CentOS Linux release 7.3.1611
RDS MySQL - db.m4.xlarge
EC2 Zabbix - m3.2xlarge

• Status

Number of hosts (enabled/disabled/templates)	1081	690 / 18 / 373
Number of items (enabled/disabled/not supported)	104097	94915 / 9037 / 145
Number of triggers (enabled/disabled [problem/ok])	60172	55861 / 4311 [470 / 55391]
Number of users (online)	139	12
Required server performance, new values per second	1240.91	

Please check that this is not ZBX-10486.

Also may be worth setting up some monitoring of Zabbix processes to pinpoint at least the type of processes which are leaking.

Hello Glebs, thanks for ansewer.

Yeah, I remember last year about nss/libcurl leak in ZBX-10486, and I'm surprise that Red Hat doesn't fix this yet!

But this issue affects zabbix 3.0 also, so this can't be related. Also, we don't use "SSL verify peer" in web scenarios (I didn't check 100% of them, but all those I looked was disabled).

I'm collecting the follow data:

• proc.mem[,,avg,php-fpm]
• proc.mem[,zabbix,avg,zabbix-cloudwatch]
• proc.mem[,zabbix,avg,zabbix_agent]
• proc.mem[,zabbix,avg,zabbix_server]

And this is the result. As you can see, the zabbix_server is increasing time after time, slowly, but increasing.

Thanks a lot!

In zabbix-32-memory.png you were losing memory much faster. Or am I missing something?

Regarding monitoring, I would suggest using sum instead of avg. Could you also add these?

• proc.mem[,zabbix,sum,poller];
• proc.mem[,zabbix,sum,trapper];
• etc. (for every process type you see in ps aux | grep zabbix_server)

Hello Glebs!

The memory leak has stoped after kernel, glibc, libnss and libcurl upgrade:

The process that you asked are working good, also. I'll let them collecting and re-open this issue, only if something went out of expectation.

Thanks a lot!

Hm, interesting... I'm glad the issue is resolved in the end!