A few days ago, just discussed this problem with a group of friends, he is the SNMP switch when the diagram is intermittent.
I also have this case, but when monitoring Tomcat, Jstat consumes more than 90 CPU,
Cause the normal monitoring project to break the diagram.
In fact, it let us wait for the seconds, then the meaning constantly superimposed 15s, is not 5 minutes after not out of the picture,
Or is the load high not out of the picture?
Case
Zabbix_server.log
12961:20170512:113910.624 Zabbix Agent Item "Java.discovery_status[service_tomcat,all]" on host "172.x.x.x" failed: First network error, wait for seconds
Investigation
Top
Top-14:34:25 up 344 days, 21:11, 1 user, load average: 1.23, 1.09, 1.10
tasks:176 Total, 1 running, 174 sleeping, 1 stopped, 0 zombie
Cpu (s): 25.0%us, 5.0%sy, 0.0%ni, 69.7%id, 0.1%wa, 0.0%hi, 0.2%si, 0.0%st
mem:3925304k Total, 3672532k used, 252772k free, 174944k buffers
swap:4128764k Total, 79512k used, 4049252k free, 2122448k cached
PID USER PR NI VIRT RES SHR S%cpu%MEM time+ COMMAND
2484 Tomcat 0 3333m 30m 12m S 99.8 0.8 9065:51 Jstat
Describes the Jstat exception, which is the Zabbix script used to obtain Tomcat data.
[[email protected] local]# ps -ef|grep jstatroot 2483 1 0 may06 ? 00:00:00 sudo -u tomcat /usr/java/jdk/bin/jstat -gcutil 12470tomcat 2484 2483 99 May06 ? 6-07:03:49 /usr/java/jdk/bin/jstat -gcutil 12470root 17095 32501 0 14:32 pts/0 00:00:00 grep jstat[[email protected] local]# ps -ef|grep jstatroot 2483 1 0 May06 ? 00:00:00 sudo -u tomcat /usr/java/jdk/bin/jstat -gcutil 12470tomcat 2484 2483 99 may06 ? 6-07:04:20 /usr/java/jdk/bin /jstat -gcutil 12470root 17097 32501 0 14:32 Pts/0 00:00:00 grep jstat solve kill -9 2484 is normal, restart Zabbix_agentd, Observe a load of about 0.5. At first the suspect was a trigger problem, and the trigger was disabled. As to the cause of this problem, it is possible that there are a few of the Python scripts that are forced to kill the Tomcat process jstat_status.py call Jstat, is it a zombie process? Or the cause of this program's deadlock? Is there anything special about Python? 1 more needs to be done. You need to analyze zabbix server-side logs with python or elk to find a lot of not suitable for value type [numeric (unsigned)] and data type [decimal]2. A process that occupies cpu 80 above is a problem, This also needs to be added to the monitoring
The other reasons for the broken diagram are as follows:
1. Database optimization such as threads_created | 49258532.zabbix Server Optimization
650) this.width=650; "Src=" https://s2.51cto.com/wyfs02/M01/95/75/wKioL1kVW9nBxPXGAAAVErieoi0037.png-wh_500x0-wm_ 3-wmp_4-s_200992409.png "style=" Float:none; "title=" QQ picture 20170512145250.png "alt=" Wkiol1kvw9nbxpxgaaaverieoi0037.png-wh_50 "/>
650) this.width=650; "Src=" https://s2.51cto.com/wyfs02/M02/95/75/wKiom1kVW9mh7FMMAAARJCSRwQ4311.png-wh_500x0-wm_ 3-wmp_4-s_2311014138.png "style=" Float:none; "title=" QQ picture 20170512145244.png "alt=" Wkiom1kvw9mh7fmmaaarjcsrwq4311.png-wh_50 "/>
3. Refer to the menu of light and light blog, from SQL to troubleshoot HTTP://CAIGUANGGUANG.BLOG.51CTO.COM/1652935/1377089/4. Nic Problem http://www.fyluo.com/?post=1645. ZABBIX Agent executes key, obtains data time exceeds 30s reference http://blog.chinaunix.net/ Uid-8108137-id-4591738.html Solution The maximum time to modify the agent execution key, the source code limits the maximum time of 30S. To modify the source code. and tested only to modify the maximum execution time of the AGETN is not possible, the server or proxy side of the maximum execution time is also 30S. Only modify the agent side 30s,item can capture the mode of active.
The stability of the system is the precondition of the server stability.
This article is from the "Python Operation" blog, declined reprint!
Break diagram first network error, wait for seconds