[rocHPL] Add a script to run containers on multiple nodes

projects/rocHPL/run-docker.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="${PWD}"  # 修改为主节点的rocHPL目录
+
+# =====================================================================
+# 节点映射表
+# 1. 当前节点是主节点，WORKER_CONFIG不包括主节点
+# 2. 第一列是物理机hostname或IP，第二列是docker容器hostname
+# =====================================================================
+MASTER_NODE=$(hostname)
+WORKER_CONFIG="
+node2    node02
+node3    node03
+master   node04
+"
+
+# =====================================================================
+# 解析hosts
+# =====================================================================
+resolve_ip() {
+  local target=$1
+  local ip=""
+
+  # Try getent
+  if command -v getent >/dev/null 2>&1; then
+    ip=$(getent ahosts "$target" 2>/dev/null | awk '{print $1}' | head -n 1)
+  fi
+
+  # Try ping
+  if [ -z "$ip" ] && command -v ping >/dev/null 2>&1; then
+    ip=$(ping -c 1 -n "$target" 2>/dev/null | awk -F'[()]' '/PING/{print $2}')
+  fi
+
+  # Fallback to /etc/hosts
+  if [ -z "$ip" ]; then
+    ip=$(awk -v h="$target" '!/^#/ && ($2 == h || $3 == h) {print $1; exit}' /etc/hosts)
+  fi
+
+  echo "$ip"
+}
+
+MASTER_IP=$(resolve_ip ${MASTER_NODE})
+DOCKER_ADD_HOSTS="--add-host node01:${MASTER_IP}"
+echo "Master node: ${MASTER_NODE} -> node01 (${MASTER_IP})"
+
+WORKER_NODE_ARR=()
+while read -r phys_host std_name rest; do
+  [[ -z "$phys_host" || "$phys_host" == \#* ]] && continue
+  
+  IP=$(resolve_ip "$phys_host")
+  if [ -z "$IP" ]; then
+    echo "Error: failed to resolve IP for worker node '$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})"
+done <<< "$WORKER_CONFIG"
+
+WORKER_NODES=$(IFS=,; echo "${WORKER_NODE_ARR[*]}")
+
+ALL_NODES="${MASTER_NODE}"
+[ -n "$WORKER_NODES" ] && ALL_NODES="${ALL_NODES},${WORKER_NODES}"
+
+# =====================================================================
+# 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 [ "$1" == "-f" ]; then
+  echo "[2/3] Force cleaning up old container..."
+  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 ${WORKSPACE_DIR}
+DOCKER_ARGS="--name=${CONTAINER_NAME} \
+  -v /opt/hyhal:/opt/hyhal:ro \
+  -v /root/.ssh:/root/.ssh \
+  -w /workspace \
+  ${DOCKER_ADD_HOSTS} \
+  --network=host \
+  --ipc=host \
+  --device=/dev/kfd \
+  --device=/dev/mkfd \
+  --device=/dev/dri \
+  --shm-size=512G \
+  --privileged \
+  --group-add video \
+  --cap-add=SYS_PTRACE \
+  -u root \
+  --security-opt seccomp=unconfined"
+
+docker run -itd \
+  ${DOCKER_ARGS} \
+  -v ${WORKSPACE_DIR}:/workspace \
+  ${IMAGE_NAME} \
+  bash -c "mkdir -p /run/sshd && /usr/sbin/sshd -p ${SSH_PORT}; sleep infinity"
+
+if [ -n "$WORKER_NODES" ]; then
+  pdsh -w ${WORKER_NODES} -S "docker run -itd \
+    ${DOCKER_ARGS} \
+    ${IMAGE_NAME} \
+    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"