A configuration layer on top of Buildbot configuration dict.
You pass a simplified dict and it will return a new dict with all objects required for Buildbot configuration.
This project tries to simplify buildbot configuration by defining a set of common rules to be used by all builders. It reduced flexibility in favor or simplicity.
Steps are just shell commands which take arguments and environment variables.
It provide a few specialized steps (ex to get the source code) but is not designed for custom steps written in Python.
Once the buildbot configuration is defined as a simple dict it should be easy for any member of the team to maintain it (add slaves or builders).
For configuration the Buildbot you will still need to read buildbot documentation.
At this stage it is designed to be used for private buildmasters.
It provides the following features:
- Get single
- Multiple projects
- Slaves definition
- Multiple slaves for a builder.
- HTTP web status
- Try schedulers with username/password authentication
- Send build results to GitHub Pull Request state
- Send email notifications
I plan to add:
- Run steps in a Python virtualenv
- All defining Python version for virtualenv
- Scheduler based on GitHub PR commits
- Scheduler based on GitHub commits, to replace polling scheduler.
- IRC status... for example when master builds fail
It does not provide (patches welcomed):
- SVN, BZR, CVS...etc
- Get code from multiple sources
This project was inspired by Travis CI simple configuration file. I wanted to have something similar for Buildbot.
There are already a few project which can convert a travis.yml into a buildmaster configuration so this project is not designed this purpose.
I wanted multiple projects on a single Buildbot instance and with slaves on multiple operating systems.
I also wanted the powers of buildbot try which allows to test a patch without committing the changes.
I also wanted to have separate steps based on each builder, so that the documentation or static code analysis builders will not have to run the same steps as for the code test.
The reason why you ended up with buildbot is to get a builder which will run a set of steps, on a particular slave with option environment variables.
Normal buildbot builder are magically created based on an environment and a set of steps. environments are created based on a set of slaves and a set of environment variables.
Continue reading for info about slaves, environments, projects, steps.
A slave is the normal buildslave which has a name, password, max_builds. They a defined using the slaves root key.
You have the optional DEFAULT slave definition with settings for all slaves.
I hope that slaves are simple to understand, so here is an example:
{ OTHER_CONFIGS: {}, 'slaves': { # `DEFAULT` values are applied to all slaves and can be overwritten # for each slave. DEFAULT: { 'password': 'password', 'max_builds': 1, 'notify_on_missing': '[email protected]', }, # We have 2 linux slaved with default configuration. 'bs1c-lnx-ubuntu1404-x64-29': {}, 'bs1i-lnx-debian7-x86-37': {}, # This is a windows slave which uses a different password. 'bs1i-win-2012-x64-34': { 'password': 'other-password' }, # We have some linux slaves which can run many tasks. 'bsmeta1b-lnx-ubuntu1204-x86-30': {'max_builds': 100}, 'bsmeta2i-lnx-ubuntu1204-x86-32': {'max_builds': 100}, }, }
The concept of environment is inspired by Travis CI environments.
You create an environment by associated a set of (similar) slaves with a set of environment variables.
A slave can be part of multiple environments.
Environments are latest combined with projects and steps to create a builder.
You will define in a single place all the environments used by your projects. In this way you can share similar environments (builders) between multiple projects. Later, in project definition you can choose what environments are associated with a specific project.
The final builder generated from environment and the project will be named: PROJECTNAME-ENVNAME
All values from the environment dictionary are copied to builders, with the exception of the slaves list.
The builder will run only on one of the slaves associated with the environment, based on a random rule.
To increase availability you can define multiple slaves for an environment,hence for a builder.
You have the optional DEFAULT environment definition with settings for all environments.
