drkns is a simple build tool aimed at easing the continuous integration of monorepos through dependency aware caching.
Monorepositories contain many projects, it rarely makes sense to rebuild a every project when changes are committed.
But we can't limit ourselves to run builds and tests in directories where changes occurred because some projects depends on others.
You wild kid!
Confucius
Check a dummy project on which our tests are based.
From your monorepo root, in drkns.yml:
dependencies:
project1: projects/project1
project2: projects/project2
library1: library/library1
Sample projects/project1/drkns.yml:
checkSteps:
installDependencies: pipenv install
test: pipenv run pytest
buildSteps:
build: pipenv run paver build
deploy: pipenv run paver deploy_to_pypi
dependencies:
- ../../library/library1
drkns check # Check the configuration file
drkns list # Check list the identified steps
drkns run # Run everything
echo '.drknspersistence/*' >> .gitignore
With environment variables:
- DRKNS_S3_PATH: S3 path to a directory you can write to
- AWS_ACCESS_KEY_ID
- AWS_DEFAULT_REGION
- AWS_SECRET_ACCESS_KEY
pip install drkns
drkns check # We never know
drkns sync in --delete # Get past execution statuses from S3
drkns debug # If needed, you can check which steps are
# going to run here
drkns run # Run all steps
drkns clean # Will remove four weeks old execution statuses to
# limit disk usage
drkns sync out --delete # Persist all execution statuses from S3,
# **use --delete** to also delete data removed by
# `clean`, or not push old data that has been removed
# in the remote storage.
We all like stable builds but sometimes life happens and things break where they should not. We would love to relaunch our CI without having to just randomly edit a file to ignore the caching.
If you have a DRKNS_S3_PATH environment variable you can remove it from the
commands.
drkns sync in s3://my-drkns/project --delete
drkns sync forget dir1_project2/fedd32a-adbcc445555
drkns sync out s3://my-drkns/project --delete
The --delete flag is required, as S3 synchronization otherwise ignores files
that have been removed locally.
Each drkns.yml forms a unit, it should represent a specific unit of your
code, a directory containing a lib or a project, that has specific build steps
or can be depended on.
drkns.yml accepted fields:
directory: string, Current directory by default, only used to compute the hash associated with this build filedependencies: strings sequence, path containing anotherdrkns.ymlfile to load, you can also only indicate a directory name.checkSteps: dictionary of steps, see below.buildSteps: dictionary of steps, see below.cleanupSteps: dictionary of steps, see below.
Steps :
- Have a
commandto run, string, required. - Have a
backgroundflag, default to false. - Can also be a single string, the
commandargument. - Are ordered.
- Are deduplicated at execution. E.g.
library/front-util/drkns.ymlis required by bothprojects/aviation-customer/drkns.ymlandprojects/retail-customer/drkns.ymlit will only appear in the first one to be parsed. Note that this can impact.gitignoreresolution.
Three different step types are available, by order of execution:
checkSteps: steps necessary to check the integrity of theunit. Will always be ran if no checkSteps has already failed in thisunit. Typically used to build your application, your docker images and run your unit tests. If a steps fail, the following ones will not be executed.buildSteps: steps executed only ifcheckSteps, dependencies'checkStepsand previous buildSteps succeeded. You can use this to upload or deploy the build artifacts. Beware there is no guarantee thatdependencieswill be executed, there execution might have been startedcleanupSteps: steps executed no matter what (except if theunithas not changed since last execution). You can suppress things that have taken too much memory or things in the like.
If any of your steps used the background attributes they are killed after the cleanup steps at the end of the execution of your `unit.
The point of drkns is to not rerun everything for every change.
Upon execution a .drknspersistence folder in the directory of each unit will
be created, this folder will contain the execution status of each step for a
specific unit, step and a specific hash.
At each execution a hash for the unit is computed, it changes if files within the unit or its dependencies hashes have changed. If it is present in the persisted files, it will prevent the execution.
Often, it happens that the execution of steps will alter some files in the directory changing the mentioned above hash.
It is not necessarily something that we wish.
You can add partial or full file or directory names (as long as they have a
trailing /) to be ignored in a .drknsignore file.
A unit will ignore specific patterns.
Warning: .drknsignore are also inherited from parents. Because of this
path are not supported to avoid ambiguity.
