因为最近线上的hadoop集群从mrv1升级到mrv2了,监控模板也跟着变动了。。
线上是200台左右的集群,模块采用了link的方式来添加,即一个模板下link大量的模块,然后主机添加到这个模板里。
这样算下来一台机器差不多有 215个item.
为了增加NM的监控,也采用了link的方式来连接模板,在页面上link时发现一直返回一个空白页。
为了快速上线,改变了下方法,使用了host.update的api,直接把host link到NM的模板。
回过头来看这个问题:
在通过页面link模板时,其实也是调用了zabbix template相关的api(具体调用了template.update方法)
直接通过脚本来调用api测试:
测试脚本:
#!/usr/bin/env python
import urllib2
import sys
import json
def requestJason(url,values):
data = json.dumps(values)
print data
req = urllib2.Request(url, data, {'Content-Type': 'application/json-rpc'})
response = urllib2.urlopen(req, data)
data_get = response.read()
output = json.loads(data_get)
print output
try:
message = output['result']
except:
message = output['error']['data']
quit()
print json.dumps(message)
return output
def authenticate(url, username, password):
values = {'jsonrpc': '2.0',
'method': 'user.login',
'params': {
'user': username,
'password': password
},
'id': '0'
}
idvalue = requestJason(url,values)
return idvalue['result']
def getTemplate(hostname,url,auth):
values = {'jsonrpc': '2.0',
'method': 'template.get',
'params': {
'output': "extend",
'filter': {
'host': hostname
}
},
'auth': auth,
'id': '2'
}
output = requestJason(url,values)
print output['result'][0]['hostid']
return output['result'][0]['hostid']
def changeTemplate(idx,id_list,url,auth):
values = {'jsonrpc': '2.0',
'method': 'template.update',
'params': {
"templateid":idx,
"templates":id_list
},
'auth': auth,
'id': '2'
}
output = requestJason(url,values)
print output
def main():
id_list = []
hostname = "Vipshop_Template_OS_Linux_Hadoop_Datanode_Pro"
url = 'xxxx'
username = 'admin'
password = 'xxxx'
auth = authenticate(url, username, password)
idx = getTemplate(hostname,url,auth)
temlist = ['Vipshop_Template_LB_Tengine_8090','Vipshop_Template_Redis_6379','Vipshop_Template_Redis_6380','Vipshop_Template_Redis_6381','Vipshop_Template_Redis_6382','Vipshop_Template_Redis_6383']
for tem in temlist:
idtemp = getTemplate(tem,url,auth)
id_list.append({"templateid":idtemp})
print id_list
#id_list = [{"templateid":'10843'},{"templateid":"10554"},{"templateid":"10467"},{"templateid":"10560"},{"templateid":"10566"},{"templateid":"10105"}]
changeTemplate(idx,id_list,url,auth)
if __name__ == '__main__':
main()
脚本结果:
urllib2.HTTPError: HTTP Error 500: Internal Server Error
因为api其实是发送了一个jason格式的post请求,手动使用curl来验证:
curl -vvv -i -X POST -H 'Content-Type:application/json' -d
'{"params": {"templates": [{"templateid": "10117"}, {"templateid": "10132"}, {"templateid": "10133"}, {"templateid": "10134"},
{"templateid": "10135"}, {"templateid": "10136"}], "templateid": "10464"}, "jsonrpc": "2.0", "method": "template.update", "auth": "421a04b400e859834357b5681a586a5f", "id": "2"}'
http://zabbix.idc.vipshop.com/api_jsonrpc.php
返回500错误(即后端php处理时遇到错误导致),调整php的配置,把日志改成debug格式:
php-fpm.conf:
log_level = debug
在error log中发现如下错误:
[04-May-2014 14:04:32.115189] WARNING: pid 6270, fpm_request_check_timed_out(), line 271: [pool www] child 6294, script '/apps/svr/zabbix/wwwroot/api_jsonrpc.php' (request: "POST /api_jsonrpc.php") executing too slow (1.269946 sec), logging
[04-May-2014 14:04:32.115327] DEBUG: pid 6270, fpm_got_signal(), line 72: received SIGCHLD
[04-May-2014 14:04:32.115371] NOTICE: pid 6270, fpm_children_bury(), line 227: child 6294 stopped for tracing
[04-May-2014 14:04:32.115385] NOTICE: pid 6270, fpm_php_trace(), line 142: about to trace 6294
[04-May-2014 14:04:32.115835] NOTICE: pid 6270, fpm_php_trace(), line 170: finished trace of 6294
[04-May-2014 14:04:32.115874] DEBUG: pid 6270, fpm_event_loop(), line 409: event module triggered 1 events
[04-May-2014 14:04:35.318614] WARNING: pid 6270, fpm_stdio_child_said(), line 166: [pool www] child 6294 said into stderr: "NOTICE: sapi_cgi_log_message(), line 663: PHP message: PHP Fatal error: Allowed memory size of 134217728 bytes exhausted (tried to allocate 512 bytes) in /apps/svr/zabbix/wwwroot/api/classes/CItem.php on line 1088"
[04-May-2014 14:04:35.318665] DEBUG: pid 6270, fpm_event_loop(), line 409: event module triggered 1 events
即在做link模板时,需要把相关的数据放在php的内存中,而默认的设置是128M,如果在item和host比较多的时候,很容易就会超过这个限制。
更改为
重新测试,返回了502 Bad Gateway错误,即后端执行超时导致。
error log:
[04-May-2014 14:50:21.318071] WARNING: pid 4131, fpm_request_check_timed_out(), line 281: [pool www] child 4147, script '/apps/svr/zabbix/wwwroot/api_jsonrpc.php' (request: "POST /api_jsonrpc.php") execution timed out (10.030883 sec), terminating
执行时间超过request_terminate_timeout 设置。导致502产生。
更改 request_terminate_timeout = 1800(默认是10s),max_execution_time = 0(默认30s),重新测试。ok.
小结:
zabbix不同于一般的线上应用,在调用api做更新时,是一个batch的行为,对内存和执行时间有一定的要求。
因此要合理的设置php的相关参数,在debug的时候调低日志级别并开启slow log来方便定位问题.