def check_instance_state(self):
'''
获取所有的instance状态,报告哪些有问题
'''
service_role_map = {}
#get all instance
session = database.getSession()
for instance in session.query(Instance):
result,msg = self.check_instance(instance)
self.update_result(result,msg)
if not service_role_map.has_key(instance.service) :
service_role_map[instance.service] = []
service_role_map[instance.service].append(instance)
for (service,roles) in service_role_map.items():
if service == "zookeeper" :
zk_leader_port = database.get_service_conf(session,"zookeeper","zookeeper_leader_port")
result,msg = self.check_zk_leader(zk_leader_port,roles)
self.update_result(result,msg)
elif service == "hbase" :
#TODO
hbase_master_info_port = database.get_service_conf(session,"hbase","hbase_master_info_port")
result,msg = self.check_hbase_master(hbase_master_info_port,roles)
self.update_result(result,msg)
session.close()
return (self.state,self.msg)
def check_instance_state(self):
'''
获取所有的instance状态,报告哪些有问题
'''
service_role_map = {}
#get all instance
session = database.getSession()
for instance in session.query(Instance):
#检查实例
key_word,msg = self.check_instance(instance)
if key_word != None:
self.update_result(key_word, msg)
#采集服务
if not service_role_map.has_key(instance.service) :
service_role_map[instance.service] = []
service_role_map[instance.service].append(instance)
for (service,roles) in service_role_map.items():
if service == "zookeeper" :
zk_leader_port = database.get_service_conf(session,"zookeeper","zookeeper_leader_port")
result,msg = self.check_zk_leader(zk_leader_port,roles)
if not result:
key_word = "cluster(zk no leader)"
self.update_result(key_word, msg)
elif service == "hbase" :
#TODO
hbase_master_info_port = database.get_service_conf(session,"hbase","hbase_master_info_port")
result,msg = self.check_hbase_master(hbase_master_info_port,roles)
if not result:
key_word = "cluster(hbase no leader)"
self.update_result(key_word, msg)
session.close()
return self.alarm_list
def get_cluster_name():
cluster_name = ""
session = database.getSession()
cluster_name = database.get_service_conf(session, "ganglia",
"cluster_name")
session.close()
return cluster_name
def namenode_web(self):
session = database.getSession()
nm_web_port = database.get_service_conf(session,"hdfs","dfs_namenode_http_address_port")
nms = []
for inst in session.query(Instance).filter(Instance.role == "namenode"):
nms.append(inst.host)
session.close()
return namenode_web.namenode_web(nms, nm_web_port)
def init_config(self):
'''
初始化关于获取端口等的内容
默认从数据库中获取,由于可能只使用监控
所以本函数可以直接修改为固定值
'''
session = database.getSession()
insts = session.query(Instance).filter(Instance.role=="resourcemanager")
for inst in insts:
self.rmhost = inst.host;
self.rmport=database.get_service_conf(session,"yarn","yarn_rm_webapp_port")
insts = session.query(Instance).filter(Instance.role=="historyserver")
for inst in insts:
self.hshost = inst.host;
self.hsport = database.get_service_conf(session,"yarn","mapreduce_jobhistory_webapp_port")
session.close()
def resourcemanager_web(self):
session = database.getSession()
rm_port = database.get_service_conf(session,"yarn","yarn_rm_webapp_port")
rms = []
for inst in session.query(Instance).filter(Instance.role == "resourcemanager"):
rms.append(inst.host)
session.close()
if len(rms) == 0 :
return (contants.ALARM_ERROR, u"%s 检查不到有resourcemanager" % self.rule.name )
if len(rms) != 1 :
return (contants.ALARM_ERROR, u"%s 检查到有多个resourcemanager %s" % (self.rule.name, ",".join(rms) ) )
return resourcemanager_web.resourcemanager_web(rms[0],rm_port)
def init_config(self):
'''
初始化关于获取端口等的内容
默认从数据库中获取,由于可能只使用监控
所以本函数可以直接修改为固定值
'''
session = database.getSession()
insts = session.query(Instance).filter(
Instance.role == "resourcemanager")
for inst in insts:
self.rmhost = inst.host
self.rmport = database.get_service_conf(session, "yarn",
"yarn_rm_webapp_port")
insts = session.query(Instance).filter(
Instance.role == "historyserver")
for inst in insts:
self.hshost = inst.host
self.hsport = database.get_service_conf(
session, "yarn", "mapreduce_jobhistory_webapp_port")
session.close()
def check_instance_state(self):
'''
获取所有的instance状态,报告哪些有问题
'''
service_role_map = {}
#get all instance
session = database.getSession()
for instance in session.query(Instance):
#检查实例
key_word, msg = self.check_instance(instance)
if key_word != None:
self.update_result(key_word, msg)
#采集服务
if not service_role_map.has_key(instance.service):
service_role_map[instance.service] = []
service_role_map[instance.service].append(instance)
for (service, roles) in service_role_map.items():
if service == "zookeeper":
zk_leader_port = database.get_service_conf(
session, "zookeeper", "zookeeper_leader_port")
result, msg = self.check_zk_leader(zk_leader_port, roles)
if not result:
key_word = "cluster(zk no leader)"
self.update_result(key_word, msg)
elif service == "hbase":
#TODO
hbase_master_info_port = database.get_service_conf(
session, "hbase", "hbase_master_info_port")
result, msg = self.check_hbase_master(hbase_master_info_port,
roles)
if not result:
key_word = "cluster(hbase no leader)"
self.update_result(key_word, msg)
session.close()
return self.alarm_list
def resourcemanager_web(self):
session = database.getSession()
rm_port = database.get_service_conf(session,"yarn","yarn_rm_webapp_port")
rms = []
for inst in session.query(Instance).filter(Instance.role == "resourcemanager"):
rms.append(inst.host)
session.close()
alarm_list = []
if len(rms) == 0 :
key_word = "cluster(no rm)"
msg = u"%s 检查不到有resourcemanager" % self.rule.name
return [{"key_word":key_word,"msg":msg}]
if len(rms) != 1 :
key_word = "cluster(too much rm)"
msg = u"%s 检查到有多个resourcemanager %s" % (self.rule.name, ",".join(rms) )
return [{"key_word":key_word,"msg":msg}]
return resourcemanager_web.resourcemanager_web(rms[0],rm_port)
def get_cluster_name():
cluster_name = ""
session = database.getSession()
cluster_name = database.get_service_conf(session,"ganglia","cluster_name")
session.close()
return cluster_name
