Merge pull request #2627 from taosdata/feature/sangshuduo/random-file-fail

add --random-file-fail-factor argument to cover more scenarios.

src/dnode/src/dnodeSystem.c

@@ -20,6 +20,7 @@
 #include "tglobal.h"
 #include "dnodeInt.h"
 #include "dnodeMain.h"
+#include "tfile.h"
 
 static void signal_handler(int32_t signum, siginfo_t *sigInfo, void *context);
 static sem_t exitSem;
@@ -67,6 +68,18 @@ int32_t main(int32_t argc, char *argv[]) {
         taosSetAllocMode(TAOS_ALLOC_MODE_DETECT_LEAK, NULL, true);
       }
     }
+#endif
+#ifdef TAOS_RANDOM_FILE_FAIL
+    else if (strcmp(argv[i], "--random-file-fail-factor") == 0) {
+      if ( (i+1) < argc ) {
+        int factor = atoi(argv[i+1]);
+        printf("The factor of random failure is %d\n", factor);
+        taosSetRandomFileFailFactor(factor);
+      } else {
+        printf("Please specify a number for random failure factor!");
+        exit(EXIT_FAILURE);
+      }
+    }
 #endif
   }
 

src/util/inc/tfile.h

@@ -18,6 +18,7 @@
 
 #ifdef TAOS_RANDOM_FILE_FAIL
 
+void taosSetRandomFileFailFactor(int factor);
 ssize_t taos_tread(int fd, void *buf, size_t count);
 ssize_t taos_twrite(int fd, void *buf, size_t count);
 off_t taos_lseek(int fd, off_t offset, int whence);

src/util/src/tfile.c

@@ -26,40 +26,51 @@
 
 #include "os.h"
 
-#define RANDOM_FILE_FAIL_FACTOR 5
+#ifdef TAOS_RANDOM_FILE_FAIL
+
+static int random_file_fail_factor = 20;
+
+void taosSetRandomFileFailFactor(int factor)
+{
+  random_file_fail_factor = factor;
+}
+#endif
 
 ssize_t taos_tread(int fd, void *buf, size_t count)
 {
 #ifdef TAOS_RANDOM_FILE_FAIL
-  if (rand() % RANDOM_FILE_FAIL_FACTOR == 0) {
-    errno = EIO;
-    return -1;
+  if (random_file_fail_factor > 0) {
+    if (rand() % random_file_fail_factor == 0) {
+      errno = EIO;
+      return -1;
+    }
   }
 #endif
-
   return tread(fd, buf, count);
 }
 
 ssize_t taos_twrite(int fd, void *buf, size_t count)
 {
 #ifdef TAOS_RANDOM_FILE_FAIL
-  if (rand() % RANDOM_FILE_FAIL_FACTOR == 0) {
-    errno = EIO;
-    return -1;
+  if (random_file_fail_factor > 0) {
+    if (rand() % random_file_fail_factor == 0) {
+      errno = EIO;
+      return -1;
+    }
   }
 #endif
-
   return twrite(fd, buf, count);
 }
 
 off_t taos_lseek(int fd, off_t offset, int whence)
 {
 #ifdef TAOS_RANDOM_FILE_FAIL
-  if (rand() % RANDOM_FILE_FAIL_FACTOR == 0) {
-    errno = EIO;
-    return -1;
+  if (random_file_fail_factor > 0) {
+    if (rand() % random_file_fail_factor == 0) {
+      errno = EIO;
+      return -1;
+    }
   }
 #endif
-
   return lseek(fd, offset, whence);
 }

