def add_host(asyncId, hosts, login):
#保留10秒后清除
session = database.getSession()
idMap = {}
progress = 0
progressStep = int((100 - 10) / len(hosts))
#创建任务
for hostName in hosts:
taskId = database.build_task(session, "async", "add_host", hostName,
"machine", "connect")
idMap[hostName] = taskId
async_push(asyncId, "progressMsg", "build task")
progress = 10
async_set(asyncId, "progress", progress)
#交给ansible运行
async_push(
asyncId, "progressMsg",
"ansible connect and install yum python json.As the yum update automatically,it could be long..."
)
progress = 20
async_set(asyncId, "progress", progress)
newlogin = get_default_login(session, login)
# (user,port,passwd,sudopasswd)
for host in hosts:
taskId = idMap[host]
database.update_task(session, taskId, Task.STATUS_RUNNING,
Task.RESULT_UNFINISH, "")
#即使不实际执行ansible
#也可以获取相关的信息
#要求信息是准确可用的
import ansible_lib
ret = ansible_lib.connect_host(hosts, newlogin[0], newlogin[1],
newlogin[2], newlogin[3])
#假连接
#ret=fade_connect_host(hosts,newlogin[0],newlogin[1],newlogin[2],newlogin[3])
app_log.info("connect with" + str(hosts) + " with :" + str(newlogin))
#处理结果
success = []
failed = []
prepareTaskids = []
for hostName in ret:
taskId = idMap[hostName]
(result, msg, info) = ret[hostName]
if result:
#成功连接，添加到host表
success.append(hostName)
host = Host(hostName, Host.STATUS_CONNECTED)
host.ip = info['ip']
host.cpu = info['cpu']
host.mem = info['mem']
host.disk = info['disk']
host.rack = info['rack']
session.add(host)
session.commit()
if config.install_manager:
#提交一个prepare任务,假如运行成功将修改状态到ready
prepareTaskid = database.build_task(session, "ansible",
"prepare", hostName,
"prepare", "prepare")
prepareTaskids.append(prepareTaskid)
callback_lib.add_callback(session, prepareTaskid,
"changeHostToReady")
#假如登录信息跟默认的不一致，保存这些登录信息到数据库
save_login(session, hostName, login)
#更新任务状态
database.update_task(session, taskId, Task.STATUS_FINISH,
Task.RESULT_SUCCESS, "")
progress += progressStep
async_push(asyncId, "progressMsg", "finish connect " + hostName)
async_set(asyncId, "progress", progress)
else:
failed.append(hostName)
database.update_task(session, taskId, Task.STATUS_FINISH,
Task.RESULT_FAILED, msg)
#全部完成
async_push(asyncId, "progressMsg",
"success:" + ",".join(success) + " failed:" + ",".join(failed))
async_set(asyncId, "progress", "100")
#发送消息到worker
retMsg = ""
msg = ','.join([str(id) for id in prepareTaskids])
if not mm.send(msg):
retMsg = "send message to worker error"
#10秒后清除内存任务
time.sleep(10)
async_remove(asyncId)
session.close()
def add_host(asyncId,hosts,login):
#保留10秒后清除
session=database.getSession()
idMap = {}
progress = 0 ;
progressStep = int( (100-10)/len(hosts) )
#创建任务
for hostName in hosts:
taskId = database.build_task(session,"async","add_host",hostName,"machine","connect")
idMap[hostName] = taskId
async_push(asyncId,"progressMsg","build task")
progress = 10
async_set(asyncId,"progress",progress)
#交给ansible运行
async_push(asyncId,"progressMsg","ansible connect and install yum python json.As the yum update automatically,it could be long...")
progress = 20
async_set(asyncId,"progress",progress)
newlogin = get_default_login(session, login)
# (user,port,passwd,sudopasswd)
for host in hosts:
taskId = idMap[host]
database.update_task(session,taskId,Task.STATUS_RUNNING,Task.RESULT_UNFINISH,"")
if config.fade_windows:
ret=fade_connect_host(hosts,newlogin[0],newlogin[1],newlogin[2],newlogin[3])
else:
import ansible_lib
ret=ansible_lib.connect_host(hosts,newlogin[0],newlogin[1],newlogin[2],newlogin[3])
app_log.info("connect with"+str(hosts)+" with :"+str(newlogin) )
#处理结果
success=[]
failed=[]
prepareTaskids=[]
for hostName in ret:
taskId = idMap[hostName]
(result,msg,info) = ret[hostName]
if result:
#成功连接，添加到host表
success.append(hostName)
host = Host(hostName,Host.STATUS_CONNECTED)
host.ip=info['ip']
host.cpu=info['cpu']
host.mem=info['mem']
host.disk=info['disk']
host.rack=info['rack']
session.add(host)
session.commit()
if config.install_manager:
#提交一个prepare任务,假如运行成功将修改状态到ready
prepareTaskid = database.build_task(session,"ansible","prepare",hostName,"prepare","prepare")
prepareTaskids.append(prepareTaskid)
callback_lib.add_callback(session,prepareTaskid,"changeHostToReady")
#假如登录信息跟默认的不一致，保存这些登录信息到数据库
save_login(session, hostName, login)
#更新任务状态
database.update_task(session,taskId,Task.STATUS_FINISH,Task.RESULT_SUCCESS,"")
progress += progressStep
async_push(asyncId,"progressMsg","finish connect "+hostName)
async_set(asyncId,"progress",progress)
else:
failed.append(hostName)
