fix process control script

2026-03-23 16:11:17 +08:00
parent 0acc58bbe1
commit 6f7ee762ec
1 changed files with 257 additions and 24 deletions
--- a/set_process.sh
+++ b/set_process.sh
@@ -8,9 +8,216 @@

 #前置all的重置或者新增都是由稳态的第一个进程来处理，所有进程收到这条消息后先判断自己的进程号是否是1，而且是稳态，否则不处理，所有操作均由这个进程完成，

+if [ -z "$SETSID" ]; then
+    export SETSID=1
+    nohup setsid "$0" "$@" >> /tmp/set_process_detach.log 2>&1 < /dev/null &
+    exit 0
+fi
+
+# 关闭从父进程继承来的 socket / pipe / 文件描述符
+for fd_path in /proc/$$/fd/*; do
+    fd_num=$(basename "$fd_path")
+    case "$fd_num" in
+        0|1|2) ;;
+        *) eval "exec ${fd_num}>&-" 2>/dev/null ;;
+    esac
+done
+
 # 设置日志文件路径
 LOGFILE="$FEP_ENV/dat/log/start_fe.log"

+INI_FILE="/FeProject/etc/config/mykafka.ini"
+
+LOCK_FILE="/tmp/set_process.lock"
+
+if [ -f "$LOCK_FILE" ]; then
+    old_pid=$(cat "$LOCK_FILE")
+    if ps -p "$old_pid" > /dev/null 2>&1; then
+        echo "Already running: $old_pid"
+        exit 1
+    else
+        echo "Stale lock found, removing"
+        rm -f "$LOCK_FILE"
+    fi
+fi
+
+echo $$ > "$LOCK_FILE"
+trap "rm -f $LOCK_FILE" EXIT
+
+get_ini_value() {
+    local key="$1"
+    local line
+
+    line=$(grep -E "^[[:space:]]*${key}=" "$INI_FILE" | tail -n 1)
+    [ -z "$line" ] && return 1
+
+    # 去掉 key=
+    line="${line#*=}"
+    # 去掉首尾空格
+    line=$(echo "$line" | sed 's/^[[:space:]]*//; s/[[:space:]]*$//')
+
+    [ -n "$line" ] || return 1
+    echo "$line"
+    return 0
+}
+
+log() {
+    echo "$@" | tee -a "$LOGFILE"
+}
+
+check_local_port_released() {
+    local PORT="$1"
+
+    for retry in $(seq 1 3); do
+        for i in $(seq 1 20); do
+            if ! ss -lntp 2>/dev/null | grep -q ":${PORT}[[:space:]]"; then
+                log "[OK] Local port $PORT released"
+                return 0
+            fi
+            log "[WAIT] Local port $PORT still in use... ($i/20)"
+            sleep 1
+        done
+
+        log "[RETRY] Local port $PORT not released, retry $retry/3"
+        sleep 1
+    done
+
+    log "[FAIL] Local port $PORT still in use after retries:"
+    ss -lntp 2>/dev/null | grep ":${PORT}[[:space:]]" | tee -a "$LOGFILE"
+
+    return 1
+}
+
+check_remote_conn_released() {
+    local REMOTE_IP="$1"
+    local REMOTE_PORT="$2"
+
+    for retry in $(seq 1 3); do
+        for i in $(seq 1 20); do
+            if ! ss -ntp 2>/dev/null \
+                | grep "${REMOTE_IP}:${REMOTE_PORT}" \
+                | grep -E 'pt61850netd_pqfe|fe_watchdog|fe_main' >/dev/null; then
+                log "[OK] Remote ${REMOTE_IP}:${REMOTE_PORT} released"
+                return 0
+            fi
+
+            log "[WAIT] Remote ${REMOTE_IP}:${REMOTE_PORT} still exists... ($i/20)"
+            ss -ntp 2>/dev/null \
+                | grep "${REMOTE_IP}:${REMOTE_PORT}" \
+                | grep -E 'pt61850netd_pqfe|fe_watchdog|fe_main' \
+                | tee -a "$LOGFILE"
+            sleep 1
+        done
+
+        log "[RETRY] Remote ${REMOTE_IP}:${REMOTE_PORT} not released, retry $retry/3"
+        sleep 1
+    done
+
+    log "[FAIL] Remote ${REMOTE_IP}:${REMOTE_PORT} still exists:"
+    ss -ntp 2>/dev/null \
+        | grep "${REMOTE_IP}:${REMOTE_PORT}" \
