cannot create /tmp/zookeeper/zookeeper_server.pid

最近使用zookeeper发现, 首次启动zookeeper时, 都会遇到一个错误:

$ bin/zkServer.sh start

JMX enabled by default

Using config: /home/nauhcud/workspace/zookeeper/zookeeper/bin/../conf/zoo.cfg

Starting zookeeper ... bin/zkServer.sh: 162: cannot create /tmp/zookeeper/zookeeper_server.pid: Directory nonexistent

FAILED TO WRITE PID

查看进程发现zookeeper进程存在, 并可以正常使用. dataDir "/tmp/zookeeper/" 也在. 

kill掉zookeeper进程重新启动, 一切正常.

删掉dataDir, 问题依旧.

查看启动脚本, 发现start逻辑如下:

start)

...

    nohup $JAVA "-Dzookeeper.log.dir=${ZOO_LOG_DIR}" "-Dzookeeper.root.logger=${ZOO_LOG4J_PROP}" \

    -cp "$CLASSPATH" $JVMFLAGS $ZOOMAIN "$ZOOCFG" > "$_ZOO_DAEMON_OUT" 2>&1 < /dev/null &

    if [ $? -eq 0 ] 

    then

      if /bin/echo -n $! > "$ZOOPIDFILE"

      then

        sleep 1

        echo STARTED

      else

        echo FAILED TO WRITE PID   #这句说明写入Pid出现了问题

        exit 1

      fi  

...

纵观整个脚本, dataDir没有出现, 因此, dataDir应该是zookeeper进程内部建立的, 并且有一定延迟, 因此将zookeeper进程id写入到dataDir下的pidfile时, dataDir还没有建立好, 因此就出现了上述情况. 

解决办法很简单, 在写入pid之前先判断一下datadir是否存在, 让zookeeper有时间做完初始化, 然后再将pid写入即可.

start)

...

    nohup $JAVA "-Dzookeeper.log.dir=${ZOO_LOG_DIR}" "-Dzookeeper.root.logger=${ZOO_LOG4J_PROP}" \

    -cp "$CLASSPATH" $JVMFLAGS $ZOOMAIN "$ZOOCFG" > "$_ZOO_DAEMON_OUT" 2>&1 < /dev/null &

    zkpid=$!;

    if [ $? -eq 0 ] 

    then

      while [ ! -d `dirname $ZOOPIDFILE` ]

      do  

        sleep 1;

      done

      if /bin/echo -n $zkpid > "$ZOOPIDFILE"

      then

        sleep 1

        echo STARTED

      else

        echo FAILED TO WRITE PID

        exit 1

      fi  

...