def modules(cc_biz_id, set_id_list=None):
"""获取业务下的模块列表，按机器数量排序"""
with ignored(Exception):
ips = set([])
if set_id_list and (not hasattr(set_id_list, "__iter__")):
set_id_list = [set_id_list]
if set_id_list is not None:
set_id_list = ";".join(map(str, set_id_list))
module_list = client.cc.get_modules_by_property(
app_id=cc_biz_id, set_id=set_id_list).get("data") or []
module_id_list = [
module_item["ModuleID"] for module_item in module_list
]
host_list = CCBiz._hosts(cc_biz_id).get("data") or []
if module_id_list:
module_info = collections.defaultdict(set)
for host in host_list:
if host["ModuleID"] in module_id_list:
module_info[host["ModuleID"]].add(host_key(host))
module_info["0"].add(host_key(host))
for set_item in module_list:
set_item["host_count"] = len(
module_info[set_item["ModuleID"]])
set_item["ModuleName"] += ' [%s]' % set_item["host_count"]
module_list = sorted(module_list,
key=lambda x: x['host_count'],
reverse=True)
return module_list, len(module_info["0"])
return [], 0
def agent_status(cc_biz_id, host_id_list):
"""获取agent状态信息
agent状态详细分成4个状态：正常，离线，未安装。已安装，无数据。
"""
result = collections.defaultdict(int)
with ignored(Exception):
ip_info_list = list()
if not hasattr(host_id_list, "__iter__"):
host_id_list = [host_id_list]
for host_id in host_id_list:
ip, plat_id = parse_host_id(host_id)
plat_id = plat_id if plat_id != "1" else "0"
ip_info_list.append({"ip": ip, "plat_id": plat_id})
if not ip_info_list:
return {}
status_list = client.job.get_agent_status(
app_id=cc_biz_id, ip_infos=ip_info_list,
is_real_time=1).get("data") or []
for info in status_list:
plat_id = info["plat_id"]
ip = info["ip"]
host_id = host_key(ip=ip, plat_id=plat_id)
exist = bool(info["status"])
if not exist:
result[host_id] = AGENT_STATUS.NOT_EXIST
continue
else:
result[host_id] = AGENT_STATUS.ON
return result
def sets(cc_biz_id):
"""获取业务下SET列表，按机器数量排序"""
# 获取set列表
with ignored(Exception):
ips = set([])
set_list = client.cc.get_sets_by_property(
app_id=cc_biz_id).get("data") or []
set_id_list = [set_item["SetID"] for set_item in set_list]
host_list = CCBiz._hosts(cc_biz_id).get("data") or []
if set_id_list:
set_info = collections.defaultdict(set)
for host in host_list:
if host["SetID"] in set_id_list:
set_info[host["SetID"]].add(host_key(host))
set_info["0"].add(host_key(host))
for set_item in set_list:
set_item["host_count"] = len(set_info[set_item["SetID"]])
set_item["SetName"] += ' [%s]' % set_item["host_count"]
set_list = sorted(set_list,
key=lambda x: x['host_count'],
reverse=True)
return set_list, len(set_info["0"])
return [], 0
def set_module_ips(cc_biz_id, set_id="", module_id=""):
"""获取业务下大区下模块的ip列表，根据平台id和ip组合去重"""
ips = set(list())
with ignored(Exception):
if module_id:
result = client.cc.get_module_host_list(app_id=cc_biz_id,
module_id=module_id)
else:
result = client.cc.get_hosts_by_property(app_id=cc_biz_id,
set_id=set_id)
if result["result"]:
for host in (result.get("data") or []):
ips.add(host_key(host))
return sorted(list(set(ips)))
def get_agent_status(request, cc_biz_id):
"""获取agent安装状态
:param request:
:param cc_biz_id:
:return:
"""
hosts_list = CCBiz.hosts(cc_biz_id).get('data') or []
hostid_list = [host_key(h) for h in hosts_list]
agent_status_info = CCBiz.agent_status(cc_biz_id, hostid_list)
if hosts_list and not len(agent_status_info):
return render_json({
'result': False,
'message': u"查询agent状态接口失败",
'data': []
})
agent_fail_cnt = agent_ok_cnt = 0
ok_ip_list = list()
fail_ip_list = list()
for hostid, status in agent_status_info.iteritems():
ip, plat_id = parse_host_id(hostid)
if status == AGENT_STATUS.ON:
agent_ok_cnt += 1
ok_ip_list.append({'ip': ip, 'cc_plat_id': plat_id})
else:
agent_fail_cnt += 1
fail_ip_list.append({'ip': ip, 'cc_plat_id': plat_id})
details = {
'agent_ok_cnt': agent_ok_cnt,
'agent_fail_cnt': agent_fail_cnt,
'ok_ip_list': ok_ip_list,
'fail_ip_list': fail_ip_list,
}
res = {
'result': True,
'message': "",
'data': agent_ok_cnt,
'details': details
}
return render_json(res)
def index(request, cc_biz_id):
"""
request.method:
get: 基础性能页面首页
post: 主机性能信息，包括主机属性，主机各指标信息。
"""
def get_access_status(cc_biz_id):
"""
获取数据平台基础性能接入进展
:return: {
"accessed": True,
"div_message": u"正在接入中",
"btn_message": u"确定"
}
"""
btn_message = u"确定"
div_message = u"您的性能指标采集任务已下发正在进行中，请稍候再试！"
# step 1 获取该业务下基础性能接入的data_id
data_id_exist = any(get_bp_data_id(cc_biz_id))
# step 2 如果data_id 不存在，则直接返回未接入
if not data_id_exist:
div_message = (u"检测到您的业务尚未开启监控，"
