def interpret_ip(monitor_cuid):
"""
@转换monitorIP
"""
host_object = get_object(host,cuid=monitor_cuid)
process_object = get_object(process,cuid=monitor_cuid)
database_object = get_object(database,cuid=monitor_cuid)
if host_object:
return host_object.ipaddr
if process_object:
return process_object.related_host_cuid.ipaddr
if database_object:
return database_object.related_host_cuid.ipaddr
def convert_monitor_object(monitor_cuid):
"""
@转换monitor对象
"""
host_object = get_object(host,cuid=monitor_cuid)
process_object = get_object(process,cuid=monitor_cuid)
database_object = get_object(database,cuid=monitor_cuid)
if host_object:
return '主机-%s'%host_object.ipaddr.encode('utf-8')
if process_object:
return '应用-%s'%process_object.label_cn.encode('utf-8')
if database_object:
return '数据库-%s'%database_object.label_cn.encode('utf-8')
def process_parse(self):
'''
@进程URL和PID监控方法
'''
try:
logger.debug('开始检测进程PID和URL')
for process_list in self.proccesses:
processobj = get_object(process,
cuid=process_list.monitor_object)
value = {}
if processobj.url_path:
url_result = cm_api.check_url(processobj.url_path)
else:
url_result = 0
pid = get_any_object(
current_process,
related_process_cuid=process_list.monitor_object)
if pid:
pid_result = 0
else:
pid_result = 1
value['related_host_cuid'] = processobj.related_host_cuid
value['pid_result'] = pid_result
value['url_result'] = url_result
value['collect_time'] = time.strftime(
ISOTIMEFORMAT, time.localtime(time.time()))
value['process'] = processobj
self.alarm_object.process_am_parse(value)
logger.debug('进程PID和URL检测结束,THE END!')
except Exception, e:
logger.error(traceback.format_exc())
def templatetags(object):
"""
int 转换为 str
"""
memory_info = get_object(current_memory,related_host_cuid=object)
if memory_info:
return object.cuid
else:
return object.cuid
def __init__(self, data):
'''
@资源采集对象初始化
'''
self.Process_Info = data.get('ProcessPlugin', None)
self.Disk_Info = data.get('DiskPlugin', None)
self.Memory_Info = data.get('MemoryPlugin', None)
self.Cpu_Info = data.get('CpuPlugin', None)
self.Host_info = data.get('host', None)
self.collect_time = data.get('collect_time', None)
self.host = get_object(host, ipaddr=self.Host_info)
self.alarm_object = Hard_AmObj()
def process_am_parse(self, value):
'''
@进程URL和PID告警解析
'''
try:
logger.debug('开始进程告警解析')
value['monitor_object'] = get_object(
monitor,
monitor_object=value['process'].cuid,
monitor_type='PROCESS')
value['tag'] = 'PROCESS'
value['title'] = '进程阈值告警'
if value['monitor_object'] and self.process_monitor_opt:
value['alarm_object'] = alarm.objects.filter(
related_monitor_cuid=value['monitor_object'].cuid,
alarm_tag=value['tag'])
if value['url_result'] == 1:
value['content'] = '进程：%s 产生告警，原因：进程URL：%s拨测失败' % (
value['process'].label_cn.encode('utf-8'),
value['process'].url_path.encode('utf-8'))
if value['alarm_object']:
logger.debug('告警已存在，更新告警时间')
self.alarm_update(value)
else:
logger.debug('开始新增告警')
self.alarm_insert(value)
elif value['pid_result'] == 1:
value['content'] = '进程：%s 产生告警，原因：进程PID不存在' % (
(value['process'].label_cn.encode('utf-8')))
if value['alarm_object']:
logger.debug('告警已存在，更新告警时间')
self.alarm_update(value)
else:
logger.debug('开始新增告警')
self.alarm_insert(value)
else:
if value['alarm_object']:
logger.debug('告警已消除，删除活动告警')
self.alarm_delete(value)
else:
logger.debug(
'主机%s不再管理范围,不进行告警解析' %
((value['related_host_cuid'].cuid).encode('utf-8')))
except Exception, e:
logger.error(traceback.format_exc())
def ping_parse(self):
'''
@PING测主机监控方法
'''
try:
logger.debug('开始PING测所有主机列表')
for host_list in self.ping_list:
value = {}
hostobj = get_object(host, cuid=host_list.monitor_object)
result = cm_api.check_host(hostobj.ipaddr)
value['related_host_cuid'] = hostobj
value['result'] = result
value['collect_time'] = time.strftime(
ISOTIMEFORMAT, time.localtime(time.time()))
self.alarm_object.ping_am_parse(value)
logger.debug('主机PING测结束,THE END!')