+        | grep -E 'pt61850netd_pqfe|fe_watchdog|fe_main' \
+        | tee -a "$LOGFILE"
+
+    return 1
+}
+
+check_value_released() {
+    local key="$1"
+    local value="$2"
+
+    [ -z "$value" ] && return 0
+
+    # 纯端口
+    if echo "$value" | grep -Eq '^[0-9]+$'; then
+        [ "$value" = "0" ] && { log "[SKIP] $key disabled"; return 0; }
+
+        log "[CHECK] $key local port: $value"
+        check_local_port_released "$value"
+        return $?
+    fi
+
+    # URL
+    if echo "$value" | grep -Eq '^[a-zA-Z]+://'; then
+        local hostport ip port
+        hostport=$(echo "$value" | sed -n 's#^[a-zA-Z]\+://\([^/]*\).*#\1#p')
+        ip="${hostport%%:*}"
+        port="${hostport##*:}"
+
+        if [ -n "$ip" ] && [ -n "$port" ] && [ "$ip" != "$port" ]; then
+            log "[CHECK] $key remote: $ip:$port"
+            check_remote_conn_released "$ip" "$port"
+            return $?
+        fi
+
+        log "[SKIP] $key invalid URL: $value"
+        return 0
+    fi
+
+    # ip:port
+    if echo "$value" | grep -Eq '^[^:]+:[0-9]+$'; then
+        local ip="${value%%:*}"
+        local port="${value##*:}"
+
+        log "[CHECK] $key remote: $ip:$port"
+        check_remote_conn_released "$ip" "$port"
+        return $?
+    fi
+
+    log "[SKIP] $key unsupported value: $value"
+    return 0
+}
+
+check_key_released() {
+    local key="$1"
+    local value
+
+    value=$(get_ini_value "$key") || {
+        log "[SKIP] $key not found"
+        return 0
+    }
+
+    log "==== Checking $key = $value ===="
+    check_value_released "$key" "$value"
+    return $?
+}
+
+check_all_resources_released() {
+    local ret=0
+
+    log "=============================="
+    log " Start checking resources..."
+    log "=============================="
+
+    # 本地端口
+    check_key_released "TestPort" || ret=1
+    check_key_released "HttpPort" || ret=1
+    check_key_released "SocketPort" || ret=1
+
+    # MQ
+    check_key_released "Ipport" || ret=1
+    check_key_released "ConsumerIpport" || ret=1
+
+    # Web
+    check_key_released "WebDevice" || ret=1
+    check_key_released "WebIcd" || ret=1
+    check_key_released "WebIntegrity" || ret=1
+    check_key_released "WebComflag" || ret=1
+    check_key_released "WebEvent" || ret=1
+    check_key_released "WebFileupload" || ret=1
+    check_key_released "WebFiledownload" || ret=1
+
+    if [ $ret -eq 0 ]; then
+        log "✅ ALL resources released"
+    else
+        log "❌ Some resources NOT released"
+    fi
+
+    return $ret
+}
+
+
+
 # 输出当前时间并打印进程停止信息
 echo "" ; echo "" 
 echo "****** `date "+%F %R:%S"` start setting Processes after 3 sec ******"
@@ -20,7 +227,7 @@ echo "****** `date "+%F %R:%S"` start setting Processes after 3 sec ******" >>"$

 # 函数检查并处理日志文件大小
 check_log_file() {
-    if [ -n "$1" ]; then
+    if [ -n "$1" ] && [ -f "$1" ]; then
        FILE_SIZE=0
        FILE_SIZE=$(du "$1" | awk '{print $1}')
        
@@ -45,25 +252,46 @@ check_log_file $LOGFILE

 # 定义查找并杀死进程的函数
 kill_process_by_name() {
-    PROCESS_NAME=$1
-    PID=$(ps -ef | grep "$PROCESS_NAME" | grep -v "grep" | awk '{print $2}')
+    local PROCESS_NAME="$1"
+    local PIDS