u"点击下面按钮开启主机指标采集，\n"
u"10-20分钟后刷新此页面即可查看到主机数据，请耐心等待！\n"
u"（未安装agent的主机，"
u"在主机详情页中按照指引完成部署并开启数据采集。）")
btn_message = u"开始采集"
return {
"accessed": data_id_exist,
"div_message": div_message,
"btn_message": btn_message
}
data = {
"hosts": [],
"update_time": datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
}
if request.method == "GET":
count_per_page = 10
# 获取云区域信息
plat_info = CCBiz.plat_info(cc_biz_id)
checkhost_id = request.GET.get("alarm_strategy_id")
return render_mako_context(request, '/monitor/performanceV2/home.html',
locals())
ts = TimeStats(u"get_host_info_list: %s" % cc_biz_id)
# 获取所有主机信息（前台分页）
ts.split(u"获取主机列表")
hosts = CCBiz.hosts(cc_biz_id).get("data") or []
host_ids = set([host_key(h) for h in hosts])
ts.split(u"获取主机agent状态")
hosts_agent_status = CCBiz.agent_status(cc_biz_id, host_ids)
default_status_info = {
'alarm': {
'level': [0, 0, 0],
},
'component': [],
}
try:
# 主机性能数据结果表字典，当性能获取需要查询之前查过的结果表的时候，
# 直接使用已有的client，节省查询时间。
# cpu 总使用率
ts.split(u"获取主机cpu 总使用率")
cpu_usage_info = HostIndex.get_host_performance(
cc_biz_id,
item="cpuusage",
category="cpu",
result_table_id="ja_gse_cpu_cpuusage")
# CPU 5分钟负载
ts.split(u"获取主机cpu 5分钟负载")
cpu_load_info = HostIndex.get_host_performance(
cc_biz_id,
item="locdavg",
category="cpu",
result_table_id="ja_gse_cpu")
# cpu 单核使用率
ts.split(u"获取主机cpu 单核使用率")
cpu_single_usage_info = HostIndex.get_host_performance(
cc_biz_id,
item="cpuusage",
category="cpu",
result_table_id="ja_gse_cpu_core_cpuusage")
# 磁盘使用量 (暂时隐藏)
# disk_used_info = get_host_performance(cc_biz_id, item="used_percent",
# category="disk", result_table_id="ja_gse_disk_used")
# IO等待
ts.split(u"获取主机IO等待")
io_util_info = HostIndex.get_host_performance(
cc_biz_id,
item="util",
category="disk",
result_table_id="ja_gse_disk_iostats")
host_list = list()
ts.split(u"处理整合所有数据")
for h in hosts:
host = Host(h)
for k, v in default_status_info.iteritems():
setattr(host, k, v)
host.is_stickied = MonitorHostSticky.objects.host_is_stickied(
host.id)
# 基础性能数据
host.cpu_usage = cpu_usage_info.get(host.id)
host.cpu_single_usage = cpu_single_usage_info.get(host.id)
host.cpu_load = cpu_load_info.get(host.id)
host.io_util = io_util_info.get(host.id)
if host.io_util and isinstance(host.io_util.get("val"), list):
io_util = dict()
for _io_util in host.io_util["val"]:
for k, v in _io_util.iteritems():
io_util[k] = v
host.io_util["val"] = max(io_util.values())
host.status = hosts_agent_status.get(host.id, AGENT_STATUS.UNKNOWN)
index_list = [
host.cpu_usage, host.cpu_single_usage, host.cpu_load,
host.io_util
]
if host.status == AGENT_STATUS.ON and not any(index_list):
host.status = AGENT_STATUS.NO_DATA
# 控制前端页面展示的参数
# 是否被选中
host.checked = False
# 是否显示（按属性分组时）
host._is_show = True
host_list.append(host)
host_list.sort(key=lambda x: x.status, reverse=True)
data["hosts"] = host_list
return render_json(ok_data(data))
except (TableNotExistException, SqlQueryException) as e:
# 前端引导用户接入
if not hosts:
data["need_access"] = False
data["access_div_message"] = (
u"检测到当前业务没有主机，请前往 %s 快速部署主机！" %
href_link(u"Agent安装", AGENT_SETUP_URL))
data["access_btn_message"] = u"确定"
else:
host_list = list()
for h in hosts:
host = Host(h)
for k, v in default_status_info.iteritems():
setattr(host, k, v)
host.is_stickied = MonitorHostSticky.objects.host_is_stickied(
host.id)
# 基础性能数据
host.cpu_usage = None
host.cpu_single_usage = None
host.cpu_load = None
host.io_util = None
host.status = hosts_agent_status.get(host.id,
AGENT_STATUS.UNKNOWN)
if host.status == AGENT_STATUS.ON:
host.status = AGENT_STATUS.NO_DATA
# 控制前端页面展示的参数
# 是否被选中
host.checked = False
# 是否显示（按属性分组时）
host._is_show = True
host_list.append(host)
host_list.sort(key=lambda x: x.status, reverse=True)
data["hosts"] = host_list
access_status = get_access_status(cc_biz_id)
data["need_access"] = not access_status.get("accessed")
if isinstance(e, SqlQueryException):
logger.exception(u"数据平台查询失败：%s" % e)
data["access_div_message"] = u"数据查询异常，请联系管理员"
else:
data["access_div_message"] = access_status.get("div_message")
data["access_btn_message"] = access_status.get("btn_message")
return render_json(failed_data(u"", data))
# except SqlQueryException as e:
# logger.exception(u"数据平台查询失败：%s" % e)
# return render_json(failed_data(u"数据查询异常", data))
except Exception as e:
logger.exception(u"拉取主机性能信息失败: %s" % e)
return render_json(failed_data(u"系统错误", data))
finally:
ts.stop()
time_stats_info = ts.display()
logger.warning(time_stats_info)