Skip to content

add support for sandbox #494

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
wedu-nvidia wants to merge 2 commits into
base: main
Choose a base branch
from
Open

add support for sandbox #494

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
0a0c50f
add support for sandbox
wedu-nvidia Apr 17, 2026
4db0994
Add sandbox support to Ray templates and update tests
wedu-nvidia Apr 17, 2026
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
27 changes: 27 additions & 0 deletions nemo_run/run/ray/templates/ray.sub.j2
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -303,6 +303,33 @@ while ! (srun {% if heterogeneous %}--het-group=0 {% endif %}--overlap --nodes=1
elapsed_time=$((elapsed_time + 2))
done

# Run sandbox in parallel across all allocated nodes when explicitly configured.
if [[ -n "${SANDBOX_CONTAINER:-}" ]] && [[ -n "${SANDBOX_COMMAND:-}" ]]; then
SANDBOX_PORTS_DIR="${SANDBOX_PORTS_DIR:-$LOG_DIR/sandbox}"
mkdir -p "$SANDBOX_PORTS_DIR"
echo "[INFO] Starting sandbox across all nodes in parallel (ports_dir=$SANDBOX_PORTS_DIR)..."
srun {% if heterogeneous %}--het-group=0 {% endif %}--output "$LOG_DIR/sandbox.log" \
--error "$LOG_DIR/sandbox.log" \
--container-image="$SANDBOX_CONTAINER" \
--container-mounts="$SANDBOX_PORTS_DIR:$SANDBOX_PORTS_DIR" \
--container-env=SANDBOX_PORTS_DIR,LISTEN_PORT,NGINX_PORT \
--no-container-mount-home \
--mpi=pmix \
-A "$SLURM_JOB_ACCOUNT" \
-p "$SLURM_JOB_PARTITION" \
--wait=60 \
--kill-on-bad-exit=1 \
--overlap \
--nodes="$SLURM_JOB_NUM_NODES" \
--ntasks-per-node=1 \
--export=ALL,SANDBOX_PORTS_DIR=$SANDBOX_PORTS_DIR,LISTEN_PORT=${SANDBOX_PORT:-6000},NGINX_PORT=${SANDBOX_PORT:-6000} \
bash -c "$SANDBOX_COMMAND" &
SRUN_PIDS["sandbox"]=$!
echo "[INFO] Sandbox started in background (PID: ${SRUN_PIDS["sandbox"]})"
else
echo "[INFO] SANDBOX_CONTAINER or SANDBOX_COMMAND not defined, skipping sandbox startup"
fi

NUM_ACTORS=$((gpus_per_node * SLURM_JOB_NUM_NODES))

# Start Ray worker nodes
Expand Down
27 changes: 27 additions & 0 deletions nemo_run/run/ray/templates/ray_enroot.sub.j2
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -312,6 +312,33 @@ while true; do
elapsed_time=$((elapsed_time + 2))
done

# Run sandbox in parallel on the head node (overlap) when explicitly configured.
if [[ -n "${SANDBOX_CONTAINER:-}" ]] && [[ -n "${SANDBOX_COMMAND:-}" ]]; then
SANDBOX_PORTS_DIR="${SANDBOX_PORTS_DIR:-$LOG_DIR/sandbox}"
mkdir -p "$SANDBOX_PORTS_DIR"
echo "[INFO] Starting sandbox on head node in parallel (ports_dir=$SANDBOX_PORTS_DIR)..."
srun --output "$LOG_DIR/sandbox.log" \
--error "$LOG_DIR/sandbox.log" \
--container-image="$SANDBOX_CONTAINER" \
--container-mounts="$SANDBOX_PORTS_DIR:$SANDBOX_PORTS_DIR" \
--container-env=SANDBOX_PORTS_DIR,LISTEN_PORT,NGINX_PORT \
--no-container-mount-home \
--mpi=pmix \
-A "$SLURM_JOB_ACCOUNT" \
-p "$SLURM_JOB_PARTITION" \
--wait=60 \
--kill-on-bad-exit=1 \
--overlap \
--nodes="$SLURM_JOB_NUM_NODES" \
--ntasks-per-node=1 \
--export=ALL,SANDBOX_PORTS_DIR=$SANDBOX_PORTS_DIR,LISTEN_PORT=${SANDBOX_PORT:-6000},NGINX_PORT=${SANDBOX_PORT:-6000} \
bash -c "$SANDBOX_COMMAND" &
SRUN_PIDS["sandbox"]=$!
echo "[INFO] Sandbox started in background (PID: ${SRUN_PIDS["sandbox"]})"
else
echo "[INFO] SANDBOX_CONTAINER or SANDBOX_COMMAND not defined, skipping sandbox startup"
fi

NUM_ACTORS=$((gpus_per_node * SLURM_JOB_NUM_NODES))

# Start Ray worker nodes
Expand Down
27 changes: 27 additions & 0 deletions test/core/execution/artifacts/expected_ray_cluster.sub
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -290,6 +290,33 @@ while ! (srun --overlap --nodes=1 --ntasks=1 -w $head_node test -f $LOG_DIR/STAR
elapsed_time=$((elapsed_time + 2))
done