In addition, the following environment values are always set:
- CI=true
- BUILDBOT=true
- COMMIT=current_revision
- BRANCH=name_of_the_branch
- BUILD_NUMBER=buildbot_build_number
- BUILDER_NAME=name_of_buildbot_builder
- BUILD_DIR=directory_where_test_is_executed
- TEST_ENVIRONMENT=name_of_builders_environment
- TEST_ARGUMENTS=builder_test_property
TEST_ARGUMENTS are not present in builds triggered by gatekeepers. This is done on purpose to avoid variable builds for gatekeepers. TEST_ARGUMENTS are copied in group builders, as well as skip steps.
The following environment variables are set if there is a property with the same name, but in lowercase:
- GITHUB_TOKEN
- GITHUB_PULL_ID
- CODECOV_TOKEN
- TEST_AUTHOR
Example:
{ OTHER_CONFIGS: {}, 'environments': { # These env args are applied to all environments. DEFAULT: { 'TEST_TYPE': 'normal' }, # We want to run tests on Linux # but you can create environments based on distro / CPU, etc # When combined with project named `brink` it will create a # builder named `brink-linux`. 'linux': { 'slaves: [ 'bs1c-lnx-ubuntu1404-x64-29', 'bs1a-lnx-centos7-x64-31' ], }, # This will generated a special builder which will overwrite the # default environment variable `TEST_TYPE` and also add a new one. # For project `brink` it will create builder `brink-leaks`. 'leaks': { 'slaves': [ 'bs1c-lnx-ubuntu1404-x64-29', 'bs1i-win-2012-x64-34', ] 'TEST_TYPE': 'leaks', 'CHEVAH_GC': 'yes', }, # In case an environment has no extra environment variables you can # defined it as a list of slaves. # This will create builder `brink-meta-director`. 'meta-director': [ 'bsmeta1b-lnx-ubuntu1204-x86-30', 'bsmeta2i-lnx-ubuntu1204-x86-32', ], }, }
You can use the same Buildbot installation for multiple projects. Projects don't have direct access to slaves but rather use the environments to run project's code on a slave.
A project defines a set of project specific options:
- repo - the url used to get project source
- github_slug - used to publish GitHub commit status
- poll_interval - number of seconds to wait for change source scheduler
It also defines a set of steps, a set of groups and a set of gatekeepers which are explained later.
A DEFAULT project can be used do define default values for all projects. These values will be used when a project does not specify a specific configuration.
Here is an example for defining 2 projects. Steps, groups and gatekeepers are explained later:
{ OTHER_CONFIGS: {} 'projects': { DEFAULT: { 'steps': [DEFAULT_STEPS_FOR_ALL_PROJECTS], 'poll_interval': 60, }, # `brink` project will use the default steps # will will check for changes on master much often. 'brink': { 'repo': 'http://git.chevah.com/brink.git', 'github_slug': 'chevah/brink', 'poll_interval': 30, 'groups': BRINK_GROUPS_EXPLAINED_LATER, 'gatekeepers': BRINK_GK_EXPLAINED_LATER, }, # `compat` project has a different set of steps. 'compat': { 'repo': 'http://git.chevah.com/compat.git', 'github_slug': 'chevah/compat', 'steps': [COMPAT_SPECIFIC_STEPS], 'groups': COMPAT_GROUPS_EXPLAINED_LATER, 'gatekeepers': COMPAT_GK_EXPLAINED_LATER, }, }, }
The following step types are available
- SLAVE_COMMAND - execute a shell command on slave
- SOURCE_COMMAND - get project code or apply patch
- MERGE_COMMAND - merge code with a branch
Default step type is SLAVE_COMMAND.
You can conditionally execute a step by using the optional configuration. In this case it will be executed only when when force_STEPNAME property is present on the builder and is not false.
The same set of steps are executed for all builders. In order to run different tests based on different environments/builders you should dispatch them bases on environment variable.
For example to run unit tests, pyflakes checker and documentation builder a single run_ci shell command is used which should dispatch a specialized command based on the environment variables.