-    if [ -n "$PID" ]; then
-        echo "Found process '$PROCESS_NAME' with PID: $PID"
+    PIDS=$(ps -ef | grep "$PROCESS_NAME" | grep -v grep | awk '{print $2}')
+
+    if [ -n "$PIDS" ]; then
+        echo "Found process '$PROCESS_NAME' with PID(s): $PIDS"
        echo "Killing process..."
-        kill -15 $PID
+
+        for pid in $PIDS; do
+            kill -15 "$pid" 2>/dev/null
+        done
+
        sleep 3

-        PID2=$(ps -ef | grep "$PROCESS_NAME" | grep -v "grep" | awk '{print $2}')
-        if [ -n "$PID2" ]; then
-            echo "Process still exists, force kill: $PID2"
-            kill -9 $PID2
-        fi
+        for pid in $PIDS; do
+            if ps -p "$pid" >/dev/null 2>&1; then
+                echo "Process still exists, force kill: $pid"
+                kill -9 "$pid" 2>/dev/null
+            fi
+        done
    else
        echo "Process '$PROCESS_NAME' not found."
    fi
 }

+wait_all_exit() {
+    for i in $(seq 1 30); do
+        COUNT=$(ps -ef | grep -E 'pt61850netd_pqfe|fe_watchdog' | grep -v grep | wc -l)
+        if [ "$COUNT" -eq 0 ]; then
+            echo "All FE processes exited"
+            return 0
+        fi
+        echo "Waiting FE processes exit... ($COUNT still running)"
+        sleep 1
+    done
+    echo "Timeout waiting FE processes exit"
+    return 1
+}
+
 # 功能块开始
 handle_reset() {
    # 功能：reset
@@ -73,22 +301,28 @@ handle_reset() {
 		if [ "$2" == "all" ]; then
 		
 			# 关闭旧的看门狗进程
-			kill_process_by_name "/FeProject/bin/fe_watchdog"
+			#kill_process_by_name "/FeProject/bin/fe_watchdog"
 			
 			# 关闭旧的 stat 进程
-			kill_process_by_name "/FeProject/bin/pt61850netd_pqfe -d cfg_stat_data"
+			#kill_process_by_name "/FeProject/bin/pt61850netd_pqfe -d cfg_stat_data"
 			
 			# 关闭旧的 recall 进程
-			kill_process_by_name "/FeProject/bin/pt61850netd_pqfe -d cfg_recallhis_data"
+			#kill_process_by_name "/FeProject/bin/pt61850netd_pqfe -d cfg_recallhis_data"
 			
 			# 关闭旧的 3s 进程
-			kill_process_by_name "/FeProject/bin/pt61850netd_pqfe -d cfg_3s_data"
+			#kill_process_by_name "/FeProject/bin/pt61850netd_pqfe -d cfg_3s_data"
 			
 			# 关闭旧的 comtrade 进程
-			kill_process_by_name "/FeProject/bin/pt61850netd_pqfe -d cfg_soe_comtrade"
+			#kill_process_by_name "/FeProject/bin/pt61850netd_pqfe -d cfg_soe_comtrade"
+			
+			/home/pq/FeProject/boot/stop_fe.sh
+
+			wait_all_exit || exit 1
+			
+			check_all_resources_released || exit 1
 			
 			#关闭进程后等待一段时间，防止端口占用
-			sleep 1
+			sleep 5
 		
 			# 清空 runtime.cf 中的所有进程配置
 			sed -i '/cfg_stat_data/d' /home/pq/FeProject/etc/runtime.cf
@@ -235,14 +469,13 @@ handle_add() {
 }

 # 获取当前脚本的进程ID
-CURRENT_PID=$$
-
+#CURRENT_PID=$$
 # 检查是否有其他的set_process.sh脚本正在运行，排除当前脚本
-if pgrep -f "set_process.sh" | grep -v "^$CURRENT_PID$" > /dev/null; then
-    echo "set_process.sh is already running. Exiting..."
-    echo "set_process.sh is already running. Exiting..." >>"$LOGFILE"
-    exit 1
-fi
+#if pgrep -f "set_process.sh" | grep -v "^$CURRENT_PID$" > /dev/null; then
+#    echo "set_process.sh is already running. Exiting..."
+#    echo "set_process.sh is already running. Exiting..." >>"$LOGFILE"
+#    exit 1
+#fi

 #脚本应该等待3秒钟
 sleep 3