Currently, we do not have WL test metadata compatible with CircleCI. Therefore, we can't use CircleCI's built-in split-by timing to distribute tests efficiently. This PR simply runs matching and WolframModel (which take the most time by far) each in separate executors, and splits the rest among the last two.
In order to optimize this further, we could:
Distribute tests from matching and WolframModel into separate files.
Implement JUnit XML or Cucumber JSON test result reports for WL and enable split-by timing.
Examples
Parallel executors:
Test time is reduced by ~10 minutes (over ~2x reduction):
Changes
Comments
Currently, we do not have WL test metadata compatible with CircleCI. Therefore, we can't use CircleCI's built-in split-by timing to distribute tests efficiently. This PR simply runs
matchingandWolframModel(which take the most time by far) each in separate executors, and splits the rest among the last two.In order to optimize this further, we could:
matchingandWolframModelinto separate files.split-by timing.Examples
Parallel executors:
Test time is reduced by ~10 minutes (over ~2x reduction):
compared to master:
This change is