sprintfix: 修复多 Agent 场景下 crontab 越界移除的问题 (fixed TencentBlueKing#1698)

ZhuoZhuoCrayon · Jul 31, 2023 · 1259b63 · 1259b63
1 parent 39d0805
commit 1259b63
Show file tree

Hide file tree

Showing 5 changed files with 103 additions and 114 deletions.
diff --git a/script_tools/gsectl/agent/linux/gsectl b/script_tools/gsectl/agent/linux/gsectl
@@ -184,24 +184,24 @@ watch_by_binary () {
     # 获取当前时间的分钟数
     minute=$(date +%M)
 
+    # 如果文件存在，则读取文件中记录的次数
+    if [ -f $LAST_RUN_FILE ]; then
+        run_count=$(cat $LAST_RUN_FILE)
+    else
+        run_count=0
+    fi
+
     # 判断是否为整点时间
     if [ "$minute" == "00" ]; then
-        log "The current time is on the hour, reset the counter, and restart the detection."
-        if [ -f $LAST_RUN_FILE ];then
+        if [ -f $LAST_RUN_FILE -a $run_count -gt 0 ];then
+            log "The current time is on the hour, reset the counter $run_count -> 0, and restart the detection."
             echo 0 > $LAST_RUN_FILE
         fi
     fi
 
     # 设置告警阈值
     THRESHOLD=5
 
-    # 如果文件存在，则读取文件中记录的次数
-    if [ -f $LAST_RUN_FILE ]; then
-        run_count=$(cat $LAST_RUN_FILE)
-    else
-        run_count=0
-    fi
-
     # 检查上一次脚本是否存在
     if [ -f /var/run/gsectl_check_status.pid ]; then
         pid=`cat /var/run/gsectl_check_status.pid`
@@ -344,6 +344,7 @@ start_by_crontab () {
 }
 
 stop_by_crontab () {
+    remove_crontab
     stop_by_binary
     return
 }
@@ -502,31 +503,42 @@ get_os_type () {
     fi
 }
 
-add_startup_to_boot (){
+check_rc_file () {
+    get_os_type
+    if [ -f $RC_LOCAL_FILE ]; then
+        return 0
+    elif [ -f "/etc/rc.d/rc.local" ]; then
+        RC_LOCAL_FILE="/etc/rc.d/rc.local"
+    elif [ -f "/etc/init.d/rc.local" ]; then
+        RC_LOCAL_FILE="/etc/init.d/rc.local"
+    elif [ -f "/etc/init.d/boot.local" ]; then
+        RC_LOCAL_FILE="/etc/init.d/boot.local"
+    else
+        RC_LOCAL_FILE="`readlink -f /etc/rc.local`"
+    fi
+}
+
+add_startup_to_boot () {
+
     local module=agent
 
     # 添加启动项到 rc.local
-    echo "Check startup items, and if not existing, add the startup item to rc.local"
-    get_os_type
-    if [ -f $RC_LOCAL_FILE ];then
-        if [ `egrep "local/${INSTALL_ENV}/${node_type}" $RC_LOCAL_FILE |egrep ${module} |wc -l` -eq 0 ];then
-            echo "[ -f /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl ] && /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl start ${module} 1>>/var/log/${INSTALL_ENV}_${node_type}.log 2>&1" >> $RC_LOCAL_FILE
-        fi
-    else
-        if [ -f /etc/rc.d/rc.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/rc.d/rc.local |egrep ${module} |wc -l` -eq 0 ];then
-                echo "[ -f /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl ] && /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl start ${module} 1>>/var/log/${INSTALL_ENV}_${node_type}.log 2>&1" >> $RC_LOCAL_FILE
-            fi
-        elif [ -f /etc/init.d/rc.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/init.d/rc.local |egrep ${module} |wc -l` -eq 0 ];then
-                echo "[ -f /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl ] && /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl start ${module} 1>>/var/log/${INSTALL_ENV}_${node_type}.log 2>&1" >> $RC_LOCAL_FILE
-            fi
-        elif [ -f /etc/init.d/boot.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/init.d/boot.local |egrep ${module} |wc -l` -eq 0 ];then
-                echo "[ -f /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl ] && /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl start ${module} 1>>/var/log/${INSTALL_ENV}_${node_type}.log 2>&1" >> $RC_LOCAL_FILE
-            fi
-        fi
+    echo "Check startup items, and if not existing, add the [${module}] startup item to rc.local"
+
+    check_rc_file
+    local rcfile=$RC_LOCAL_FILE
+
+    if [ $OS_TYPE == "ubuntu" ]; then
+        sed -i "\|\#\!/bin/bash|d" $rcfile
+        sed -i "1i \#\!/bin/bash" $rcfile
     fi
+
+    chmod +x $rcfile
+
+    # 先删后加，避免重复
+    sed -i "\|${WORK_HOME}/bin/gsectl start ${module}|d" $rcfile
+
+    echo "[ -f ${WORK_HOME}/bin/gsectl ] && ${WORK_HOME}/bin/gsectl start ${module} 1>>/var/log/${INSTALL_ENV}_${node_type}.log 2>&1" >>$rcfile
 }
 
