Split the JUnit tests into "new", "flaky", and the remainder.
On PR builds, "new" tests are anything that do not exist on trunk. They are run with zero tolerance for flakiness.
On trunk builds, "new" tests are anything added in the last 7 days. They are run with some tolerance for flakiness.
Another change included here is that we will not update the test catalog if any test job fails on a trunk build. We have had difficulty determining if all the tests had or not (due to timeout or failures in upstream Gradle tasks). By requiring green ":test" jobs, we can be sure that the resulting catalog will be valid.
---
The purpose of this change is to discourage contributors from adding flaky tests, but give some leeway for trunk so we have successful builds.
The "quarantinedTest" Gradle target has been consolidated into the regular "test" target. There are now some
runtime properties to control what tests are run.
* kafka.test.catalog.file: path to test catalog
* kafka.test.run.new: include new tests. this selection depends on the age of the loaded test catalog
* kafka.test.run.flaky: include tests marked as `@Flaky` (replaces the `excludeTags 'flaky'` directive)
* kafka.test.verbose: include additional logging from new JUnit classes (enabled by default if re-running GitHub workflow with debug logging enabled)
* maxTestRetries: how many retries to allow via Develocity retry plugin (default 0)
* maxTestRetryFailures: how many failures to allow before stopping retries (default 0)
Thanks to Jun Rao for inspiring the idea.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>, Ismael Juma <ismael@juma.me.uk>, Jun Rao <junrao@gmail.com>
In the case of a CI timeout, this patch uses jstack to capture thread dumps from the Gradle test workers.
These thread dumps are stored in files which are later archived by the CI workflow.
This patch also increases the compression level to 9 for our "actions/upload-artifact" steps to save a bit of storage space.
Reviewers: Chia-Ping Tsai <chia7712@gmail.com>
David Arthur