software-engineering/spark-vllm-docker
2
0
Fork 0
Code Issues Actions Packages Projects Releases Wiki Activity

Fixed a bug when solo mode failed on a standalone Spark without configured RoCE.

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2026-01-30 16:39:11 -08:00
parent 57c890b10c
commit 4a4b4e7610
1 changed files with 8 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
launch-cluster.sh
View File

@@ -144,24 +144,22 @@ for i in "${!MOD_PATHS[@]}"; do
done done
# --- Auto-Detection Logic --- # --- Auto-Detection Logic ---
# Source autodiscover module # Source autodiscover module
source "$(dirname "$0")/autodiscover.sh" source "$(dirname "$0")/autodiscover.sh"
# Perform auto-detection
detect_interfaces || exit 1
if [[ "$SOLO_MODE" == "true" ]]; then if [[ "$SOLO_MODE" == "true" ]]; then
if [[ -n "$NODES_ARG" ]]; then if [[ -n "$NODES_ARG" ]]; then
echo "Error: --solo is incompatible with -n/--nodes." echo "Error: --solo is incompatible with -n/--nodes."
exit 1 exit 1
fi fi
# Solo mode: skip node detection, just get local IP # Solo mode: skip node detection, just get local IP
detect_local_ip || exit 1 LOCAL_IP="127.0.0.1"
NODES_ARG="$LOCAL_IP" NODES_ARG="$LOCAL_IP"
PEER_NODES=() PEER_NODES=()
echo "Solo mode enabled. Skipping node detection." echo "Solo mode enabled. Skipping node detection."
else else
# Perform auto-detection
detect_interfaces || exit 1
detect_nodes || exit 1 detect_nodes || exit 1
fi fi
@@ -173,8 +171,11 @@ fi
# Split nodes into array # Split nodes into array
IFS=',' read -r -a ALL_NODES <<< "$NODES_ARG" IFS=',' read -r -a ALL_NODES <<< "$NODES_ARG"
# Detect Head IP (Local IP) if [[ "$SOLO_MODE" != "true" ]]; then
detect_local_ip || exit 1 # Detect Head IP (Local IP)
detect_local_ip || exit 1
fi
HEAD_IP="$LOCAL_IP" HEAD_IP="$LOCAL_IP"
# Verify HEAD_IP is in ALL_NODES # Verify HEAD_IP is in ALL_NODES