 add_config_to_systemd () {
@@ -541,11 +553,11 @@ After=network-online.target
 [Service]
 LimitNOFILE=512000
 LimitCORE=infinity
-WorkingDirectory=/usr/local/${INSTALL_ENV}/${node_type}/bin
-PIDFile=/usr/local/${INSTALL_ENV}/${node_type}/bin/run/${module}.pid
-ExecStart=/usr/local/${INSTALL_ENV}/${node_type}/bin/gse_agent -f /usr/local/${INSTALL_ENV}/${node_type}/etc/gse_agent.conf
-ExecReload=/usr/local/${INSTALL_ENV}/${node_type}/bin/gse_agent --reload
-ExecStop=/usr/local/${INSTALL_ENV}/${node_type}/bin/gse_agent --quit
+WorkingDirectory=${WORK_HOME}/bin
+PIDFile=${WORK_HOME}/bin/run/${module}.pid
+ExecStart=${WORK_HOME}/bin/gse_agent -f /usr/local/${INSTALL_ENV}/${node_type}/etc/gse_agent.conf
+ExecReload=${WORK_HOME}/bin/gse_agent --reload
+ExecStop=${WORK_HOME}/bin/gse_agent --quit
 Type=forking
 KillMode=process
 User=root
@@ -573,26 +585,8 @@ EOF
     fi
 
     # 删除rc.local里的启动项
-    get_os_type
-    if [ -f $RC_LOCAL_FILE ];then
-        if [ `egrep "local/${INSTALL_ENV}/${node_type}" $RC_LOCAL_FILE |wc -l` -ne 0 ];then
-            sed -i "/local\/${INSTALL_ENV}\/${node_type}/d" $RC_LOCAL_FILE
-        fi
-    else
-        if [ -f /etc/rc.d/rc.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/rc.d/rc.local |wc -l` -ne 0 ];then
-                sed -i "/local\/${INSTALL_ENV}\/${node_type}/d" /etc/rc.d/rc.local
-            fi
-        elif [ -f /etc/init.d/rc.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/init.d/rc.local |wc -l` -ne 0 ];then
-                sed -i "/local\/${INSTALL_ENV}\/${node_type}/d" /etc/init.d/rc.local
-            fi
-        elif [ -f /etc/init.d/boot.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/init.d/boot.local |wc -l` -ne 0 ];then
-                sed -i "/local\/${INSTALL_ENV}\/${node_type}/d" /etc/init.d/boot.local
-            fi
-        fi
-    fi
+    check_rc_file
+    sed -i "\|${WORK_HOME}/bin/gsectl start ${module}|d" $RC_LOCAL_FILE
 
     # 删除crontab里的watch条目
     remove_crontab

diff --git a/script_tools/gsectl/proxy/linux/gsectl b/script_tools/gsectl/proxy/linux/gsectl
@@ -610,7 +610,7 @@ start_by_crontab () {
 }
 
 stop_by_crontab () {
-
+    remove_crontab
     for module in $modules
     do
         stop_by_binary ${module}
@@ -828,31 +828,43 @@ get_os_type () {
     fi
 }
 
-add_startup_to_boot (){
+check_rc_file () {
+    get_os_type
+    if [ -f $RC_LOCAL_FILE ]; then
+        return 0
+    elif [ -f "/etc/rc.d/rc.local" ]; then
+        RC_LOCAL_FILE="/etc/rc.d/rc.local"
+    elif [ -f "/etc/init.d/rc.local" ]; then
+        RC_LOCAL_FILE="/etc/init.d/rc.local"
+    elif [ -f "/etc/init.d/boot.local" ]; then
+        RC_LOCAL_FILE="/etc/init.d/boot.local"
+    else
+        RC_LOCAL_FILE="`readlink -f /etc/rc.local`"
+    fi
+}
+
+
+add_startup_to_boot () {
+
     local module=$1
 
