Store and sort origin tests by duration to avoid a long tail problem where long running tests at the end of a test bucket end up executing alone because their duration far outlasts the rest of the tests in the bucket.

By front-loading (scheduling) long running tests first, we can ensure they run in parallel with other tests rather than executing alone at the end, which can save significant time in overall test execution.

This work builds on the scheduler work being done in TRT-2292 for test isolation and orchestration.

Implementation

• Store test duration metadata
• Sort tests by duration (longest first) when scheduling test buckets
• Ensure long running tests are scheduled early in bucket execution
• Verify this doesn't conflict with isolation requirements from TRT-2292

Expected Benefits

• Reduced overall test execution time
• Better parallelization of test buckets
• Elimination of long tail where a single long test runs alone