Problem
On main commit d02666d, API unit-test steps took 258s, 276s, and 227s across three equally sized shards. The parallel pytest sessions took 171.69s, 184.99s, and 142.46s, with a noticeable tail near completion.
Run: https://github.com/langgenius/dify/actions/runs/35502070968
The current progress-only logs do not identify slow test phases or distinguish collection time from execution. There are no retained per-test timing reports to inform balancing.
Proposed change
Print the slowest 50 test phases, retain JUnit XML for both parallel unit tests and serial controller tests, and report collection wall time per pytest process. Keep scheduling, coverage, and test selection unchanged while collecting evidence for subsequent optimizations.
Problem
On main commit d02666d, API unit-test steps took 258s, 276s, and 227s across three equally sized shards. The parallel pytest sessions took 171.69s, 184.99s, and 142.46s, with a noticeable tail near completion.
Run: https://github.com/langgenius/dify/actions/runs/35502070968
The current progress-only logs do not identify slow test phases or distinguish collection time from execution. There are no retained per-test timing reports to inform balancing.
Proposed change
Print the slowest 50 test phases, retain JUnit XML for both parallel unit tests and serial controller tests, and report collection wall time per pytest process. Keep scheduling, coverage, and test selection unchanged while collecting evidence for subsequent optimizations.