Graph-Learning-Benchmarks
[FEATURE REQUEST] Trigger tests by changes in /glb/
Is your feature request related to a problem? Please describe. Currently the tests for all dataset will not be triggered when there are changes in /glb/. We should trigger it after making the repo public.
See also for more context: https://github.com/Graph-Learning-Benchmarks/GLB-Repo/pull/179#issuecomment-1194877700
@xingjian-zhang @StevenGolden1203, when testing multiple datasets, are the tests running sequentially? And will the memory used by each dataset released once a test finished?
In other words, if we trigger testing on all the small datasets, will there be a memory issue given the Github Action worker only has 7G RAM?
BTW, the time limit for each job is 6 hours. So far we are far below this limit if we trigger test on all the small datasets.
From https://www.tutorialspoint.com/pytest/pytest_run_tests_in_parallel.htm:
By default, pytest runs tests in sequential order. In a real scenario, a test suite will have a number of test files and each file will have a bunch of tests. This will lead to a large execution time. To overcome this, pytest provides us with an option to run tests in parallel.
For the memory release, I think once a test is finished, the memory will be automatically released.
Besides, rather than run all tests simultaneously, I suggest we can use something like
pytest -n 3 ...
where -n <num> runs the tests by using multiple workers, here it is 3. We can try different numbers here and find a suitable choice.
By the way, it is possible to mark some tests as "slow" in pytest. So we can skip or run these tests after all other tests under different contexts. See https://stackoverflow.com/a/61539510
By the way, it is possible to mark some tests as "slow" in pytest. So we can skip or run these tests after all other tests under different contexts. See https://stackoverflow.com/a/61539510
This sounds great. Let's convert it into a feature request issue.
Activate the test trigger (by changes in /gli/) after #324 and #327 are resolved.