Improve linter by going deeper into pipeline generation
Problem to solve
The CI Lint currently uses YamlProcessor to validate the content passed in. The YamlProcessor represents a small part of the whole pipeline creation process and it only covers syntax errors. Using YamlProcessor today it gives a false sense of validation to users because it can return that a configuration is valid but when trying to create a pipeline it fails for some other reasons not considered before. Typical examples are logical validations that occur beyond the YamlProcessor.
Example: The following syntax is considered valid by CI Lint but it would return an error test: needs 'build' when trying to actually create the pipeline.
build:
script: echo
stage: build
rules:
- if: '$CI_MERGE_REQUEST_ID'
test:
script: echo
stage: test
needs: [build]Intended users
User experience goal
Improved understanding of errors at lint-time.
Proposal
We will introduce a new mode in Ci::CreatePipelineService called dry_run: true (which is false by default). This mode would skip persisting the pipeline while letting everything else happening. In the end, a non-persisted pipeline is returned including any errors and (recently added) warnings.
Then we change the Lint controller to use Ci::CreatePipelineService with dry_run: true and pass in the content to validate.
Further details
Permissions and Security
Documentation
Availability & Testing
- Unit test changes - Yes, please add unit tests along with new/updated logics
- Integration test changes - not required
- End-to-end test change - not required