UPDATE
There is now support for detecting changed files and running multistage workflows natively in CircleCI. See these links: https://medium.com/nerd-for-tech/mastering-monorepos-with-circlecis-new-dynamic-config-2e187fe7934c https://circleci.com/blog/introducing-dynamic-config-via-setup-workflows/
UPDATE
There is a new experimental script in branchv2
that has the following improvements:
- Improved calculation of base commit for new branches that takes into account also the history of past CI builds. Current script doesn't identify correctly the parent when creating new branches from merge commits.
- Improved diff calculation using the two-dot git diff command.
- Improved calculation of CI workflow status that considers success when all jobs succeeds. Current script considers a workflow successful when at least one job succeeds.
- Possibility to customize the list of packages with advanced path specifications, that allows to list multiple paths for each package as well as exclude certain paths.
- Possibility to provide additional custom pipeline parameters.
- Possibility to customize the number of API pages with jobs to get from CircleCI. Existing script takes only the first page with 100 jobs.
The new script is planned to be published as CircleCI orb to allow for easier integration.
Monorepo brings simplicity to the development process by having all code in one place, but raises the complexity of automated builds and deploy.
For a relatively small monorepo it can be acceptable to have builds run for each service on every change. However, if you have a monorepo with a dozen of services/components, even the smallest change can introduce big delays in the CI process making it less efficient.
This repository is an example of configuring CircleCI for a monorepo that has four services. The CircleCI configuration file has workflows defined per each service, that are triggered on every push only when the corresponding service has code changes.
The sample services/components are api
, app
, auth
and gateway
all located in the subdirectory /packages
.
For each service a CircleCI workflow with the same name is defined in .circleci/config.yml
file.
This repository relies on CircleCI API v2 changes which are currently in Preview release.
Use of the v2 API as long as this notice is in the master branch (current as of June 2019) is done at your own risk and is governed by CircleCI’s Terms of Service.
Additionally, it requires use of v2.1 configuration files as well as having Pipelines enabled.
Whenever a change is pushed to GIT, by default the ci
workflow is triggered in CircleCI.
The ci
workflow consist of a single job trigger-workflows
, which performs a checkout
and executes the circle_trigger.sh
bash script from the .circleci
folder. The circle_trigger.sh
bash script is then responsible for detecting which services contain code changes and trigger their corresponding workflow via CircleCI API 2.0.
By convention, each service is located in a separate directory in packages
.
For each service there should be a separate workflow defined in the workflows
section in .circleci/config.yml
configuration file. The workflow is matched by the service's name, which is the same as the directory name from packages
.
Each workflow is conditioned using a when
clause, that depends on a pipeline parameter. The name of the parameter is the same as the name of the service.
Each workflow that corresponds to a service is triggered only when there are code changes in the corresponding service directory. Below is a more detailed explanation of how it detects changes:
- The first step consists of finding the commit SHA of the changes for which the most recently CircleCI pipeline was triggered
- Firstly it attempts to get the latest completed CircleCI workflow for current branch (together with the commit SHA)
- If there are no builds for current branch (which is usually the case with feature branches), it looks for nearest parent branch and gets its commit SHA (using this solution)
- If there are no builds for parent branch then it uses
master
- Once it has the commit SHA of latest CircleCI workflow, it uses
git log
to list all changes between the two commits and flag those services for which changes are found in their directories.
To be able to trigger workflows via API, you need a CircleCI personal API token.
The circle_trigger.sh
script expects to find the token in CIRCLE_TOKEN
environment variable.
To prevent having the tokens published to git, you can use project environment variables.
Once you add a new service/component to your monorepo you have to do the following steps:
-
Add a directory in
packages/
that will be the root of your service/component. The name of the directory will be used as the name of the service/component. -
In
.circleci/config.yml
configuration file inparameters
section add a corresponding parameter:
parameters:
my_awesome_service: # this should be the name of your service
type: boolean
default: false
- In
.circleci/config.yml
configuration file injobs
section define all the jobs that are need by current service/component. To set the job's working directory to the directory of the service, you can use job parameter:
build:
parameters:
package_name:
type: string
working_directory: ~/project/packages/<< parameters.package_name >>
...
steps:
- checkout:
path: ~/project
- In
.circleci/config.yml
configuration file inworkflows
section add a corresponding workflow:
workflows:
when: << pipeline.parameters.my_awesome_service >> # the name of the parameter is the same as service name
jobs:
# add here the jobs used by this workflow
You have no restrictions on which jobs can be used by each workflow. It can be a service specific job, or a job reused by several workflows (services). In case you have dependencies between jobs within a workflow, you can have a custom name for the job and then use that name as a requirement:
service:
...
jobs:
- build:
name: service-build # give a name for the `build` job used in current workflow
package_name: service # name of the service passed as parameter; used to set the working directory
...
- deploy:
...
requires:
- service-build # list as a dependency the `build` job from this workflow
In CircleCI dashboard when viewing a job details you can see who triggered it. In this case it will be the team member whose API token is configured in CIRCLE_TOKEN
environment variable (described above).
It would be nicer to show there the name of the team member who made the code changes and triggered the workflow. This allows to see then who triggered a specific job, and have a better integration with CircleCI notifications.
Unfortunately CircleCI currently doesn't support user specific environment variables. One way to workaround it is to define in the CircleCI project
a variable for each team member by following a convention, and then use the built-in $CIRCLE_USERNAME
to get the name of the variable:
TOKEN_NAME=CIRCLE_TOKEN_${CIRCLE_USERNAME} # this however will fail if username contain chars like `-`, '.' etc.
CIRCLE_TOKEN=${!TOKEN_NAME}
Dependencies can be handled using circleci's advanced logic with the or
operator.
Example: if package api
depends on common
, you should add common
as a pipeline parameter and use it to define the when
parameter for api's workflow
workflows:
version: 2
# The main workflow responsible for triggering all other workflows
# in which changes are detected.
ci:
when: << pipeline.parameters.trigger >>
jobs:
- trigger-workflows
# Workflows defined for each package.
api:
when:
or:
- << pipeline.parameters.api >>
- << pipeline.parameters.common >>
jobs:
- build:
name: api-build
package_name: api
- deploy:
name: api-deploy
package_name: api
requires:
- api-build
Now, changes in package common
will also trigger build for api