Skip to content

Commit 5115edc

Browse files
committed
Docker queueing same time
1 parent 229b7a8 commit 5115edc

2 files changed

Lines changed: 34 additions & 0 deletions

File tree

runners/launch_mi300x-amd.sh

Lines changed: 17 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,9 +7,26 @@ PORT=8888
77

88
server_name="bmk-server"
99

10+
# Ensure non-root container user can access AMD GPU devices (/dev/kfd, /dev/dri)
11+
# by inheriting the host device group IDs (render/video). File permissions are
12+
# not overridden by --privileged, so group membership is required.
13+
KFD_GID=$(stat -c '%g' /dev/kfd 2>/dev/null || true)
14+
RENDER_NODE=$(ls -1 /dev/dri/renderD* 2>/dev/null | head -n1 || true)
15+
CARD_NODE=$(ls -1 /dev/dri/card* 2>/dev/null | head -n1 || true)
16+
RENDER_GID=$([[ -n "$RENDER_NODE" ]] && stat -c '%g' "$RENDER_NODE" 2>/dev/null || true)
17+
VIDEO_GID=$([[ -n "$CARD_NODE" ]] && stat -c '%g' "$CARD_NODE" 2>/dev/null || true)
18+
19+
GROUP_ARGS=""
20+
for gid in "$KFD_GID" "$RENDER_GID" "$VIDEO_GID"; do
21+
if [[ -n "$gid" ]]; then
22+
GROUP_ARGS+=" --group-add $gid"
23+
fi
24+
done
25+
1026
set -x
1127
docker run --rm --ipc=host --shm-size=16g --network=host --name=$server_name \
1228
--user $(id -u):$(id -g) \
29+
${GROUP_ARGS} \
1330
--privileged --cap-add=CAP_SYS_ADMIN --device=/dev/kfd --device=/dev/dri --device=/dev/mem \
1431
--cap-add=SYS_PTRACE --security-opt seccomp=unconfined \
1532
-v $HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE \

runners/launch_mi300x-cr.sh

Lines changed: 17 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,9 +7,26 @@ PORT=8888
77

88
server_name="bmk-server"
99

10+
# Ensure non-root container user can access AMD GPU devices (/dev/kfd, /dev/dri)
11+
# by inheriting the host device group IDs (render/video). File permissions are
12+
# not overridden by --privileged, so group membership is required.
13+
KFD_GID=$(stat -c '%g' /dev/kfd 2>/dev/null || true)
14+
RENDER_NODE=$(ls -1 /dev/dri/renderD* 2>/dev/null | head -n1 || true)
15+
CARD_NODE=$(ls -1 /dev/dri/card* 2>/dev/null | head -n1 || true)
16+
RENDER_GID=$([[ -n "$RENDER_NODE" ]] && stat -c '%g' "$RENDER_NODE" 2>/dev/null || true)
17+
VIDEO_GID=$([[ -n "$CARD_NODE" ]] && stat -c '%g' "$CARD_NODE" 2>/dev/null || true)
18+
19+
GROUP_ARGS=""
20+
for gid in "$KFD_GID" "$RENDER_GID" "$VIDEO_GID"; do
21+
if [[ -n "$gid" ]]; then
22+
GROUP_ARGS+=" --group-add $gid"
23+
fi
24+
done
25+
1026
set -x
1127
docker run --rm --ipc=host --shm-size=16g --network=host --name=$server_name \
1228
--user $(id -u):$(id -g) \
29+
${GROUP_ARGS} \
1330
--privileged --cap-add=CAP_SYS_ADMIN --device=/dev/kfd --device=/dev/dri --device=/dev/mem \
1431
--cap-add=SYS_PTRACE --security-opt seccomp=unconfined \
1532
-v $HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE \

0 commit comments

Comments
 (0)