except Exception, e:
logger.error(traceback.format_exc())
def __init__(self):
'''
@获取告警监控阈值
'''
self.memory_monitor_opt = get_object(monitor_opt,
monitor_type='MEMORY')
self.disk_monitor_opt = get_object(monitor_opt, monitor_type='DISK')
self.cpu_monitor_opt = get_object(monitor_opt, monitor_type='CPU')
self.ping_monitor_opt = get_object(monitor_opt, monitor_type='PING')
self.process_monitor_opt = get_object(monitor_opt,
monitor_type='PROCESS')
self.heart_monitor_opt = get_object(monitor_opt, monitor_type='HEART')
def ping_am_parse(self, value):
'''
@主机PING测告警解析
'''
try:
logger.debug('开始主机PING测告警解析')
value['monitor_object'] = get_object(
monitor,
monitor_object=value['related_host_cuid'].cuid,
monitor_type='PING')
value['tag'] = 'PING'
value['title'] = 'PING测阈值告警'
if value['monitor_object'] and self.ping_monitor_opt:
value['alarm_object'] = alarm.objects.filter(
related_monitor_cuid=value['monitor_object'].cuid,
alarm_tag=value['tag'])
if float(value['result']) > float(
self.ping_monitor_opt.monitor_alarm_value):
value[
'content'] = 'PING测主机：%s ,当前丢包率:%s%%,产生告警，原因：ping测丢包率大于%s%%' % (
value['related_host_cuid'].ipaddr.encode('utf-8'),
float(value['result']),
self.ping_monitor_opt.monitor_alarm_value)
if value['alarm_object']:
logger.debug('告警已存在，更新告警时间')
self.alarm_update(value)
else:
logger.debug('开始新增告警')
self.alarm_insert(value)
else:
if value['alarm_object']:
logger.debug('告警已消除，删除活动告警')
self.alarm_delete(value)
else:
logger.debug(
'主机%s不再管理范围,不进行告警解析' %
((value['related_host_cuid'].cuid).encode('utf-8')))
except Exception, e:
logger.error(traceback.format_exc())
def disk_am_parse(self, value):
'''
@硬盘告警解析
'''
try:
logger.debug('开始硬盘告警解析')
value['monitor_object'] = get_object(
monitor,
monitor_object=value['related_host_cuid'].cuid,
monitor_type='DISK')
value['tag'] = value['device_name']
value['title'] = '硬盘阈值告警'
if value['monitor_object'] and self.disk_monitor_opt:
value['alarm_object'] = alarm.objects.filter(
related_monitor_cuid=value['monitor_object'].cuid,
alarm_tag=value['tag'])
if float(value['device_percent']) > float(
self.disk_monitor_opt.monitor_alarm_value):
value[
'content'] = '当前磁盘：%s 磁盘利用率:%s%%,产生告警，原因：磁盘利用率大于%s%%' % (
value['device_name'].encode('utf-8'),
float(value['device_percent']),
float(self.cpu_monitor_opt.monitor_alarm_value))
if value['alarm_object']:
logger.debug('告警已存在，更新告警时间、告警次数')
self.alarm_update(value)
else:
logger.debug('开始新增告警')
self.alarm_insert(value)
else:
if value['alarm_object']:
logger.debug('告警已消除，删除活动告警')
self.alarm_delete(value)
else:
logger.debug(
'主机%s不再管理范围,不进行告警解析' %
((value['related_host_cuid'].cuid).encode('utf-8')))
except Exception, e:
logger.error(traceback.format_exc())
def memory_am_parse(self, value):
'''
@内存告警解析
'''
try:
logger.debug('开始内存告警解析')
value['monitor_object'] = get_object(
monitor,
monitor_object=value['related_host_cuid'].cuid,
monitor_type='MEMORY')
value['tag'] = 'memory'
value['title'] = 'MEMORY阈值告警'
if value['monitor_object'] and self.memory_monitor_opt:
value['alarm_object'] = alarm.objects.filter(
related_monitor_cuid=value['monitor_object'].cuid,
alarm_tag=value['tag'])
if float(value['memory_percent']) > float(
self.cpu_monitor_opt.monitor_alarm_value):
value['content'] = '内存利用率:%s%%,产生告警，原因：内存利用率大于%s%%' % (
float(value['memory_percent']),
float(self.memory_monitor_opt.monitor_alarm_value))
if value['alarm_object']:
logger.error('告警已存在，更新告警时间')
self.alarm_update(value)
else:
logger.error('开始新增告警')
self.alarm_insert(value)
else:
if value['alarm_object']:
logger.error('告警已消除，删除活动告警')
self.alarm_delete(value)
else:
logger.debug(
'主机%s不再管理范围,不进行告警解析' %
((value['related_host_cuid'].cuid).encode('utf-8')))
except Exception, e:
logger.error(traceback.format_exc())
def oracle_mviews_parse(self):
'''
@数据库物化视图刷新监控方法
'''
logger.debug('开始物化视图刷新状态监测')
for mview in self.mview_list:
value = {}
database_info = get_object(database, cuid=mview.monitor_object)
value['related_host_cuid'] = database_info.related_host_cuid
value['database'] = database_info
value['monitor_object'] = mview
value['collect_time'] = time.strftime(ISOTIMEFORMAT,
time.localtime(time.time()))
sql = "select mview_name,last_refresh,last_sync_exception from meta_mview where isactive='Y'"
result_list = cm_api.exec_sql(database_info.user_name,
database_info.passwd,
database_info.ipaddr,
database_info.port,
database_info.sid, sql)
for result in result_list:
value['mview_name'] = result[0]
value['last_refresh_time'] = result[1]
value['last_sync_exception'] = result[2]
self.alarm_object.mview_am_parse(value)