[cluster] Update hostname in container

projects/cluster/docker-cluster-up.sh

@@ -3,15 +3,14 @@ set -e
 
 # =====================================================================
 # 节点映射表
-# 1. 当前节点是主节点，WORKER_CONFIG不包括主节点
+# 1. 当前节点是主节点，会挂载一个工作目录
 # 2. 第一列是物理机hostname或IP，第二列是docker容器hostname
 # =====================================================================
-MASTER_NODE=$(hostname)
-DOCKER_MASTER=node01
-WORKER_CONFIG="
-node1    node02
-node2    node03
-node3    node04
+CLUSTER_CONFIG="
+$(hostname)    node01
+node1          node02
+node2          node03
+node3          node04
 "
 
 # =====================================================================
@@ -78,25 +77,47 @@ resolve_ip() {
   echo "$ip"
 }
 
-MASTER_IP=$(resolve_ip ${MASTER_NODE})
-DOCKER_ADD_HOSTS="--add-host ${DOCKER_MASTER}:${MASTER_IP}"
-echo "[INFO] Master node: ${MASTER_NODE} -> ${DOCKER_MASTER} (${MASTER_IP})"
-
+MASTER_NODE=$(hostname)
+DOCKER_MASTER=""
+MASTER_IP=""
+DOCKER_ADD_HOSTS=""
+MAPPING_STR=""
 WORKER_NODE_ARR=()
+
+# 读取hostname映射表
 while read -r phys_host std_name rest; do
   [[ -z "$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'!"
+    echo "[Error] Failed to resolve IP for node '$phys_host'!"
     exit 1
   fi
   
+  # 所有节点的主机记录都需要加上
   DOCKER_ADD_HOSTS="${DOCKER_ADD_HOSTS} --add-host ${std_name}:${IP}"
-  WORKER_NODE_ARR+=("$phys_host")
   
-  echo "[INFO] Worker node: ${phys_host} -> ${std_name} (${IP})"
-done <<< "$WORKER_CONFIG"
+  if [ -z "$MAPPING_STR" ]; then
+    MAPPING_STR="${phys_host}:${std_name}"
+  else
+    MAPPING_STR="${MAPPING_STR},${phys_host}:${std_name}"
+  fi
+
+  # 判断是否是本物理机（主节点）
+  if [ "$phys_host" = "$MASTER_NODE" ]; then
+    DOCKER_MASTER=$std_name
+    MASTER_IP=$IP
+    echo "[INFO] Master node: ${phys_host} -> ${std_name} (${IP})"
+  else
+    WORKER_NODE_ARR+=("$phys_host")
+    echo "[INFO] Worker node: ${phys_host} -> ${std_name} (${IP})"
+  fi
+done <<< "$CLUSTER_CONFIG"
+
+if [ -z "$DOCKER_MASTER" ]; then
+  echo "[Error] Current node '$MASTER_NODE' is not found in CLUSTER_CONFIG!"
+  exit 1
+fi
 
 WORKER_NODES=$(IFS=,; echo "${WORKER_NODE_ARR[*]}")
 
@@ -121,6 +142,7 @@ fi
 echo "[INFO] Starting docker containers..."
 mkdir -p ${WORKDIR}
 DOCKER_ARGS="--name=${CONTAINER_NAME} \
+  -e NODE_MAPPING=${MAPPING_STR} \
   -v /opt/hyhal:/opt/hyhal:ro \
   -v /root/.ssh:/root/.ssh \
   -w /workspace \
@@ -141,13 +163,13 @@ docker run -itd \
   ${DOCKER_ARGS} \
   -v ${WORKDIR}:/workspace \
   ${IMAGE_NAME} \
-  bash -c "mkdir -p /run/sshd && /usr/sbin/sshd -p ${SSH_PORT}; sleep infinity"
+  bash -c "hostname ${DOCKER_MASTER} && 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'"
+    bash -c 'PHYS=\$(hostname); for m in \${NODE_MAPPING//,/ }; do [ \"\${m%%:*}\" = \"\$PHYS\" ] && hostname \${m##*:} && break; done; mkdir -p /run/sshd && /usr/sbin/sshd -p ${SSH_PORT}; sleep infinity'"
 fi
   
 echo "[INFO] All containers are ready!"