The ignored elements are passed to dirhash's ignore argument. The relevant
documentation is available at https://pypi.org/project/dirhash/ .
Sample .drknsignore:
*.tmp
build/
Persistence is enabled through S3 buckets.
You need a DRKNS_S3_PATH environment variable, or provide it as a command line
argument.
export DRKNS_S3_PATH=s3://some-drkns-bucket/someProjectDirectory
drkns expects AWS environment variables
to be defined: AWS_ACCESS_KEY_ID, AWS_DEFAULT_REGION and
AWS_SECRET_ACCESS_KEY for the AWS client to work properly.
This is an advanced usage. You are invited to skip this if this is your first
dive in drkns.
Once your CI is setup, maybe you want to parallelize job execution, drkns can
generate your CI configuration, provided you write a template.
The template must be located at .drknsgeneration/*.template.*, only a single
template file must be in this directory.
If you need a more concrete example, check the dummy project sample main.template.yml.
Fields to be defined in the template, always in the form %FIELD_NAME%:
DRKNS_GROUP_BEGINand%DRKNS_GROUP_END%: around a group template definition. A group will containunitsthat are ran together. The lines containing those tags will be removed. For now, only one unit per group is supported, so do not be surprised.DRKNS_UNIT_BEGINandDRKNS_UNIT_END: around a unit template, mostly callingdrkns run %DRKNS_UNIT_NAME%. The lines containing those tags will be removed.%DRKNS_GROUP_UNITS%: where group units will be inserted - must be within a group template. The line containing this tag will be removed.%DRKNS_UNIT_NAME%: the unit's name%DRKNS_DEPENDENCY_GROUPS_NAMES{, }%: the dependency group names, with a comma separator (as an example), if you need a separator:%DRKNS_DEPENDENCY_GROUPS_NAMES?{, }%will only insert a,.%DRKNS_ALL_GROUPS_NAMES{, }%: all group names, here with a comma separator
drkns check: checksdrkns.ymlfiles by loading the one in the current directory and printing errors if anything bad happens.drkns clean: cleans older than four weeks execution steps from the persisted data.drkns debug: prints debug information, mostly to control if execution cache has been altered.drkns forget UNIT_NAME[/HASH]: forgets the previous execution ofUNIT_NAMEatHASH(or every execution if hash is not provided),allfor everything. Nota the name of the root config unit ismain. This can be used to force the new execution of some entries.drkns list: prints all the available steps plan. Beware repeating steps are removed.drkns run [--limit-output] [--force-success] [--summary] STEP_NAME: run the given steps and its requirements or all steps.drkns sync DIRECTION [--delete ][DRKNS_S3_PATH]: syncs the persisted execution cache,DIRECTIONcan beinorout.drkns generate: generates a file based on the template described above
--force-success: flag, forces exit code0.--limit-output: flag, less restrictive thansummary, limits the outputs to the failed steps' output at this execution (ignored cached ones), steps and execution statuses.--summary: flag, prints only the summary: steps and execution statuses.--delete: flag, deletes remote execution status files that do not exist locally. It is of paramount importance to use afterclean. It is a flag and not a default behavior as it makes sense to sync in / sync out at each step to prevent duplicate execution when a parallel processing is happening, or in the case the whole process is canceled.
drkns output is composed of the failed steps output only and a summary.
Output for unitName1@stepName1:
/bin/sh: line 1: stop: command not found
In case a step got canceled because of a previous failure:
Output for unitName2@stepName2:
Previous failure of dependency1 / dependency1FailingCheck
dependency1@dependency1FailingCheck: Error (restored)
dependency1@dependency1Build: Ignored
dependency1@dependency1Cleanup: OK (restored)
project1@project1Check: OK (restored)
project1@project1ShouldNotRunBuild: Ignored (restored)
project1@project1Cleanup: OK (restored)
project3@project3Build: OK
project3@project3failingCleanup: Error
project3@project3NeverExecutedCleanUp: OK
main@mainCheck1: OK
main@mainBuild1: Ignored
Step statuses:
Error: an error occurred on executionIgnored: an error occurred in a previous step, preventing this step from being executedOK: all good
(restored) indicates that the execution did not happen this round but was
cache from a previous step.