# Run sandbox in parallel across all allocated nodes when explicitly configured.
if [[ -n "${SANDBOX_CONTAINER:-}" ]] && [[ -n "${SANDBOX_COMMAND:-}" ]]; then
SANDBOX_PORTS_DIR="${SANDBOX_PORTS_DIR:-$LOG_DIR/sandbox}"
mkdir -p "$SANDBOX_PORTS_DIR"
echo "[INFO] Starting sandbox across all nodes in parallel (ports_dir=$SANDBOX_PORTS_DIR)..."
srun --output "$LOG_DIR/sandbox.log" \
--error "$LOG_DIR/sandbox.log" \
--container-image="$SANDBOX_CONTAINER" \
--container-mounts="$SANDBOX_PORTS_DIR:$SANDBOX_PORTS_DIR" \
--container-env=SANDBOX_PORTS_DIR,LISTEN_PORT,NGINX_PORT \
--no-container-mount-home \
--mpi=pmix \
-A "$SLURM_JOB_ACCOUNT" \
-p "$SLURM_JOB_PARTITION" \
--wait=60 \
--kill-on-bad-exit=1 \
--overlap \
--nodes="$SLURM_JOB_NUM_NODES" \
--ntasks-per-node=1 \
--export=ALL,SANDBOX_PORTS_DIR=$SANDBOX_PORTS_DIR,LISTEN_PORT=${SANDBOX_PORT:-6000},NGINX_PORT=${SANDBOX_PORT:-6000} \
bash -c "$SANDBOX_COMMAND" &
SRUN_PIDS["sandbox"]=$!
echo "[INFO] Sandbox started in background (PID: ${SRUN_PIDS["sandbox"]})"
else
echo "[INFO] SANDBOX_CONTAINER or SANDBOX_COMMAND not defined, skipping sandbox startup"
fi

NUM_ACTORS=$((gpus_per_node * SLURM_JOB_NUM_NODES))

# Start Ray worker nodes
Expand Down
29 changes: 28 additions & 1 deletion test/core/execution/artifacts/expected_ray_cluster_enroot.sub
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -306,6 +306,33 @@ while true; do
elapsed_time=$((elapsed_time + 2))
done

# Run sandbox in parallel on the head node (overlap) when explicitly configured.
if [[ -n "${SANDBOX_CONTAINER:-}" ]] && [[ -n "${SANDBOX_COMMAND:-}" ]]; then
SANDBOX_PORTS_DIR="${SANDBOX_PORTS_DIR:-$LOG_DIR/sandbox}"
mkdir -p "$SANDBOX_PORTS_DIR"
echo "[INFO] Starting sandbox on head node in parallel (ports_dir=$SANDBOX_PORTS_DIR)..."
srun --output "$LOG_DIR/sandbox.log" \
--error "$LOG_DIR/sandbox.log" \
--container-image="$SANDBOX_CONTAINER" \
--container-mounts="$SANDBOX_PORTS_DIR:$SANDBOX_PORTS_DIR" \
--container-env=SANDBOX_PORTS_DIR,LISTEN_PORT,NGINX_PORT \
--no-container-mount-home \
--mpi=pmix \
-A "$SLURM_JOB_ACCOUNT" \
-p "$SLURM_JOB_PARTITION" \
--wait=60 \
--kill-on-bad-exit=1 \
--overlap \
--nodes="$SLURM_JOB_NUM_NODES" \
--ntasks-per-node=1 \
--export=ALL,SANDBOX_PORTS_DIR=$SANDBOX_PORTS_DIR,LISTEN_PORT=${SANDBOX_PORT:-6000},NGINX_PORT=${SANDBOX_PORT:-6000} \
bash -c "$SANDBOX_COMMAND" &
SRUN_PIDS["sandbox"]=$!
echo "[INFO] Sandbox started in background (PID: ${SRUN_PIDS["sandbox"]})"
else
echo "[INFO] SANDBOX_CONTAINER or SANDBOX_COMMAND not defined, skipping sandbox startup"
fi

NUM_ACTORS=$((gpus_per_node * SLURM_JOB_NUM_NODES))

# Start Ray worker nodes
Expand Down Expand Up @@ -495,4 +522,4 @@ EOF
echo " bash $CLUSTER_DIR/scripts/${SLURM_JOB_ID}-attach.sh 1 # to attach to worker 1"
echo " bash $CLUSTER_DIR/scripts/${SLURM_JOB_ID}-attach.sh 2 # to attach to worker 2, etc."
sleep infinity
fi
fi
27 changes: 27 additions & 0 deletions test/core/execution/artifacts/expected_ray_cluster_ssh.sub
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -295,6 +295,33 @@ while ! (srun --overlap --nodes=1 --ntasks=1 -w $head_node test -f $LOG_DIR/STAR
elapsed_time=$((elapsed_time + 2))
done

