Skip to content

Fix reached_preemption crash when JAX preemption service is disabled #3458

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
marsunique wants to merge 1 commit into
base: main
Choose a base branch
from
+21 −3
Open

Fix reached_preemption crash when JAX preemption service is disabled #3458

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 8 additions & 3 deletions checkpoint/orbax/checkpoint/_src/multihost/multihost.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -423,9 +423,14 @@ def reached_preemption(step: int) -> bool:
if is_proxy_pathways_backend():
return False

preemption_sync_point_reached = multihost_utils.reached_preemption_sync_point(
step
)
try:
preemption_sync_point_reached = (
multihost_utils.reached_preemption_sync_point(step)
)
except RuntimeError:
# JAX raises RuntimeError when the preemption service is disabled
# via jax_enable_preemption_service config. Treat as no preemption.
return False
_maybe_log_reached_preemption(step, preemption_sync_point_reached)
return preemption_sync_point_reached

Expand Down
13 changes: 13 additions & 0 deletions checkpoint/orbax/checkpoint/_src/multihost/multihost_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -244,5 +244,18 @@ def setUp(self):
super().setUp()


class ReachedPreemptionTest(parameterized.TestCase):

@mock.patch.object(multihost, 'is_proxy_pathways_backend', return_value=False)
@mock.patch.object(multihost.multihost_utils, 'reached_preemption_sync_point')
def test_returns_false_when_service_disabled(
self, mock_sync_point, mock_is_proxy
):
mock_sync_point.side_effect = RuntimeError(
'Preemption sync manager has not been initialized.'
)
self.assertFalse(multihost.reached_preemption(step=5))


if __name__ == '__main__':
multiprocess_test.main()