Skip to content
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

[core][compiled graphs] Support nsight.nvtx profiling #49392

Merged
merged 2 commits into from
Dec 26, 2024
Merged

[core][compiled graphs] Support nsight.nvtx profiling #49392

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f805ebc
[core][compiled graphs] Support nsight.nvtx profiling
ruisearch42 Dec 20, 2024
883a1e3
up
ruisearch42 Dec 21, 2024
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
21 changes: 19 additions & 2 deletions python/ray/dag/compiled_dag_node.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,10 @@
from ray.dag.dag_operation_future import GPUFuture, DAGOperationFuture, ResolvedFuture
from ray.experimental.channel.cached_channel import CachedChannel
from ray.experimental.channel.communicator import Communicator
from ray.dag.constants import RAY_CGRAPH_VISUALIZE_SCHEDULE
from ray.dag.constants import (
RAY_CGRAPH_ENABLE_NVTX_PROFILING,
RAY_CGRAPH_VISUALIZE_SCHEDULE,
)
import ray
from ray.exceptions import RayTaskError, RayChannelError
from ray.experimental.compiled_dag_ref import (
Expand Down Expand Up @@ -153,6 +156,17 @@ def do_exec_tasks(
for task in tasks:
task.prepare(overlap_gpu_communication=overlap_gpu_communication)

if RAY_CGRAPH_ENABLE_NVTX_PROFILING:
try:
import nvtx
except ImportError:
raise ImportError(
"Please install nvtx to enable nsight profiling. "
"You can install it by running `pip install nvtx`."
)
nvtx_profile = nvtx.Profile()
nvtx_profile.enable()

done = False
while True:
if done:
Expand All @@ -163,6 +177,9 @@ def do_exec_tasks(
)
if done:
break

if RAY_CGRAPH_ENABLE_NVTX_PROFILING:
nvtx_profile.disable()
except Exception:
logging.exception("Compiled DAG task exited with exception")
raise
Expand Down Expand Up @@ -1577,14 +1594,14 @@ def _get_or_compile(
executable_tasks.sort(key=lambda task: task.bind_index)
self.actor_to_executable_tasks[actor_handle] = executable_tasks

# Build an execution schedule for each actor
from ray.dag.constants import RAY_CGRAPH_ENABLE_PROFILING

if RAY_CGRAPH_ENABLE_PROFILING:
exec_task_func = do_profile_tasks
else:
exec_task_func = do_exec_tasks

# Build an execution schedule for each actor
self.actor_to_execution_schedule = self._build_execution_schedule()
for actor_handle, executable_tasks in self.actor_to_executable_tasks.items():
self.worker_task_refs[actor_handle] = actor_handle.__ray_call__.options(
Expand Down
7 changes: 7 additions & 0 deletions python/ray/dag/constants.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,13 @@
# Feature flag to turn on profiling.
RAY_CGRAPH_ENABLE_PROFILING = os.environ.get("RAY_CGRAPH_ENABLE_PROFILING", "0") == "1"

# Feature flag to turn on NVTX (NVIDIA Tools Extension Library) profiling.
# With this flag, Compiled Graph uses nvtx to automatically annotate and profile
# function calls during each actor's execution loop.
RAY_CGRAPH_ENABLE_NVTX_PROFILING = (
os.environ.get("RAY_CGRAPH_ENABLE_NVTX_PROFILING", "0") == "1"
)

# Feature flag to turn on visualization of the execution schedule.
RAY_CGRAPH_VISUALIZE_SCHEDULE = (
os.environ.get("RAY_CGRAPH_VISUALIZE_SCHEDULE", "0") == "1"
Expand Down
Loading