* [FEATURE] Docker simplification and configuration generation
The Authelia binary now will attempt to generate configuration based on the latest template assuming that the config location specified on startup does not exist. If a file based backend is selected and the backend cannot be found similarly it will generate a `user_database.yml` based a template.
This will allow more seamless bootstrapping of an environment no matter the deployment method.
We have also squashed the Docker volume requirement down to just `/config` thus removing the requirement for `/var/lib/authelia` this is primarily in attempts to simplify the Docker deployment.
Users with the old volume mappings have two options:
1. Change their mappings to conform to `/config`
2. Change the container entrypoint from `authelia --config /config/configuration.yml` to their old mapping
* Adjust paths relative to `/etc/authelia` and simplify to single volume for compose
* Add generation for file backend based user database
* Refactor Docker volumes and paths to /config
* Refactor Docker WORKDIR to /app
* Fix integration tests
* Update BREAKING.md for v4.20.0
* Run go mod tidy
* Fix log_file_path in miscellaneous.md docs
* Generate config and userdb with 0600 permissions
* Fix log_file_path in config.template.yml
* [CI] Add Codecov support
* [CI] Capture backend coverage from integration tests
* [CI] Remove unnecessary artifacts for coverage build
* [CI] Only run coverage elements where necessary
* [CI] Simplify post-command hook
* Fix yarn dependencies and collect coverage
* [CI] Include cmd/authelia/ path in coverage
* [CI] Exclude internal/suites/ in coverage
Closes#1061.
Prior to this change if there was a branch/PR build which had not yet published manifests and a master build running simultaneously, assuming the master build finished publishing manifests before former it would clean up the architecture tagged containers (-{amd64,arm32v7,arm64v8}) which would result in the manifest step failing for the branch or PR build.
These should not be considered in either of the clean up steps because they're removed as part of a successful manifest being published.
Pushes to master and tagged releases will have now have explicit dependencies for steps. This is specifically to prevent darwin based builds holding up execution of other steps which should not have a dependence.
If we have multiple builds to master that intend to deploy AUR packages or documentation, we must ensure that the jobs are locked and executed sequentially, not simultaneously. If they were to run simultaneously this has the ability to cause a race condition when attempting to commit the respective steps.
* [CI] Lint all builds except tagged commits to satisfy branch protection
* [CI] Add automatic retries for linting failures
This is to treat any issues with the reviewdog API server and occasional failures we are seeing.
* [FEATURE] Embed static assets in Go binary
* Refactor/consolidate code and specify public_html via configuration
* Update docs and config template for assets
* Update AUR package pre-requisites and systemd unit
* Include static assets as Buildkite and GitHub artifacts
* Remove references to PUBLIC_DIR
* Only serve assets via embedded filesystem and remove configuration references
* Update authelia-scripts helper to build the embedded filesystem
* Mock the embedded filesystem for unit tests
Add to gitignore to ensure this isn't overwritten.
* Move go:generate to satisfy linter
* [CI] Introduce linting for branch commits with reviewdog
This utilises the GitHub checks API and could be a potential candidate instead of in-line PR reviews.
* [CI] Change reporter to `github-check`
* [CI] Adjust linting in-line PR commentary to execute with linting step
This will ensure that notes pertaining to a version in the BREAKING.md will be published in each of the respective github releases.
All information from:
'## Breaking in $TAG' until the next '## Breaking in $TAG' is included.
* [Buildkite] Optimise pipeline for tagged deployments
Ensure Unit and Integration testing is bypassed for tagged builds.
* Apply suggestions from code review
Co-Authored-By: Clément Michaud <clement.michaud34@gmail.com>
Prior to this change all PR's which are merged into master would result in another run of the Unit and Integration testing.
This is not necessary because all steps have to pass for a PR to be accepted in to master, this will save significant time for deployments to master and reduce overall load to the Buildkite workers.