionaries of real devices. To allow the unit tests to run, some device
dictionaries are exported into lava_scheduler_app/tests/devices
but there
is no guarantee that any of these would work with any real devices, even of
the declared device-type.
For example, the Cambridge lab stores each device dictionary in git at
https://git.linaro.org/lava/lava-lab.git and you can look at the configuration
of staging
as a reference:
https://git.linaro.org/lava/lava-lab.git/tree/HEAD:/staging.validation.linaro.org/lava/pipeline/devices
The lava-dispatcher
codebase also has local device configuration files in
order to support the dispatcher unit tests. These are not Jinja2 format,
these are YAML - the same YAML as would be sent to the dispatcher by the
relevant master after rendering the Jinja2 templates on that master. There is
no guarantee that any of the device-type or device configurations in the
lava-dispatcher
codebase would work with any real devices, even of the
declared device-type.
The best way to protect your investment on LAVA is to contribute your changes back. This way you don’t have to maintain the changes you need by yourself, and you don’t run the risk of LAVA changed in a way that is incompatible with your changes.
Upstream uses Debian, see Developing LAVA on Debian for more information.
The LAVA software team use Jira for long term planning for new features and concepts. The JIRA instance used by LAVA is https://projects.linaro.org/browse/LAVA and anonymous access is available for anyone interested in LAVA to find out more about the future direction of LAVA. Not all features are available at this stage but all LAVA issues are visible individually. Not all issues will necesarily be delivered exactly as described, many descriptions are written well in advance of delivery of the feature.
Many git commit messages within the LAVA codebase contain references to JIRA
issues as LAVA-123
etc. All references like this can be appended to a basic
URL to find the details of that issue: https://projects.linaro.org/browse/
.
e.g. the addition of this section on JIRA relates to LAVA-735
which can be
viewed as https://projects.linaro.org/browse/LAVA-735
Within JIRA, there is a hierarchy of issues. EPIC is the highest level to group similar issues. Stories are each within a single EPIC and sub-tasks can exist within a single Story.
This information is made available for interest and to make our development process open to the community. If you have comments or questions about anything visible within the LAVA project, please subscribe to one of the mailing lists and ask your questions there. For bugs in the current release, please continue to file bug reports using Bugzilla.
Many stories contain comments linking directly to one or more gerrit reviews related to that story. When the review is merged, the story will be marked as resolved with a Fix Version matching the git tag of the release containing the fix from the review.
The LAVA software team use git review
to manage contributions. Each review
is automatically tested against all the unit tests. All reviews must pass all
unit tests before being considered for merging into the master branch. The
contributor is responsible for making the changes necessary to allow the unit
tests to pass and to keep the review up to date with other changes in the
master branch.
To setup git review
for the first time, install the package and setup the
local git configuration. (This can take a little time.):
$ apt -y install git-review
$ cd lava-server/
$ git review -s
Important
All changes need to support both Debian unstable and Debian stable - currently Jessie. This often includes multiple versions of django and other supporting packages. Automated unit tests are run on stable (with backports).
The master branch may be significantly ahead of the latest packages available from Debian (unstable or stable backports) which are based on the release branch. Use the LAVA repositories and/or Developer build versions to ensure that your instance is up to date with master.
See also
If you’d like to offer a patch (whether it is a bug fix, documentation update, new feature or even a simple typo fix) it is best to follow this simple check-list:
Clone the master branch of the correct project.
Create a new, clean, local branch based on master:
$ git checkout -b fixupbranch
Add your code, change any existing files as needed.
Commit your changes on the local branch.
Checkout the master branch and git pull
Checkout your existing local branch:
$ git checkout fixupbranch
rebase your local branch against updated master:
$ git rebase master
Fix any merge conficts. #. Send the patch to the Linaro Code Review system (gerrit):
$ git review
If successful, you will get a link to a review with the lava-team
already added as reviewers.
The unit tests will automatically start and you will be notified by email of the results and a link to the output which is useful if the tests fail.
See also
Patch Submissions and workflow for detailed information on running the unit tests and other static code analysis tools before submitting the review.
You are welcome to use the bug tracker of your chosen distribution. The maintainer for the packages in that distribution should Register with Linaro as a Community contributor with Linaro (or already be part of Linaro) to be able to forward bug reports and patches into the upstream LAVA systems.
If you, or anyone on your team, would like to register with Linaro directly, this will allow you to file an upstream bug, submit code for review by the LAVA team, etc. Register at the following url:
If you are considering large changes, it is best to register and also to subscribe to the lava-devel mailing list and talk to us on IRC:
irc.freenode.net
#linaro-lava
You can use the GitHub mirrors of git.linaro.org
to fork the LAVA packages
and make pull requests. Remember to make your change against the master
branch, not the github default branch of release
. Production releases are
based on the release
branch.
It is worth sending an email to the lava-devel mailing list, so that someone can migrate the pull request to a review.
Note
The process of creating or updating the review is not currently
linked to the github pull request process. You will need to respond to
comments on the review which will not appear in the pull request. LAVA
is not developed on github, the code is simply mirrored to github from
git.linaro.org
when a release is made. (So the github mirror of
master
can also be significantly behind current master
from
git.linaro.org
. Reviews use git.linaro.org
.)
See also