[cluster] Add a lightweight container launcher

projects/rocHPL/run-docker.sh → projects/cluster/docker-cluster-up.sh

 #!/bin/bash
 set -e
 
-IMAGE_NAME=harbor.sourcefind.cn:5443/dcu/admin/base/vllm:0.11.0-ubuntu22.04-dtk26.04-0130-py3.10-20260204
-CONTAINER_NAME=rochpl-dtk26-20260204
-SSH_PORT=3333           # 确保端口可用
-WORKSPACE_DIR="${PWD}"  # 修改为主节点的rocHPL目录
-
 # =====================================================================
 # 节点映射表
 # 1. 当前节点是主节点，WORKER_CONFIG不包括主节点
@@ -19,6 +14,44 @@ node3    node03
 master   node04
 "
 
+# =====================================================================
+# 默认值
+# =====================================================================
+IMAGE_NAME=harbor.sourcefind.cn:5443/dcu/admin/base/vllm:0.11.0-ubuntu22.04-dtk26.04-0130-py3.10-20260204
+CONTAINER_NAME=cluster-dtk26-20260204
+SSH_PORT=3333     # 确保端口可用
+WORKDIR="${PWD}"  # 修改为主节点的rocHPL/rocHPCG等应用目录
+FORCE_RM=0
+SKIP_PULL=0
+
+usage() {
+  echo "Usage: $0 [-f|--force] [-i|--image IMAGE] [-n|--name NAME] [-p|--port PORT] [-w|--workdir WORKDIR]"
+  echo "  -f, --force              Force remove existing container before starting"
+  echo "  -i, --image    IMAGE     Docker image name (default: ${IMAGE_NAME})"
+  echo "  -n, --name     NAME      Container name across all nodes (default: ${CONTAINER_NAME})"
+  echo "  -p, --port     PORT      SSH port inside container (default: ${SSH_PORT})"
+  echo "  -w, --workdir  WORKDIR   Workspace directory to mount for the master node (default: \$PWD)"
+  echo "  --no-pull                Skip pulling the image"
+  echo "  -h, --help               Show this help message"
+  exit 1
+}
+
+OPTS=$(getopt -o fi:n:p:w: --long force,image:,name:,port:,workdir:,no-pull -n "$0" -- "$@") || usage
+eval set -- "${OPTS}"
+
+while true; do
+  case "$1" in
+    -f|--force)      FORCE_RM=1; shift ;;
+    -i|--image)      IMAGE_NAME="$2"; shift 2 ;;
+    -n|--name)       CONTAINER_NAME="$2"; shift 2 ;;
+    -p|--port)       SSH_PORT="$2"; shift 2 ;;
+    -w|--workdir)    WORKDIR="$2"; shift 2 ;;
+    --no-pull)       SKIP_PULL=1; shift ;;
+    --)              shift; break ;;
+    *)               usage ;;
+  esac
+done
+
 # =====================================================================
 # 解析hosts
 # =====================================================================
@@ -46,7 +79,7 @@ resolve_ip() {
 
 MASTER_IP=$(resolve_ip ${MASTER_NODE})
 DOCKER_ADD_HOSTS="--add-host ${DOCKER_MASTER}:${MASTER_IP}"
-echo "Master node: ${MASTER_NODE} -> ${DOCKER_MASTER} (${MASTER_IP})"
+echo "[INFO] Master node: ${MASTER_NODE} -> ${DOCKER_MASTER} (${MASTER_IP})"
 
 WORKER_NODE_ARR=()
 while read -r phys_host std_name rest; do
@@ -54,14 +87,14 @@ while read -r phys_host std_name rest; do
   
   IP=$(resolve_ip "$phys_host")
   if [ -z "$IP" ]; then
-    echo "Error: failed to resolve IP for worker node '$phys_host'!"
+    echo "[Error] Failed to resolve IP for worker node '$phys_host'!"
     exit 1
   fi
   
   DOCKER_ADD_HOSTS="${DOCKER_ADD_HOSTS} --add-host ${std_name}:${IP}"
   WORKER_NODE_ARR+=("$phys_host")
   
-  echo "Worker node: ${phys_host} -> ${std_name} (${IP})"
+  echo "[INFO] Worker node: ${phys_host} -> ${std_name} (${IP})"
 done <<< "$WORKER_CONFIG"
 
 WORKER_NODES=$(IFS=,; echo "${WORKER_NODE_ARR[*]}")
@@ -72,18 +105,20 @@ ALL_NODES="${MASTER_NODE}"
 # =====================================================================
 # 启动容器
 # =====================================================================
-echo "[1/3] Pulling the image on each node..."
-pdsh -w ${ALL_NODES} -S "docker pull ${IMAGE_NAME} >/dev/null"
+if [ "${SKIP_PULL}" == "1" ]; then
+  echo "[INFO] Image pulling skipped."
+else
+  echo "[INFO] Pulling the image on each node..."
+  pdsh -w ${ALL_NODES} -S "docker pull ${IMAGE_NAME} >/dev/null"
+fi
 
-if [ "$1" == "-f" ]; then
-  echo "[2/3] Force cleaning up old container..."
+if [ "${FORCE_RM}" == "1" ]; then
+  echo "[INFO] Force removing existing containers..."
   pdsh -w ${ALL_NODES} -S "docker rm -f ${CONTAINER_NAME} 2>/dev/null || true"
-else
-  echo "[2/3] Skipping cleanup."
 fi
 
-echo "[3/3] Starting containers..."
-mkdir -p ${WORKSPACE_DIR}
+echo "[INFO] Starting docker containers..."
+mkdir -p ${WORKDIR}
 DOCKER_ARGS="--name=${CONTAINER_NAME} \
   -v /opt/hyhal:/opt/hyhal:ro \
   -v /root/.ssh:/root/.ssh \
@@ -103,7 +138,7 @@ DOCKER_ARGS="--name=${CONTAINER_NAME} \
 
 docker run -itd \
   ${DOCKER_ARGS} \
-  -v ${WORKSPACE_DIR}:/workspace \
+  -v ${WORKDIR}:/workspace \
   ${IMAGE_NAME} \
   bash -c "mkdir -p /run/sshd && /usr/sbin/sshd -p ${SSH_PORT}; sleep infinity"
 
@@ -114,6 +149,6 @@ if [ -n "$WORKER_NODES" ]; then
     bash -c 'mkdir -p /run/sshd && /usr/sbin/sshd -p ${SSH_PORT}; sleep infinity'"
 fi
   
-echo "All containers are ready!"
-echo "To access the container on the current node, run:"
-echo "docker exec -it ${CONTAINER_NAME} bash"
+echo "[INFO] All containers are ready!"
+echo "[INFO] To access the container on the current node, run:"
+echo "       docker exec -it ${CONTAINER_NAME} bash"