tests/pytest/util/dnodes-random-fail.py

+###################################################################
+#           Copyright (c) 2016 by TAOS Technologies, Inc.
+#                     All rights reserved.
+#
+#  This file is proprietary and confidential to TAOS Technologies.
+#  No part of this file may be reproduced, stored, transmitted,
+#  disclosed or used in any form or by any means other than as
+#  expressly provided by the written permission from Jianhui Tao
+#
+###################################################################
+
+# -*- coding: utf-8 -*-
+
+import sys
+import os
+import os.path
+import subprocess
+from util.log import *
+
+
+class TDSimClient:
+    def __init__(self):
+        self.testCluster = False
+
+        self.cfgDict = {
+            "numOfLogLines": "100000000",
+            "numOfThreadsPerCore": "2.0",
+            "locale": "en_US.UTF-8",
+            "charset": "UTF-8",
+            "asyncLog": "0",
+            "anyIp": "0",
+            "sdbDebugFlag": "135",
+            "rpcDebugFlag": "135",
+            "tmrDebugFlag": "131",
+            "cDebugFlag": "135",
+            "udebugFlag": "135",
+            "jnidebugFlag": "135",
+            "qdebugFlag": "135",
+            }
+
+    def init(self, path):
+        self.__init__()
+        self.path = path
+
+    def getLogDir(self):
+        self.logDir = "%s/sim/psim/log" % (self.path)
+        return self.logDir
+
+    def getCfgDir(self):
+        self.cfgDir = "%s/sim/psim/cfg" % (self.path)
+        return self.cfgDir
+
+    def setTestCluster(self, value):
+        self.testCluster = value
+
+    def addExtraCfg(self, option, value):
+        self.cfgDict.update({option: value})
+
+    def cfg(self, option, value):
+        cmd = "echo '%s %s' >> %s" % (option, value, self.cfgPath)
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+    def deploy(self):
+        self.logDir = "%s/sim/psim/log" % (self.path)
+        self.cfgDir = "%s/sim/psim/cfg" % (self.path)
+        self.cfgPath = "%s/sim/psim/cfg/taos.cfg" % (self.path)
+
+        cmd = "rm -rf " + self.logDir
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+    
+        cmd = "mkdir -p " + self.logDir
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+        cmd = "rm -rf " + self.cfgDir
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+        cmd = "mkdir -p " + self.cfgDir
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+        cmd = "touch " + self.cfgPath
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+        if self.testCluster:
+            self.cfg("masterIp", "192.168.0.1")
+            self.cfg("secondIp", "192.168.0.2")
+        self.cfg("logDir", self.logDir)
+
+        for key, value in self.cfgDict.items():
+            self.cfg(key, value)
+
+        tdLog.debug("psim is deployed and configured by %s" % (self.cfgPath))
+
+
+class TDDnode:
+    def __init__(self, index):
+        self.index = index
+        self.running = 0
+        self.deployed = 0
+        self.testCluster = False
+        self.valgrind = 0
+
+    def init(self, path):
+        self.path = path
+
+    def setTestCluster(self, value):
+        self.testCluster = value
+
+    def setValgrind(self, value):
+        self.valgrind = value
+
+    def getDataSize(self):
+        totalSize = 0
+
+        if (self.deployed == 1):
+            for dirpath, dirnames, filenames in os.walk(self.dataDir):
+                for f in filenames:
+                    fp = os.path.join(dirpath, f)
+
+                    if not os.path.islink(fp):
+                        totalSize = totalSize + os.path.getsize(fp)
+
+        return totalSize
+
+    def deploy(self):
+        self.logDir = "%s/sim/dnode%d/log" % (self.path, self.index)
+        self.dataDir = "%s/sim/dnode%d/data" % (self.path, self.index)
+        self.cfgDir = "%s/sim/dnode%d/cfg" % (self.path, self.index)
+        self.cfgPath = "%s/sim/dnode%d/cfg/taos.cfg" % (
+            self.path, self.index)
+
+        cmd = "rm -rf " + self.dataDir
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+        cmd = "rm -rf " + self.logDir