Steps are defined inside the project's steps key:
from chevah.buildbot_configuration_builder.builder import ( MERGE_COMMAND, SOURCE_COMMAND, ) { OTHER_CONFIGS: {} 'projects': { 'brink': { 'steps': [ # Get source based on project settings. {'type': SOURCE_COMMAND}, # Merge with master. {'type': MERGE_COMMAND , 'branch': 'master'}, # Option clean the build folder. { 'name': 'clean', 'command': ['bash','./paver.sh', 'clean'], # This step is only executed when `force_clean=1` # builder property is defined. 'optional': True, }, # Build dependencies { 'name': 'deps', 'command': ['bash','./paver.sh', 'deps'], 'add_environment': { 'SOME_VAR': 'some-value', 'OTHER_VAR': Interpolate('%(prop:builder)s'), }, }, # Run tests { 'name': 'test', 'command': ['bash', './paver.sh', 'run_ci'], }, ], }, }, }
You can group multiple builders into a group. For example you can create a group named pr which will trigger all builders required to validate a pull request commit and another group named post-commit which will trigger all builders required to check the committed code. Or you can group them in 'supported' and 'experimental' builders.
For each member of a group a dedicated builder is created. This builder will execute the steps associated with this project. The builder will be named PROJECT_NAME-ENV_NAME.
A builder can be part of multiple groups.
The group will have a dedicated group builder which will trigger a build for all builders form the group and report the results once all builders are done.
The builder associated with GROUP_NAME for PROJECT_NAME will be named PROJECT_NAME-group-GROUP_NAME.
Groups are defined inside the project's groups key:
{ OTHER_CONFIGS: {} 'projects': { 'brink': { 'groups': { # This will create a builder named `brink-group-post-commit` # and we can use this builder to trigger multiple builders. 'post-commit': [ # Here is a list with environment names, NOT slave names. # This will created the following builders: # brink-leaks, brink-linux-x86, brink-windows-x64 'leaks', 'linux-x86', 'windows-x64' ], # This will create a builder named `brink-group-supported` # Since `leaks` builder is very slow we don't run to check # if a changes is ready for review. 'supported': [ 'linux-x86', 'windows-x64', ], # This will create a group named `brink-group-unstable`. 'unstable': [ 'solaris-x86', 'freebsd-x64', ], }, }, }, }
Gatekeepers are specialized builders which will not use the project's steps.
The following step types are available:
- SEQUENTIAL_GROUP - run all builders from a group one after another and wait for all builders to end.
- PARALLEL_GROUP - run all builders in parallel and wait for all to end.
- MASTER_COMMAND - run a shell command on master
- DIRECTORY_UPLOAD - upload a folder on master
By default gatekeeper builders are triggered by try schedulers. You can change this to trigger the builder based on changes on a branch using 'scheduler': 'master' option.
Gatekeepers are defined inside the project's gatekeepers key:
{ OTHER_CONFIGS: {} 'projects': { 'brink': { 'gatekeepers': { # Post commit builder which will run all tests # in sequential mode. 'master': { # Run this builder after changes are done on master branch. 'scheduler': 'master', 'stable_timer': 300, 'steps': [ { # Trigger each builde from a group, one after another. 'type': SEQUENTIAL_GROUP, # Name of the step show in logs. 'name': 'step name all', # Name of the triggered group. 'target': 'post-commit', }, { # Upload local folder `dist` on buildmaster. 'type': DIRECTORY_UPLOAD, 'name': 'upload_production', 'source': 'dist', 'destination': '/srv/buildmaster/upload/production', 'optional': True, }, { # Execute shell command on buildmaster. 'type': MASTER_COMMAND, 'name': 'Fix permissions', 'command': ['chmod', '-R', '755', '/srv/buildmaster/upload'], }, ], 'notifications': { # Once tests on master are done notify everyone on # both success or failure. 