Skip to content

Fail fast when overlap wait times out #20

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
TianyeGGBond wants to merge 1 commit into
base: zhenyu/miles-mvp-e2e
Choose a base branch
from
Open

Fail fast when overlap wait times out #20

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
26 changes: 11 additions & 15 deletions rlix/pipeline/miles_pipeline.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -554,10 +554,9 @@ def _wait_for_overlap_engines_offloaded(self, allocated_train_gpus, *, timeout_s
rollout_manager.get_engine_states.remote(target_indices)
)
except Exception as exc: # noqa: BLE001
logger.warning(
"_wait_for_overlap_engines_offloaded: get_engine_states failed: %r", exc
)
return
raise RuntimeError(
"_wait_for_overlap_engines_offloaded: get_engine_states failed"
) from exc
uniq = {states.get(i, "?") for i in target_indices}
if uniq.issubset({"offloaded", "shell"}):
logger.info(
Expand All @@ -567,10 +566,9 @@ def _wait_for_overlap_engines_offloaded(self, allocated_train_gpus, *, timeout_s
break
time.sleep(0.1)
else:
logger.warning(
"_wait_for_overlap_engines_offloaded: state timeout after %.1fs; "
"engines %s still in state=%r",
timeout_s, target_indices, uniq,
raise RuntimeError(
"_wait_for_overlap_engines_offloaded: state timeout after "
f"{timeout_s:.1f}s; engines {target_indices} still in state={uniq!r}"
)

# Phase 2: probe nvidia-smi for OS-level free memory on the
Expand Down Expand Up @@ -607,13 +605,11 @@ def _wait_for_overlap_engines_offloaded(self, allocated_train_gpus, *, timeout_s
)
return
time.sleep(0.5)
logger.warning(
"_wait_for_overlap_engines_offloaded: free-mem timeout after %.1fs; "
"min_free_gb=%.2f below %.1f GB target on GPUs %s — wake_up may OOM",
timeout_s,
last_min_free_gb if last_min_free_gb is not None else float("nan"),
target_free_gb,
target_gpu_ids,
raise RuntimeError(
"_wait_for_overlap_engines_offloaded: free-mem timeout after "
f"{timeout_s:.1f}s; min_free_gb="
f"{last_min_free_gb if last_min_free_gb is not None else float('nan'):.2f} "
f"below {target_free_gb:.1f} GB target on GPUs {target_gpu_ids}"
)

@staticmethod
Expand Down