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+        cmd = "rm -rf " + self.cfgDir
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+        cmd = "mkdir -p " + self.dataDir
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+        cmd = "mkdir -p " + self.logDir
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+        cmd = "mkdir -p " + self.cfgDir
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+        cmd = "touch " + self.cfgPath
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+        if self.testCluster:
+            self.startIP()
+
+        if self.testCluster:
+            self.cfg("masterIp", "192.168.0.1")
+            self.cfg("secondIp", "192.168.0.2")
+            self.cfg("publicIp", "192.168.0.%d" % (self.index))
+            self.cfg("internalIp", "192.168.0.%d" % (self.index))
+            self.cfg("privateIp", "192.168.0.%d" % (self.index))
+        self.cfg("dataDir", self.dataDir)
+        self.cfg("logDir", self.logDir)
+        self.cfg("numOfLogLines", "100000000")
+        self.cfg("mnodeEqualVnodeNum", "0")
+        self.cfg("walLevel", "1")
+        self.cfg("statusInterval", "1")
+        self.cfg("numOfTotalVnodes", "64")
+        self.cfg("numOfMnodes", "3")
+        self.cfg("numOfThreadsPerCore", "2.0")
+        self.cfg("monitor", "0")
+        self.cfg("maxVnodeConnections", "30000")
+        self.cfg("maxMgmtConnections", "30000")
+        self.cfg("maxMeterConnections", "30000")
+        self.cfg("maxShellConns", "30000")
+        self.cfg("locale", "en_US.UTF-8")
+        self.cfg("charset", "UTF-8")
+        self.cfg("asyncLog", "0")
+        self.cfg("anyIp", "0")
+        self.cfg("dDebugFlag", "135")
+        self.cfg("mDebugFlag", "135")
+        self.cfg("sdbDebugFlag", "135")
+        self.cfg("rpcDebugFlag", "135")
+        self.cfg("tmrDebugFlag", "131")
+        self.cfg("cDebugFlag", "135")
+        self.cfg("httpDebugFlag", "135")
+        self.cfg("monitorDebugFlag", "135")
+        self.cfg("udebugFlag", "135")
+        self.cfg("jnidebugFlag", "135")
+        self.cfg("qdebugFlag", "135")
+        self.deployed = 1
+        tdLog.debug(
+            "dnode:%d is deployed and configured by %s" %
+            (self.index, self.cfgPath))
+
+    def getBuildPath(self):
+        selfPath = os.path.dirname(os.path.realpath(__file__))
+
+        if ("community" in selfPath):
+            projPath = selfPath[:selfPath.find("community")]
+        else:
+            projPath = selfPath[:selfPath.find("tests")]
+
+        for root, dirs, files in os.walk(projPath):
+            if ("taosd" in files):
+                rootRealPath = os.path.dirname(os.path.realpath(root))
+                if ("packaging" not in rootRealPath):
+                    buildPath = root[:len(root)-len("/build/bin")]
+                    break
+        return buildPath
+
+    def start(self):
+        buildPath = self.getBuildPath()
+
+        if (buildPath == ""):
+            tdLog.exit("taosd not found!")
+        else:
+            tdLog.info("taosd found in %s" % buildPath)
+
+        binPath = buildPath + "/build/bin/taosd"
+
+        if self.deployed == 0:
+            tdLog.exit("dnode:%d is not deployed" % (self.index))
+
+        if self.valgrind == 0:
+            cmd = "nohup %s -c %s > /dev/null 2>&1 & " % (
+                binPath, self.cfgDir)
+        else:
+            valgrindCmdline = "valgrind --tool=memcheck --leak-check=full --show-reachable=no --track-origins=yes --show-leak-kinds=all -v --workaround-gcc296-bugs=yes"
+
+            cmd = "nohup %s %s -c %s --random-file-fail-factor 5 2>&1 & " % (
+                valgrindCmdline, binPath, self.cfgDir)
+
+            print(cmd)
+
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+        self.running = 1
+        tdLog.debug("dnode:%d is running with %s " % (self.index, cmd))
+
+        tdLog.debug("wait 5 seconds for the dnode:%d to start." % (self.index))
+        time.sleep(5)
+
+    def stop(self):
+        if self.valgrind == 0:
+            toBeKilled = "taosd"
+        else:
+            toBeKilled = "valgrind.bin"
+
+        if self.running != 0:
+            psCmd = "ps -ef|grep -w %s| grep -v grep | awk '{print $2}'" % toBeKilled
+            processID = subprocess.check_output(
+                psCmd, shell=True).decode("utf-8")
+
+            while(processID):
+                killCmd = "kill -INT %s > /dev/null 2>&1" % processID
+                os.system(killCmd)
+                time.sleep(1)
+                processID = subprocess.check_output(
+                    psCmd, shell=True).decode("utf-8")
+            for port in range(6030, 6041):
+                fuserCmd = "fuser -k -n tcp %d" % port
+                os.system(fuserCmd)
+            if self.valgrind:
+                time.sleep(2)
+
+            self.running = 0
+            tdLog.debug("dnode:%d is stopped by kill -INT" % (self.index))
+
+    def forcestop(self):
+        if self.valgrind == 0:
+            toBeKilled = "taosd"
+        else:
+            toBeKilled = "valgrind.bin"
+
+        if self.running != 0:
+            psCmd = "ps -ef|grep -w %s| grep -v grep | awk '{print $2}'" % toBeKilled
+            processID = subprocess.check_output(
+                psCmd, shell=True).decode("utf-8")
+
+            while(processID):
+                killCmd = "kill -KILL %s > /dev/null 2>&1" % processID
+                os.system(killCmd)
+                time.sleep(1)
+                processID = subprocess.check_output(
+                    psCmd, shell=True).decode("utf-8")
+            for port in range(6030, 6041):
+                fuserCmd = "fuser -k -n tcp %d" % port
+                os.system(fuserCmd)
+            if self.valgrind:
+                time.sleep(2)
+
+            self.running = 0
+            tdLog.debug("dnode:%d is stopped by kill -KILL" % (self.index))
+
+    def startIP(self):
+        cmd = "sudo ifconfig lo:%d 192.168.0.%d up" % (self.index, self.index)
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+    def stopIP(self):
+        cmd = "sudo ifconfig lo:%d 192.168.0.%d down" % (
+            self.index, self.index)
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+    def cfg(self, option, value):
+        cmd = "echo '%s %s' >> %s" % (option, value, self.cfgPath)
+        if os.system(cmd) != 0:
+            tdLog.exit(cmd)
+
+    def getDnodeRootDir(self, index):
+        dnodeRootDir = "%s/sim/psim/dnode%d" % (self.path, index)
+        return dnodeRootDir
+
+    def getDnodesRootDir(self):
+        dnodesRootDir = "%s/sim/psim" % (self.path)
+        return dnodesRootDir
+
+
+class TDDnodes:
+    def __init__(self):
+        self.dnodes = []
+        self.dnodes.append(TDDnode(1))
+        self.dnodes.append(TDDnode(2))
+        self.dnodes.append(TDDnode(3))
+        self.dnodes.append(TDDnode(4))
+        self.dnodes.append(TDDnode(5))
+        self.dnodes.append(TDDnode(6))
+        self.dnodes.append(TDDnode(7))
+        self.dnodes.append(TDDnode(8))
+        self.dnodes.append(TDDnode(9))
+        self.dnodes.append(TDDnode(10))
+        self.simDeployed = False
+
+    def init(self, path):
+        psCmd = "ps -ef|grep -w taosd| grep -v grep | awk '{print $2}'"
+        processID = subprocess.check_output(psCmd, shell=True).decode("utf-8")
+        while(processID):
+            killCmd = "kill -KILL %s > /dev/null 2>&1" % processID
+            os.system(killCmd)
+            time.sleep(1)
+            processID = subprocess.check_output(
+                psCmd, shell=True).decode("utf-8")
+
+        psCmd = "ps -ef|grep -w valgrind.bin| grep -v grep | awk '{print $2}'"
+        processID = subprocess.check_output(psCmd, shell=True).decode("utf-8")
+        while(processID):
+            killCmd = "kill -KILL %s > /dev/null 2>&1" % processID
+            os.system(killCmd)
+            time.sleep(1)
+            processID = subprocess.check_output(
+                psCmd, shell=True).decode("utf-8")
+
+        binPath = os.path.dirname(os.path.realpath(__file__))
+        binPath = binPath + "/../../../debug/"
+        tdLog.debug("binPath %s" % (binPath))
+        binPath = os.path.realpath(binPath)
+        tdLog.debug("binPath real path %s" % (binPath))
+
+        # cmd = "sudo cp %s/build/lib/libtaos.so /usr/local/lib/taos/" % (binPath)
+        # tdLog.debug(cmd)
+        # os.system(cmd)
+
+        # cmd = "sudo cp %s/build/bin/taos /usr/local/bin/taos/" % (binPath)
+        # if os.system(cmd) != 0 :
+        #  tdLog.exit(cmd)
+        # tdLog.debug("execute %s" % (cmd))
+
+        # cmd = "sudo cp %s/build/bin/taosd /usr/local/bin/taos/" % (binPath)
+        # if os.system(cmd) != 0 :
+        # tdLog.exit(cmd)
+        # tdLog.debug("execute %s" % (cmd))
+
+        if path == "":
+            # self.path = os.path.expanduser('~')
+            self.path = os.path.abspath(binPath + "../../")
+        else:
+            self.path = os.path.realpath(path)
+
+        for i in range(len(self.dnodes)):
+            self.dnodes[i].init(self.path)
+
+        self.sim = TDSimClient()
+        self.sim.init(self.path)
+
+    def setTestCluster(self, value):
+        self.testCluster = value
+
+    def setValgrind(self, value):
+        self.valgrind = value
+
+    def deploy(self, index):
+        self.sim.setTestCluster(self.testCluster)
+
+        if (self.simDeployed == False):
+            self.sim.deploy()
+            self.simDeployed = True
+
+        self.check(index)
+        self.dnodes[index - 1].setTestCluster(self.testCluster)
+        self.dnodes[index - 1].setValgrind(self.valgrind)
+        self.dnodes[index - 1].deploy()
+
+    def cfg(self, index, option, value):
+        self.check(index)
+        self.dnodes[index - 1].cfg(option, value)
+
+    def start(self, index):
+        self.check(index)
+        self.dnodes[index - 1].start()
+
+    def stop(self, index):
+        self.check(index)
+        self.dnodes[index - 1].stop()
+
+    def getDataSize(self, index):
+        self.check(index)
+        return self.dnodes[index - 1].getDataSize()
+
+    def forcestop(self, index):
+        self.check(index)
+        self.dnodes[index - 1].forcestop()
+
+    def startIP(self, index):
+        self.check(index)
+
+        if self.testCluster:
+            self.dnodes[index - 1].startIP()
+
+    def stopIP(self, index):
+        self.check(index)
+
+        if self.dnodes[index - 1].testCluster:
+            self.dnodes[index - 1].stopIP()
+
+    def check(self, index):
+        if index < 1 or index > 10:
+            tdLog.exit("index:%d should on a scale of [1, 10]" % (index))
+
+    def stopAll(self):
+        tdLog.info("stop all dnodes")
+        for i in range(len(self.dnodes)):
+            self.dnodes[i].stop()
+
+        psCmd = "ps -ef | grep -w taosd | grep 'root' | grep -v grep | awk '{print $2}'"
+        processID = subprocess.check_output(psCmd, shell=True).decode("utf-8")
+        if processID:
+            cmd = "sudo systemctl stop taosd"
+            os.system(cmd)
+        # if os.system(cmd) != 0 :
+        # tdLog.exit(cmd)
+        psCmd = "ps -ef|grep -w taosd| grep -v grep | awk '{print $2}'"
+        processID = subprocess.check_output(psCmd, shell=True).decode("utf-8")
+        while(processID):
+            killCmd = "kill -KILL %s > /dev/null 2>&1" % processID
+            os.system(killCmd)
+            time.sleep(1)
+            processID = subprocess.check_output(
+                psCmd, shell=True).decode("utf-8")
+
+        psCmd = "ps -ef|grep -w valgrind.bin| grep -v grep | awk '{print $2}'"
+        processID = subprocess.check_output(psCmd, shell=True).decode("utf-8")
+        while(processID):
+            killCmd = "kill -KILL %s > /dev/null 2>&1" % processID
+            os.system(killCmd)
+            time.sleep(1)
+            processID = subprocess.check_output(
+                psCmd, shell=True).decode("utf-8")
+
+        # if os.system(cmd) != 0 :
+        # tdLog.exit(cmd)
+
+    def getDnodesRootDir(self):
+        dnodesRootDir = "%s/sim" % (self.path)
+        return dnodesRootDir
+
+    def getSimCfgPath(self):
+        return self.sim.getCfgDir()
+
+    def getSimLogPath(self):
+        return self.sim.getLogDir()
+
+    def addSimExtraCfg(self, option, value):
+        self.sim.addExtraCfg(option, value)
+
+
+tdDnodes = TDDnodes()

tests/script/sh/exec-random-fail.sh

+#!/bin/bash
+
+# if [ $# != 4 || $# != 5 ]; then 
+  # echo "argument list need input : "
+  # echo "  -n nodeName"
+  # echo "  -s start/stop"
+  # echo "  -c clear"
+  # exit 1
+# fi
+
+NODE_NAME=
+EXEC_OPTON=
+CLEAR_OPTION="false"
+while getopts "n:s:u:x:ct" arg 
+do
+  case $arg in
+    n)
+      NODE_NAME=$OPTARG
+      ;;
+    s)
+      EXEC_OPTON=$OPTARG
+      ;;
+    c)
+      CLEAR_OPTION="clear"
+      ;;
+    t)
+      SHELL_OPTION="true"
+      ;;
+    u)
+      USERS=$OPTARG
+      ;;
+    x)
+      SIGNAL=$OPTARG
+      ;;
+    ?)
+      echo "unkown argument"
+      ;;
+  esac
+done
+
+SCRIPT_DIR=`dirname $0`
+cd $SCRIPT_DIR/../
+SCRIPT_DIR=`pwd`
+
+IN_TDINTERNAL="community"
+if [[ "$SCRIPT_DIR" == *"$IN_TDINTERNAL"* ]]; then
+  cd ../../..
+else
+  cd ../../
+fi
+
+TAOS_DIR=`pwd`
+TAOSD_DIR=`find . -name "taosd"|grep bin|head -n1`
+
+if [[ "$TAOSD_DIR" == *"$IN_TDINTERNAL"* ]]; then
+  BIN_DIR=`find . -name "taosd"|grep bin|head -n1|cut -d '/' --fields=2,3`
+else
+  BIN_DIR=`find . -name "taosd"|grep bin|head -n1|cut -d '/' --fields=2`
+fi
+
+BUILD_DIR=$TAOS_DIR/$BIN_DIR/build
+
+SIM_DIR=$TAOS_DIR/sim
+NODE_DIR=$SIM_DIR/$NODE_NAME
+EXE_DIR=$BUILD_DIR/bin
+CFG_DIR=$NODE_DIR/cfg
+LOG_DIR=$NODE_DIR/log
+DATA_DIR=$NODE_DIR/data
+MGMT_DIR=$NODE_DIR/data/mgmt
+TSDB_DIR=$NODE_DIR/data/tsdb
+
+TAOS_CFG=$NODE_DIR/cfg/taos.cfg
+
+echo ------------ $EXEC_OPTON $NODE_NAME
+
+TAOS_FLAG=$SIM_DIR/tsim/flag
+if [ -f "$TAOS_FLAG" ]; then 
+  EXE_DIR=/usr/local/bin/taos
+fi
+
+if [ "$CLEAR_OPTION" = "clear" ]; then 
+  echo rm -rf $MGMT_DIR $TSDB_DIR  
+  rm -rf $TSDB_DIR
+  rm -rf $MGMT_DIR
+fi
+
+if [ "$EXEC_OPTON" = "start" ]; then 
+  echo "ExcuteCmd:" $EXE_DIR/taosd -c $CFG_DIR
+  
+  if [ "$SHELL_OPTION" = "true" ]; then 
+    nohup valgrind --log-file=${LOG_DIR}/valgrind.log --tool=memcheck --leak-check=full --show-reachable=no  --track-origins=yes --show-leak-kinds=all  -v  --workaround-gcc296-bugs=yes   $EXE_DIR/taosd -c $CFG_DIR > /dev/null 2>&1 &   
+  else
+    nohup $EXE_DIR/taosd -c $CFG_DIR --random-file-fail-factor 5 > /dev/null 2>&1 & 
+  fi
+  
+else
+  #relative path
+  RCFG_DIR=sim/$NODE_NAME/cfg
+  PID=`ps -ef|grep taosd | grep $RCFG_DIR | grep -v grep | awk '{print $2}'`
+  while [ -n "$PID" ]
+  do
+    if [ "$SIGNAL" = "SIGINT" ]; then 
+      echo try to kill by signal SIGINT
+      kill -SIGINT $PID
+    else
+      echo try to kill by signal SIGKILL
+      kill -9 $PID
+    fi
+    sleep 1
+    PID=`ps -ef|grep taosd | grep $RCFG_DIR | grep -v grep | awk '{print $2}'`
+  done 
+fi
+