Skip to content

[NIXL] Fix NIXL UCX worker node placement #922

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

Merged
podkidyshev merged 1 commit into from
Jun 11, 2026
Merged

[NIXL] Fix NIXL UCX worker node placement #922

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
9 changes: 8 additions & 1 deletion src/cloudai/workloads/common/nixl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -335,7 +335,14 @@ def gen_nixlbench_srun_commands(self, test_cmd: list[str], backend: str) -> list
nnodes, _ = self.get_cached_nodes_spec()
if nnodes > 1:
cmds = [
[*prefix_part, "--overlap", f"--relative={idx}", *tpn_part, *bash_part] for idx in range(nnodes)
[
*prefix_part,
"--overlap",
f"--nodelist=$(scontrol show hostname $SLURM_JOB_NODELIST | sed -n '{idx + 1}p')",
*tpn_part,
*bash_part,
]
for idx in range(nnodes)
]
else:
cmds *= max(2, nnodes)
Expand Down
4 changes: 2 additions & 2 deletions tests/ref_data/nixl-kvbench.sbatch
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,9 +22,9 @@ timeout 60 bash -c "until curl -s $NIXL_ETCD_ENDPOINTS/health > /dev/null 2>&1;
echo "ETCD ($NIXL_ETCD_ENDPOINTS) was unreachable after 60 seconds";
exit 1
}
srun --export=ALL --mpi=pmix --container-image=url.com/docker:tag --container-mounts=__OUTPUT_DIR__/output:/cloudai_run_results,__OUTPUT_DIR__/install:/cloudai_install,__OUTPUT_DIR__/output --overlap --relative=0 --ntasks-per-node=1 --ntasks=1 -N1 bash -c "source __OUTPUT_DIR__/output/env_vars.sh; path/to/python path/to/kvbench_script.sh profile --backend UCX --etcd_endpoints http://$NIXL_ETCD_ENDPOINTS" &
srun --export=ALL --mpi=pmix --container-image=url.com/docker:tag --container-mounts=__OUTPUT_DIR__/output:/cloudai_run_results,__OUTPUT_DIR__/install:/cloudai_install,__OUTPUT_DIR__/output --overlap --nodelist=$(scontrol show hostname $SLURM_JOB_NODELIST | sed -n '1p') --ntasks-per-node=1 --ntasks=1 -N1 bash -c "source __OUTPUT_DIR__/output/env_vars.sh; path/to/python path/to/kvbench_script.sh profile --backend UCX --etcd_endpoints http://$NIXL_ETCD_ENDPOINTS" &
sleep 15
srun --export=ALL --mpi=pmix --container-image=url.com/docker:tag --container-mounts=__OUTPUT_DIR__/output:/cloudai_run_results,__OUTPUT_DIR__/install:/cloudai_install,__OUTPUT_DIR__/output --overlap --relative=1 --ntasks-per-node=1 --ntasks=1 -N1 bash -c "source __OUTPUT_DIR__/output/env_vars.sh; path/to/python path/to/kvbench_script.sh profile --backend UCX --etcd_endpoints http://$NIXL_ETCD_ENDPOINTS"
srun --export=ALL --mpi=pmix --container-image=url.com/docker:tag --container-mounts=__OUTPUT_DIR__/output:/cloudai_run_results,__OUTPUT_DIR__/install:/cloudai_install,__OUTPUT_DIR__/output --overlap --nodelist=$(scontrol show hostname $SLURM_JOB_NODELIST | sed -n '2p') --ntasks-per-node=1 --ntasks=1 -N1 bash -c "source __OUTPUT_DIR__/output/env_vars.sh; path/to/python path/to/kvbench_script.sh profile --backend UCX --etcd_endpoints http://$NIXL_ETCD_ENDPOINTS"
kill -TERM $etcd_pid
timeout 60 bash -c "while kill -0 $etcd_pid 2>/dev/null; do sleep 1; done" || {
echo "Failed to kill ETCD (pid=$etcd_pid) within 60 seconds";
Expand Down
4 changes: 2 additions & 2 deletions tests/ref_data/nixl_bench.sbatch
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,9 +22,9 @@ timeout 60 bash -c "until curl -s $NIXL_ETCD_ENDPOINTS/health > /dev/null 2>&1;
echo "ETCD ($NIXL_ETCD_ENDPOINTS) was unreachable after 60 seconds";
exit 1
}
srun --export=ALL --mpi=pmix --container-image=url.com/docker:2 --container-mounts=__OUTPUT_DIR__/output:/cloudai_run_results,__INSTALL_DIR__:/cloudai_install,__OUTPUT_DIR__/output --overlap --relative=0 --ntasks-per-node=1 --ntasks=1 -N1 bash -c "source __OUTPUT_DIR__/output/env_vars.sh; ./nixlbench --etcd-endpoints=http://$NIXL_ETCD_ENDPOINTS --backend=UCX" &
srun --export=ALL --mpi=pmix --container-image=url.com/docker:2 --container-mounts=__OUTPUT_DIR__/output:/cloudai_run_results,__INSTALL_DIR__:/cloudai_install,__OUTPUT_DIR__/output --overlap --nodelist=$(scontrol show hostname $SLURM_JOB_NODELIST | sed -n '1p') --ntasks-per-node=1 --ntasks=1 -N1 bash -c "source __OUTPUT_DIR__/output/env_vars.sh; ./nixlbench --etcd-endpoints=http://$NIXL_ETCD_ENDPOINTS --backend=UCX" &
sleep 15
srun --export=ALL --mpi=pmix --container-image=url.com/docker:2 --container-mounts=__OUTPUT_DIR__/output:/cloudai_run_results,__INSTALL_DIR__:/cloudai_install,__OUTPUT_DIR__/output --overlap --relative=1 --ntasks-per-node=1 --ntasks=1 -N1 bash -c "source __OUTPUT_DIR__/output/env_vars.sh; ./nixlbench --etcd-endpoints=http://$NIXL_ETCD_ENDPOINTS --backend=UCX"
srun --export=ALL --mpi=pmix --container-image=url.com/docker:2 --container-mounts=__OUTPUT_DIR__/output:/cloudai_run_results,__INSTALL_DIR__:/cloudai_install,__OUTPUT_DIR__/output --overlap --nodelist=$(scontrol show hostname $SLURM_JOB_NODELIST | sed -n '2p') --ntasks-per-node=1 --ntasks=1 -N1 bash -c "source __OUTPUT_DIR__/output/env_vars.sh; ./nixlbench --etcd-endpoints=http://$NIXL_ETCD_ENDPOINTS --backend=UCX"
kill -TERM $etcd_pid
timeout 60 bash -c "while kill -0 $etcd_pid 2>/dev/null; do sleep 1; done" || {
echo "Failed to kill ETCD (pid=$etcd_pid) within 60 seconds";
Expand Down
3 changes: 2 additions & 1 deletion tests/workloads/nixl_bench/test_command_gen_strategy_slurm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -241,7 +241,8 @@ def test_gen_nixl_srun_command(
assert "-N1" in cmd
if backend == "UCX":
if nnodes > 1:
assert f"--relative={idx}" in cmd
assert f"--nodelist=$(scontrol show hostname $SLURM_JOB_NODELIST | sed -n '{idx + 1}p')" in cmd
assert "--relative" not in cmd
else:
assert "--relative" not in cmd
assert "--nodelist=$SLURM_JOB_MASTER_NODE" in cmd
Expand Down
Loading