- yml syntax
- Auto-complete
- Run command
- Jobs
- Tools
- Team approval workflow
- Giphy generator
- Custom github actions
- When to create github actions
- Github actions return value
- Github actions exit code
- Github action types
- Github action metadata
- Github actions - communicating with the Host by using logging commands
- Public vs. private actions
- Github actions - runtime
- Create JavaScript action using TypeScript
- Create Docker action - auto release milestone
- Enable debug level in GitHub actions
- Publishing actions
- Versioning
- links
https://onlineyamltools.com/convert-yaml-to-json
In online github editor make sure that cursor is set in the right place - only then auto complete appears.
No auto-complete:
Auto-complete:
VSCode auto-complete seems to be fine:
This will run only one command npm install and second command npm test will be silently skipped:
run:
npm install
npm test
Use |
to run multiple commands:
run: |
npm install
npm test
To run next command only if previous was successful:
run: |
npm install && npm test
run: |
npm install && npm test
working-directory: |
./src/project
This is helpful because every run
commands starts by default in the checked out directory, for example: /home/runner/work/github-action-publish-npm/github-action-publish-npm
. If we do not want use working-directory
then we have to go with single run
command, for example:
- run: |
cd my-workspace && npm install && npm run build @kicaj29/lib1 --prod && cd dist/kicaj29/lib1 && npm publish
Each job executes on fresh instance of a virtual environment that`s why jobs do not share anything (like file system). To share artifacts in the same workflow between different jobs use actions upload and download to share to build output which is needed to run the unit tests. https://docs.github.com/en/actions/guides/storing-workflow-data-as-artifacts
Use needs: [job-name]
to control when to start particular jobs.
Use https://github-actions-hero.vercel.app/ and its playground to visualize a workflow. In this view it is easy to understand which parts are executed in parallel. https://github-actions-hero.vercel.app/playground
- easy way to see when enough approvals has been achieved
- branch protections: https://github.com/kicaj29/github-actions-demo/settings/branches
- required review approvals
- matrix build
- save build artifacts
- dedicated test job
Using https://developers.giphy.com/dashboard/?create=true create a GIPHY_TOKEN and add it to the secrets:
If we will add comment that starts with /giphy
then selected git will be added by this workflow:
Actions run in their own isolated process.
When to create github actions (not only re-use reason):
- Control flow: you need more control flow structures then simple if conditions
Without github action:
run: |
npm install && npm test
working-directory: |
./src/project
run: |
npm install && npm test
working-directory: |
./src/another
run: |
npm install && npm test
working-directory: |
./src/yet-another
With github action:
name: Build
uses: actions/npm-build@v1
with:
working-directories:
- src/project
- src/another
- src.yet-another
- Diagnostic: you need to handle errors to make them easier to diagnose
In workflow it is not possible to specify job that runs only when another job failed:
jobs:
install:
steps:
- name: Clean install
run: npm ci
build:
needs: install
steps:
- name: Build
run: npm run build
error:
steps:
- name: Handle error
run: ...
- Complexity
Workflows are declarative and are designed to specify a processes in high level of abstraction. Details should be handled outside of the workflow in actions where are better tools to handle complexity.
For example running a command line with many parameters could be encapsulated in github action:
- name: Deploy
env:
USER: ${{secrets.USER}}
PWD: ${{secrets.PWD}}
run: |
curls -Ss \
--fail \
-X POST \
-T path/to/app.zip \
-u $USER:$PWD \
https://example.com/app
- Re-use
jobs:
build_and_deploy:
steps:
- name: Build & Deploy
uses: ./npm-deploy@v1
build_and_deploy_another:
steps:
- name: Build & Deploy
uses: ./npm-deploy@v1
and_another:
steps:
- name: Build & Deploy
uses: ./npm-deploy@v1
${{ steps id outputs name }}
id: step name?
name: name of the output parameter?
exit 0 - success exit 1 - fail
The workflow looks on action exit code to decide whether should continue with the next step or should stop with an error.