# Run sandbox in parallel across all allocated nodes when explicitly configured.
if [[ -n "${SANDBOX_CONTAINER:-}" ]] && [[ -n "${SANDBOX_COMMAND:-}" ]]; then
SANDBOX_PORTS_DIR="${SANDBOX_PORTS_DIR:-$LOG_DIR/sandbox}"
mkdir -p "$SANDBOX_PORTS_DIR"
echo "[INFO] Starting sandbox across all nodes in parallel (ports_dir=$SANDBOX_PORTS_DIR)..."
srun --output "$LOG_DIR/sandbox.log" \
--error "$LOG_DIR/sandbox.log" \
--container-image="$SANDBOX_CONTAINER" \
--container-mounts="$SANDBOX_PORTS_DIR:$SANDBOX_PORTS_DIR" \
--container-env=SANDBOX_PORTS_DIR,LISTEN_PORT,NGINX_PORT \
--no-container-mount-home \
--mpi=pmix \
-A "$SLURM_JOB_ACCOUNT" \
-p "$SLURM_JOB_PARTITION" \
--wait=60 \
--kill-on-bad-exit=1 \
--overlap \
--nodes="$SLURM_JOB_NUM_NODES" \
--ntasks-per-node=1 \
--export=ALL,SANDBOX_PORTS_DIR=$SANDBOX_PORTS_DIR,LISTEN_PORT=${SANDBOX_PORT:-6000},NGINX_PORT=${SANDBOX_PORT:-6000} \
bash -c "$SANDBOX_COMMAND" &
SRUN_PIDS["sandbox"]=$!
echo "[INFO] Sandbox started in background (PID: ${SRUN_PIDS["sandbox"]})"
else
echo "[INFO] SANDBOX_CONTAINER or SANDBOX_COMMAND not defined, skipping sandbox startup"
fi

NUM_ACTORS=$((gpus_per_node * SLURM_JOB_NUM_NODES))

# Start Ray worker nodes
Expand Down
27 changes: 27 additions & 0 deletions test/core/execution/artifacts/expected_ray_het_cluster.sub
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -319,6 +319,33 @@ while ! (srun --het-group=0 --overlap --nodes=1 --ntasks=1 -w $head_node test -f
elapsed_time=$((elapsed_time + 2))
done

# Run sandbox in parallel across all allocated nodes when explicitly configured.
if [[ -n "${SANDBOX_CONTAINER:-}" ]] && [[ -n "${SANDBOX_COMMAND:-}" ]]; then
SANDBOX_PORTS_DIR="${SANDBOX_PORTS_DIR:-$LOG_DIR/sandbox}"
mkdir -p "$SANDBOX_PORTS_DIR"
echo "[INFO] Starting sandbox across all nodes in parallel (ports_dir=$SANDBOX_PORTS_DIR)..."
srun --het-group=0 --output "$LOG_DIR/sandbox.log" \
--error "$LOG_DIR/sandbox.log" \
--container-image="$SANDBOX_CONTAINER" \
--container-mounts="$SANDBOX_PORTS_DIR:$SANDBOX_PORTS_DIR" \
--container-env=SANDBOX_PORTS_DIR,LISTEN_PORT,NGINX_PORT \
--no-container-mount-home \
--mpi=pmix \
-A "$SLURM_JOB_ACCOUNT" \
-p "$SLURM_JOB_PARTITION" \
--wait=60 \
--kill-on-bad-exit=1 \
--overlap \
--nodes="$SLURM_JOB_NUM_NODES" \
--ntasks-per-node=1 \
--export=ALL,SANDBOX_PORTS_DIR=$SANDBOX_PORTS_DIR,LISTEN_PORT=${SANDBOX_PORT:-6000},NGINX_PORT=${SANDBOX_PORT:-6000} \
bash -c "$SANDBOX_COMMAND" &
SRUN_PIDS["sandbox"]=$!
echo "[INFO] Sandbox started in background (PID: ${SRUN_PIDS["sandbox"]})"
else
echo "[INFO] SANDBOX_CONTAINER or SANDBOX_COMMAND not defined, skipping sandbox startup"
fi

NUM_ACTORS=$((gpus_per_node * SLURM_JOB_NUM_NODES))

# Start Ray worker nodes
Expand Down
7 changes: 5 additions & 2 deletions test/run/ray/test_slurm_ray_request.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -285,8 +285,11 @@ def test_container_configurations(self):

# Should use cluster_dir as default workdir
assert "--container-workdir=/tmp/test_jobs/test-ray-cluster" in script
# Should not contain container-image flag when none specified
assert "--container-image" not in script
# The main Ray cluster script should not contain a container-image flag
# when none is specified on the executor. Ignore the optional sandbox
# stanza, which may carry its own container-image placeholder.
pre_sandbox_script = script.split("# Run sandbox", 1)[0]
assert "--container-image" not in pre_sandbox_script

def test_special_mount_handling(self):
"""Test materialize handles special RUNDIR_SPECIAL_NAME mounts."""
Expand Down
Loading