database.update_task(session,taskId,Task.STATUS_FINISH,Task.RESULT_FAILED,msg)
#全部完成
async_push(asyncId,"progressMsg","success:"+",".join(success)+" failed:"+",".join(failed))
async_set(asyncId,"progress","100")
#发送消息到worker
retMsg = ""
msg = ','.join([str(id) for id in prepareTaskids])
if not mm.send(msg):
retMsg = "send message to worker error"
#10秒后清除内存任务
time.sleep(10)
async_remove(asyncId)
session.close()
def deal_alarm_list(self, alarm_list):
new_key_word, old_key_word = self.diff_key_word(alarm_list)
mail_center.push_key_word_map(self.key_word_map)
session = database.getSession()
# 更新数据库的alarm_now表
for alarm_now in session.query(AlarmNow):
session.delete(alarm_now)
for (key_word, info) in self.key_word_map.items():
end = key_word.find("(")
host = key_word[0:end]
session.add(AlarmNow(key_word, host, info["msg"], "ERROR", info["count"], int(time.time())))
session.commit()
# 根据连续告警次数尝试进行修复动作
try:
fix_list = []
for auto_fix in session.query(AlarmAutofix):
fix_list.append(auto_fix.format())
for (key_word, count) in self.key_word_map.items():
for auto_fix in fix_list:
match = auto_fix["pattern"].match(key_word)
if match and (count == auto_fix["count"] or count == auto_fix["count"] * 2):
end = key_word.find("(")
host = key_word[0:end]
log.info(" build the auto fix tasks: %s %s %s " % (host, auto_fix["role"], auto_fix["task"]))
database.build_task(
session, "auto_fix", auto_fix["service"], host, auto_fix["role"], auto_fix["task"]
)
except:
log.exception("autofix catch exception")
try:
# 发出邮件 记录动作
ignore_key_word = session.query(AlarmAssist).filter(AlarmAssist.name == "ignore_key_word").first()
ignore_list = []
if ignore_key_word != None:
# ignore_list = ignore_key_word.value.split(",")
for str in ignore_key_word.value.split(","):
if len(str) > 1:
ignore_list.append(re.compile(str))
# log.info(ignore_list)
for alarm_state in new_key_word:
key_word = alarm_state["key_word"]
session.add(AlarmList(alarm_state["key_word"], "", alarm_state["msg"], "ERROR", int(time.time())))
if self.is_match(ignore_list, key_word):
log.info("ignore %s" % key_word)
else:
self._callback(alarm_state)
for alarm_state in old_key_word:
key_word = alarm_state["key_word"]
session.add(AlarmList(alarm_state["key_word"], "", alarm_state["msg"], "INFO", int(time.time())))
if self.is_match(ignore_list, key_word):
log.info("ignore %s" % key_word)
else:
self._callback(alarm_state)
session.commit()
except:
log.exception("deal callback catch exception")
session.close()
def deal_alarm_list(self, alarm_list):
new_key_word, old_key_word = self.diff_key_word(alarm_list)
mail_center.push_key_word_map(self.key_word_map)
session = database.getSession()
#更新数据库的alarm_now表
for alarm_now in session.query(AlarmNow):
session.delete(alarm_now)
for (key_word, info) in self.key_word_map.items():
end = key_word.find('(')
host = key_word[0:end]
session.add(
AlarmNow(key_word, host, info['msg'], "ERROR", info['count'],
int(time.time())))
session.commit()
#根据连续告警次数尝试进行修复动作
try:
fix_list = []
for auto_fix in session.query(AlarmAutofix):
fix_list.append(auto_fix.format())
for (key_word, count) in self.key_word_map.items():
for auto_fix in fix_list:
match = auto_fix['pattern'].match(key_word)
if match and (count == auto_fix['count']
or count == auto_fix['count'] * 2):
end = key_word.find('(')
host = key_word[0:end]
log.info(" build the auto fix tasks: %s %s %s " %
(host, auto_fix['role'], auto_fix['task']))
database.build_task(session, "auto_fix",
auto_fix['service'], host,
auto_fix['role'], auto_fix['task'])
except:
log.exception("autofix catch exception")
try:
#发出邮件 记录动作
ignore_key_word = session.query(AlarmAssist).filter(
AlarmAssist.name == "ignore_key_word").first()
ignore_list = []
if ignore_key_word != None:
#ignore_list = ignore_key_word.value.split(",")
for str in ignore_key_word.value.split(","):
if len(str) > 1:
ignore_list.append(re.compile(str))
#log.info(ignore_list)
for alarm_state in new_key_word:
key_word = alarm_state['key_word']
session.add(
AlarmList(alarm_state['key_word'], "", alarm_state['msg'],
"ERROR", int(time.time())))
if self.is_match(ignore_list, key_word):
log.info("ignore %s" % key_word)
else:
self._callback(alarm_state)
for alarm_state in old_key_word:
key_word = alarm_state['key_word']
session.add(
AlarmList(alarm_state['key_word'], "", alarm_state['msg'],
"INFO", int(time.time())))
if self.is_match(ignore_list, key_word):
log.info("ignore %s" % key_word)
else:
self._callback(alarm_state)
session.commit()
except:
log.exception("deal callback catch exception")
session.close()