     # 添加启动项到 rc.local
-    echo "Check startup items, and if not existing, add the startup item to rc.local"
-    get_os_type
-    if [ -f $RC_LOCAL_FILE ];then
-        if [ `egrep "local/${INSTALL_ENV}/${node_type}" $RC_LOCAL_FILE |egrep ${module} |wc -l` -eq 0 ];then
-            echo "[ -f /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl ] && /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl start ${module} 1>>/var/log/${INSTALL_ENV}_${node_type}.log 2>&1" >> $RC_LOCAL_FILE
-        fi
-    else
-        if [ -f /etc/rc.d/rc.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/rc.d/rc.local |egrep ${module} |wc -l` -eq 0 ];then
-                echo "[ -f /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl ] && /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl start ${module} 1>>/var/log/${INSTALL_ENV}_${node_type}.log 2>&1" >> $RC_LOCAL_FILE
-            fi
-        elif [ -f /etc/init.d/rc.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/init.d/rc.local |egrep ${module} |wc -l` -eq 0 ];then
-                echo "[ -f /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl ] && /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl start ${module} 1>>/var/log/${INSTALL_ENV}_${node_type}.log 2>&1" >> $RC_LOCAL_FILE
-            fi
-        elif [ -f /etc/init.d/boot.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/init.d/boot.local |egrep ${module} |wc -l` -eq 0 ];then
-                echo "[ -f /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl ] && /usr/local/${INSTALL_ENV}/${node_type}/bin/gsectl start ${module} 1>>/var/log/${INSTALL_ENV}_${node_type}.log 2>&1" >> $RC_LOCAL_FILE
-            fi
-        fi
+    echo "Check startup items, and if not existing, add the [${module}] startup item to rc.local"
+
+    check_rc_file
+    local rcfile=$RC_LOCAL_FILE
+
+    if [ $OS_TYPE == "ubuntu" ]; then
+        sed -i "\|\#\!/bin/bash|d" $rcfile
+        sed -i "1i \#\!/bin/bash" $rcfile
     fi
+
+    chmod +x $rcfile
+
+    # 先删后加，避免重复
+    sed -i "\|${WORK_HOME}/bin/gsectl start ${module}|d" $rcfile
+
+    echo "[ -f ${WORK_HOME}/bin/gsectl ] && ${WORK_HOME}/bin/gsectl start ${module} 1>>/var/log/${INSTALL_ENV}_${node_type}.log 2>&1" >>$rcfile
 }
 
 add_config_to_systemd () {
@@ -875,11 +887,11 @@ After=network-online.target
 [Service]
 LimitNOFILE=512000
 LimitCORE=infinity
-WorkingDirectory=/usr/local/${INSTALL_ENV}/${node_type}/bin
-PIDFile=/usr/local/${INSTALL_ENV}/${node_type}/bin/run/${module}.pid
-ExecStart=/usr/local/${INSTALL_ENV}/${node_type}/bin/gse_${module} -f /usr/local/${INSTALL_ENV}/${node_type}/etc/gse_${config}.conf
-ExecReload=/usr/local/${INSTALL_ENV}/${node_type}/bin/gse_${module} --reload
-ExecStop=/usr/local/${INSTALL_ENV}/${node_type}/bin/gse_${module} --quit
+WorkingDirectory=${WORK_HOME}/bin
+PIDFile=${WORK_HOME}/bin/run/${module}.pid
+ExecStart=${WORK_HOME}/bin/gse_${module} -f ${WORK_HOME}/etc/gse_${config}.conf
+ExecReload=${WORK_HOME}/bin/gse_${module} --reload
+ExecStop=${WORK_HOME}/bin/gse_${module} --quit
 Type=forking
 KillMode=process
 User=root
@@ -909,26 +921,8 @@ EOF
     [ -f /tmp/${INSTALL_ENV}_${node_type}_${module}.service ] && rm /tmp/${INSTALL_ENV}_${node_type}_${module}.service
 
