hiro/ray
1
0
Fork 0
mirror of https://github.com/vale981/ray synced 2025-03-05 10:01:43 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

[Test] Fix broken test_base_trainer (#27855)

Browse source 
The test was written incorrectly. This root cause was that the trainer & worker both requires 1 CPU, meaning pg requires {CPU: 1} * 2 resources.

And when the max fraction is 0.001, we only allow up to 1 CPU for pg, so we cannot schedule the requested pgs in any case.
This commit is contained in:
SangBin Cho 2022-08-15 23:50:18 +09:00 • committed by GitHub
parent a2c168cd6d
commit d654636bfc
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 7 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
python/ray/train/tests/test_base_trainer.py
View file

@ -183,9 +183,13 @@ def test_reserved_cpus(ray_start_4_cpus):
)
tune.run(trainer.as_trainable(), num_samples=4)
# TODO(ekl/sang) this currently fails.
# Check we don't deadlock with too low of a fraction either.
scale_config = ScalingConfig(num_workers=1, _max_cpu_fraction_per_node=0.01)
# Needs to request 0 CPU for the trainer otherwise the pg
# will require {CPU: 1} * 2 resources, which means
# _max_cpu_fraction_per_node == 0.01 cannot schedule it
# (because this only allows to have 1 CPU for pg per node).
scale_config = ScalingConfig(
num_workers=1, _max_cpu_fraction_per_node=0.01, trainer_resources={"CPU": 0}
)
trainer = DummyTrainer(
train_loop,
scaling_config=scale_config,