Pros:
- fast: run directly on the host machine
- development support: easy to integrate with the github environment
Cons:
- require setup: any external dependencies must be installed
Pros:
- you can package everything in a single container and the workflow runs it as a unit
- flexible: can run any software
- consistent: action and its dependencies are bundled together so you do not need to worry about installing anything on the host
Cons:
- the are slower then JavaScript actions: it takes time do download and start a container
- linux only
metadata: action name, descriptions, input and output params, if it is js or docker action.
Metadata are stored in file action.yml
.
name: 'My super action'
description: 'Super description'
author: 'Jacek'
inputs:
parameter-name:
description: 'An input parameter.'
required: true # by default params are optional
outputs:
output-value-name:
description: 'A value returned by the action.'
# for JS actions:
runs:
using: 'node12'
main: 'lib/main.js' # entry point
# for docker actions:
runs:
using: 'docker'
image: 'Dockerfile'
# or:
runs:
using: 'dockerfile'
image: 'docker://image:tag'
# if we want publish an action in the github marketplace:
branding:
icon: 'power' # it has to be one of icon offered by http://feathericons.com/
color: 'yellow'
Logging commands are instructions for the workflow host that are embedded in the build log.
WORKFLOW -> BUILD LOG -> HOST
Syntax:
::log-command param1=arg,param2=arg::command value
Sample logging commands:
::set-env name=DEBUG::1
::set-output name=name_of_param::value
::add-path::/path/to/directory # adds path to the PATH variable of the host
::debug:: file=name, line=0,col=0::message # file, line, col are optional
::warning:: file=name, line=0,col=0::message # file, line, col are optional
::error:: file=name, line=0,col=0::message # file, line, col are optional
::add-mask::message #masked message (for example for secret value)
::add-mask::$VARIABLE #masked message (for example for secret value)
::stops-commands::token # tells the host to stop interpreting the logs as commands
::token:: # tells the host to start interpreting the logs as commands
Log:
Download action repository...
Setting environment variables...
::set-env name=DEBUG::1
[command]program --param=arg
running program with `arg`...
Done
::set-env name=DEBUG::1
means
env:
DEBUG:1
- public: workflow and action are stored in different repositories. Create public actions if they will be used in many different workflows.
- private: workflow and action are stored in the same repo
- public and private actions are referenced differently from a workflow:
- public
- name: public action uses: owner/repo@ref # ref can be commit or tag (version), semantic versioning has to be used, branch name
- name: public action uses: owner/repo@12az34
- name: public action uses: owner/repo@v1.2.3
In case of docker image additionally it is possible to specify docker image url:- name: public action uses: owner/repo@master
- name: public action uses: docker://image:tag
- private
- name private action uses: ./.github/actions/... # it is convention to use such path
- public
For repository proper permissions have to be configured:
NOTE: Public in public action does it mean that source code of the action will be available for all. You can have public action that is hosted in private repository in which case it is only available to other repositories that are part of the same organization.
- first an event that triggers a workflow is fired
- next all needed actions are downloaded
- temporary workspace directory is created, every time it has different name so we have to refer to it using
$GITHUB_WORKSPACE
- to access the workspace actions can use
upload-artifact
and ```download-artifact`` actions - the workspace folder is deleted when the workflow is done
- to check what events triggered the workflow read
event.json
. This file is also temporary and can be read using$GITHUB_EVENT_PATH
.
Repo link: https://github.com/kicaj29/auto-release-draft
Repository: https://github.com/kicaj29/github-action-docker-auto-release-milestone
Create secret ACTIONS_STEP_DEBUG
with value true
. Then code like this
core.debug(`The previous version tag us ${previousTag}`);
will appear in the log output.
https://github.com/marketplace/actions/kicaj-29-auto-release-milestone
First semantic versioning should be used when we create tags.
It is recommended to refer to actions by using floating versioning, for example:
uses: actions/checkout@v2
then for example when a new version with some bug fixes is available let`s say v2.0.1 it will be automatically used by the workflow.
https://github.com/a-a-ron/github-actions-course-template
http://github.com/marketplace
https://github.com/marketplace/actions/label-approved-pull-requests
https://github.com/marketplace/actions/giphy-generator
https://lab.github.com
https://github.com/nektos/act (Run your GitHub Actions locally)
...