Hello everyone, There have been recurring discussions about minimizing the usage of caplog in unit tests. If logs are necessary for a test, we should mock them unless the test strictly requires actual log output. Otherwise, the caplog tests should be removed and logs should be mocked.
During recent large migrations, changes and ongoing CI/CD efforts, caplog has proven to be flaky in unit tests. They frequently cause red pipelines in PRs and scheduled CI/CD runs. There have been local discussions on removing caplog from unit tests. To formalize this, I propose a lazy consensus to remove and prevent caplog usage in unit tests, ensuring logs are mocked when needed and disallowing caplog without mocking unless explicitly approved. *Why should caplog be avoided?* - Big maintenance effort on CI/CD - Instability *What should we do instead if we use caplog in a unit test?* - Mocking the Log *In the Scope of this consensus* - Remove caplog usage from unit tests if possible - Mock logs and remove caplog from unit tests if possible - Exceptional cases will be subject to maintainer approval - Prevent caplog to be included in unit tests without explicitly mocking the log *Action Items related to the above Scope:* - Scan and replace caplog tests with mocking where possible - Scan and remove caplog tests where possible - Include CI check to prevent adding additional caplog tests without Mocking and/or without approval from a maintainer to allow flexibility in some exceptional cases - Create documentation with examples for implementing tests using logs without caplog (e.g., using mocking or avoiding logs entirely) The lazy consensus will be reached on 2025-02-04, (unless someone surfaces an objection). -- Bugra Ozturk
