You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
By specifying test --local_resources=gpu_memory_mb=16000 --local_resources=memory=HOST_RAM*0.6 if we have a CI runner with 48 vcpus, 1 16 GB gpu, 96 GB of ram executing tests locally, we can limit test parallelism scheduling so that we can fully utilize the runner whilst still not running out of resources.
Is there a similar way to do this via buildbarn? I know the Initial Size Class Cache exists so that users don't need to maintain this test tag heuristic. However, for users that have this data, is there any way to leverage this data in buildbarn to limit concurrency to avoid Out of memory errors?
😅 I guess this is mainly a cost optimization question to maximize machine utilization whilst minimizing instabilities.
The text was updated successfully, but these errors were encountered:
On local execution, we can add resource tags for each test to help reduce likelihood of bazel running out of resources.
Assuming we have 100 tests tagged with the following,
By specifying
test --local_resources=gpu_memory_mb=16000 --local_resources=memory=HOST_RAM*0.6
if we have a CI runner with 48 vcpus, 1 16 GB gpu, 96 GB of ram executing tests locally, we can limit test parallelism scheduling so that we can fully utilize the runner whilst still not running out of resources.Is there a similar way to do this via buildbarn? I know the
Initial Size Class Cache
exists so that users don't need to maintain this test tag heuristic. However, for users that have this data, is there any way to leverage this data in buildbarn to limit concurrency to avoid Out of memory errors?😅 I guess this is mainly a cost optimization question to maximize machine utilization whilst minimizing instabilities.
The text was updated successfully, but these errors were encountered: