def checkServerProcess():
hostAndPorts = conf.get('zookeeper.hosts')
serverlist = getQuorumPeerMain(hostAndPorts.split(','))
for host in serverlist:
content = exeCmd.execJps(host)
if len(re.findall(serverlist.get(host), content)) < 1:
log.warn('{host} 节点 QuorumPeerMain 服务未运行'.format(host=host))
startZk(host, 'QuorumPeerMain')
else:
log.info('{host} 节点 QuorumPeerMain 服务正在运行'.format(host=host))
def checkServerProcess():
HostAndPorts = conf.get('flume.hosts')
flumeServerList = getFlume(HostAndPorts.split(','))
for host in flumeServerList:
content = exeCmd.execJps(host)
if len(re.findall(flumeServerList.get(host), content)) < 1:
log.warn(host + ' 节点 ' + 'flume' + ' 服务未运行')
startFlume(host, 'flume')
else:
log.info(host + ' 节点 ' + 'flume' + '服务正在运行')
def checkServerProcess():
hostAndPorts = conf.get('es.hosts')
print(hostAndPorts)
serverlist = getES(hostAndPorts.split(','))
for host in serverlist:
content = exeCmd.execJps(host)
if (len(re.findall(serverlist.get(host), content)) < 1):
log.warn(host + ' 节点 ' + 'Elasticsearch' + ' 服务未运行')
startES(host, 'Elasticsearch')
else:
log.info(host + ' 节点 ' + 'Elasticsearch ' + '服务正在运行')
def exeCheckServerProcess():
serverList = checkServerProcess()
for host in serverList:
content = exeCmd.execJps(host)
for serverName in serverList.get(host).split(','):
if len(re.findall(serverName, content)) < 1:
log.warn('{host} 节点 {serverName} 服务未运行'.format(
host=host, serverName=serverName))
startHbase(host, serverName.lower())
else:
log.info('{host} 节点 {serverName} 服务正在运行'.format(
host=host, serverName=serverName))
def checkServerProcess():
hostAndPorts = conf.get('tomcat.hosts')
serverlist, proNums = getBootstrap(hostAndPorts.split(','))
for host in serverlist:
content = exeCmd.execJps(host)
nowProNum = len(re.findall(serverlist.get(host), content))
proNum = int(proNums.get(host))
if nowProNum < proNum:
log.warn('{host} 节点 {proNum} 个 Bootstrap 服务未运行'.format(
host=host, proNum=(proNum - nowProNum)))
startBootstrap(host, 'Bootstrap')
else:
log.info('{host} 节点 {proNum} 个 Bootstrap 服务正在运行'.format(
host=host, proNum=proNum))
def checkServerProcess():
hostAndPorts = conf.get('hive.server2')
serverlist = getHiveServer2(hostAndPorts.split(','))
hostAndPorts = conf.get('hive.metastore')
serverlist = getHiveMetaStore(hostAndPorts.split(','), serverlist)
for host in serverlist:
content = exeCmd.execJps(host, 'ansible client -l {host} -a "jps -m"')
for server in serverlist.get(host).split(','):
if len(re.findall(server, content)) < 1:
log.warn('{host} 节点 {server} 服务未运行'.format(host=host,
server=server))
startHive(host, server)
else:
log.info('{host} 节点 {server} 服务正在运行'.format(host=host,
server=server))
def exeCheckServerProcess():
serverList = checkServerProcess()
startNum = 0
for host in serverList:
content = exeCmd.execJps(host)
for server in serverList.get(host).split(','):
if len(re.findall(server, content)) < 1:
log.warn('{host} 节点 {server} 服务未运行'.format(host=host, server=server))
start_hadoop(host, server.lower())
startNum += 1
else:
log.info('{host} 节点 {server} 服务正在运行'.format(host=host, server=server))
if startNum > 0:
log.warn("检测到有 {startNum} 个hadoop进程重启".format(startNum=startNum))
time_util.sleep(30)
def checkServerProcess():
webHostAndPorts = conf.get('azkaban.web.hosts')
serverProcessList = getWeb(webHostAndPorts.split(','))
execHostAndPorts = conf.get('azkaban.exe.hosts')
serverProcessList = getExe(execHostAndPorts.split(','), serverProcessList)
for host in serverProcessList:
content = exeCmd.execJps(host)
for serverName in serverProcessList.get(host).split(','):
if len(re.findall(serverName, content)) < 1:
log.warn('{host} 节点 {serverName} 服务未运行'.format(
host=host, serverName=serverName))
startAzkaban(host, serverName)
else:
log.info('{host} 节点 {serverName} 服务正在运行'.format(
host=host, serverName=serverName))
def checkServerProcess():
hostAndPorts = conf.get('kafka.hosts')
serverlist = getKafka(hostAndPorts.split(','))
proNum = conf.get('kafka.process.number')
if proNum == '':
proNum = 1
else:
proNum = int(proNum)
for host in serverlist:
content = exeCmd.execJps(host)
nowProNum = len(re.findall(serverlist.get(host), content))
if nowProNum < proNum:
log.warn('{host} 节点 {proNum} 个 Kafka 服务未运行'.format(
host=host, proNum=(proNum - nowProNum)))
startKafka(host, 'Kafka')
else:
log.info('{host} 节点 {proNum} 个 Kafka 服务正在运行'.format(
host=host, proNum=proNum))