'email_all': ['[email protected]'], }, }, # Trigger tests and send result to GitHub for pull request # status when asked by buildbot try. 'review': { 'steps': [{ 'type': PARALLEL_GROUP, 'name': 'check review step', 'target': 'supported', 'set_properties': { 'codecov_token': '1234', }, 'copy_properties': ['github_pull_id'], }], 'notifications': { 'email_all': [INTERESTED_USERS], }, 'github_send_status': True, }, # Trigger tests and merge in master on success # When asked by buildbot try. 'merge': { 'steps': [ {'type': SOURCE_COMMAND}, { 'type': PARALLEL_GROUP, 'name': 'all', 'target': 'all', }, { 'name': 'merge-commit', 'command': ['bash', './paver.sh', 'merge_commit'], }, ], 'notifications': { 'email_passing': ['[email protected]'], 'email_error': [INTERESTED_USERS], }, }, }, }, }, }
You can directly define buildmaster configuration by using the global key from root dict:
{ OTHER_CONFIGS: {} 'global': { 'title': 'ACME Project Buildbot', 'db_url': 'sqlite:///state.sqlite', 'buildbotURL': 'http://build.domain.com/', 'titleURL': 'http://www.domain.com', 'slavePortnum': 10089, 'changeHorizon': 500, 'buildHorizon': 500, 'eventHorizon': 50, 'logHorizon': 50, 'buildCacheSize': 15, 'logCompressionMethod': 'gz', 'mergeRequests': True, }, }
Web status settings are defined in the web key of the root dict:
{ OTHER_CONFIGS: {} 'web': { 'interface': '0.0.0.0', 'port': 10088, 'htpasswd': '/srv/www/buildbot.passwd', 'authorization': { 'gracefulShutdown': True, 'forceBuild': True, 'forceAllBuilds': True, 'pingBuilder': True, 'stopBuild': True, 'stopAllBuilds': True, 'cancelPendingBuild': True, }, }, }
Gatekeeper builders can be configured to send email notifications.
Email settings are defined in the email key of the root dict:
def lookup(user): """ Called with each responsible user of the build and should return the full email address. """ return user + '@domain.tld' { OTHER_CONFIGS: {} 'email': { 'server': { 'fromaddr': '[email protected]', 'relayhost': 'smtp.domain.com', 'useTls': True, 'smtpPort': 587, 'smtpUser': 'stmp-user', 'smtpPassword': 'stmp-pass', }, # This is optional and use to convert buildbot users to # email addresses. 'user_to_email_mapper': lookup, # Available placeholders for subject: # result, projectName, title, builder 'subject': '%(result)s %(builder)s', }, }
Each builder has an associated Try scheduler so you can use buildbot try to run a patch or a revision.
Group builders and gatekeeper builders also have an associated try scheduler.
They are defined in the try_scheduler root key:
{ OTHER_CONFIGS: {} 'try_scheduler': { # Port on which to listed for buildbot try requests. 'port': 10087, # Credentials accepted by buildbot try. 'credentials': [('try_user_name', 'try_user_pass')], # Default environment/slaves used to execute the try schedulers for # groups and other builders which don't have an explicit environment. 'environment': 'meta-director', }, }
You can send Buildbot's results for a specific commit to GitHub. To send them you will need a GitHub token.
Each project also need to define the GitHub slug. See projects documentation.
GitHub is configured using the github key from root dict:
{ OTHER_CONFIGS: {} 'github': { 'token': 'GITHUB-TOKEN-VALUE', }, }
This is the basic content of your master.cfg file:
# You will need to import more things for here to define steps. from chevah.buildbot_configuration_builder.builder import ( generate_configuration ) config = { 'global': {YOUR_DATA}, 'try_scheduler': {YOUR_DATA}, 'github': {YOUR_DATA}, 'web': {YOUR_DATA}, 'email': {YOUR_DATA}, 'slaves': {YOUR_DATA}, 'environments': {YOUR_DATA}, 'projects': {YOUR_DATA}, } BuildmasterConfig = generate_configuration(config)