     # 删除rc.local里的启动项
-    get_os_type
-    if [ -f $RC_LOCAL_FILE ];then
-        if [ `egrep "local/${INSTALL_ENV}/${node_type}" $RC_LOCAL_FILE |wc -l` -ne 0 ];then
-            sed -i "/local\/${INSTALL_ENV}\/${node_type}/d" $RC_LOCAL_FILE
-        fi
-    else
-        if [ -f /etc/rc.d/rc.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/rc.d/rc.local |wc -l` -ne 0 ];then
-                sed -i "/local\/${INSTALL_ENV}\/${node_type}/d" /etc/rc.d/rc.local
-            fi
-        elif [ -f /etc/init.d/rc.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/init.d/rc.local |wc -l` -ne 0 ];then
-                sed -i "/local\/${INSTALL_ENV}\/${node_type}/d" /etc/init.d/rc.local
-            fi
-        elif [ -f /etc/init.d/boot.local ];then
-            if [ `egrep "local/${INSTALL_ENV}/${node_type}" /etc/init.d/boot.local |wc -l` -ne 0 ];then
-                sed -i "/local\/${INSTALL_ENV}\/${node_type}/d" /etc/init.d/boot.local
-            fi
-        fi
-    fi
+    check_rc_file
+    sed -i "\|${WORK_HOME}/bin/gsectl start ${module}|d" $RC_LOCAL_FILE
 
     # 删除crontab里的定时任务
     remove_crontab
@@ -950,10 +944,6 @@ setup_crontab () {
     local module=$1
     local tmpcron
 
-    if [ "${module}" != "${MODULES[-1]}" ];then
-        return 0
-    fi
-
     if [ -n "`crontab -l | grep \"$WORK_HOME/bin/gsectl\"`" ];then
         echo "The watch detection entry is already in the crontab..."
         return 0
@@ -975,12 +965,14 @@ remove_crontab (){
     local tmpcron
     tmpcron=/tmp/cron.XXXXXX
 
-    crontab -l |egrep -v "$WORK_HOME" >$tmpcron
-    crontab $tmpcron && rm -f $tmpcron
+    if [ `crontab -l |egrep  "$WORK_HOME" |wc -l` -ne 0 ];then
+        crontab -l |egrep -v "$WORK_HOME" >$tmpcron
+        crontab $tmpcron && rm -f $tmpcron
 
-    # 下面这段代码是为了确保修改的crontab立即生效
-    if pgrep -x crond &>/dev/null; then
-        pkill -HUP -x crond
+        # 下面这段代码是为了确保修改的crontab立即生效
+        if pgrep -x crond &>/dev/null; then
+            pkill -HUP -x crond
+        fi
     fi
 }
 

diff --git a/script_tools/setup_agent.sh b/script_tools/setup_agent.sh
@@ -395,7 +395,8 @@ remove_crontab () {
     local tmpcron
     tmpcron=$(mktemp "$TMP_DIR"/cron.XXXXXXX)
 
-    crontab -l | grep -v "bin/gsectl"  >"$tmpcron"
+    # 仅删除关联到安装目录的 crontab，避免多 Agent 互相影响
+    crontab -l | grep -v "${AGENT_SETUP_PATH}"  >"$tmpcron"
     crontab "$tmpcron" && rm -f "$tmpcron"
 
     # 下面这段代码是为了确保修改的crontab能立即生效

diff --git a/script_tools/setup_agent.zsh b/script_tools/setup_agent.zsh
@@ -373,7 +373,8 @@ remove_crontab () {
     local tmpcron
     tmpcron=$(mktemp "$TMP_DIR"/cron.XXXXXXX)
 
-    crontab -l | grep -v "bin/gsectl"  >"$tmpcron"
+    # 仅删除关联到安装目录的 crontab，避免多 Agent 互相影响
+    crontab -l | grep -v "${AGENT_SETUP_PATH}"  >"$tmpcron"
     crontab "$tmpcron" && rm -f "$tmpcron"
 
     # 下面这段代码是为了确保修改的crontab能立即生效

diff --git a/script_tools/setup_proxy.sh b/script_tools/setup_proxy.sh
@@ -352,7 +352,8 @@ remove_crontab () {
     local tmpcron
     tmpcron=$(mktemp "$TMP_DIR"/cron.XXXXXXX)
 
-    crontab -l | grep -v "bin/gsectl"  >"$tmpcron"
+    # 仅删除关联到安装目录的 crontab，避免多 Agent 互相影响
+    crontab -l | grep -v "${AGENT_SETUP_PATH}"  >"$tmpcron"
     crontab "$tmpcron" && rm -f "$tmpcron"
 
     # 下面这段代码是为了确保修改的crontab能立即生效