history syncer can only reach 25% busy

[westbrook]

Hi, all,

I am using zabbix 2.2.2, my default 4 history syncers can only reach 25% busy no matter how many items added.

If change the default 4 to 10, it turns to 10%, that means only 1 syncer busy no matter how many items to deal.
It waists more time to deal more items,or even cause timeout to connect to port 15001, but not use more syncers at the same time.

Can anybody tell me why?

[kloczek]

Because you are using passive proxies?
Another explanation is that you have only one proxy.

[westbrook]

Thanks for reply!

But I didn't use proxy.

Actually,I use TESTTOOL(LoadRunner/JMeter) to send TCP packages direct to the server.
When the stress is little( eg. 1000vps), I can see (ps -ef|grep syncer) all the syncers working at the same time, but when stress goes up,only one (not the same) working, and the other always deal only several.

#################################
The Strees not so large, all syncers can deal the datas at the same time.

zabbixs 73642 73533 0 12:10 ? 00:00:56 /opt/OpsMonitor/zabbix-server/sbin/zabbix_server: history syncer #1 [synced 1300 items in 0.145747 sec, idle 3 sec]
zabbixs 73643 73533 1 12:10 ? 00:01:53 /opt/OpsMonitor/zabbix-server/sbin/zabbix_server: history syncer #2 [synced 18 items in 0.027077 sec, idle 5 sec]
zabbixs 73644 73533 0 12:10 ? 00:01:42 /opt/OpsMonitor/zabbix-server/sbin/zabbix_server: history syncer #3 [synced 1371 items in 0.320311 sec, idle 5 sec]
zabbixs 73645 73533 0 12:10 ? 00:01:31 /opt/OpsMonitor/zabbix-server/sbin/zabbix_server: history syncer #4 [synced 1289 items in 0.158633 sec, idle 3 sec]

#################################
It seems the stress goes up, only one syncer is busy and waisting more time, but the other is idle.

zabbixs 73642 73533 0 12:10 ? 00:01:04 /opt/OpsMonitor/zabbix-server/sbin/zabbix_server: history syncer #1 [synced 24411 items in 4.886770 sec, syncing history]
zabbixs 73643 73533 1 12:10 ? 00:01:53 /opt/OpsMonitor/zabbix-server/sbin/zabbix_server: history syncer #2 [synced 21 items in 0.063924 sec, idle 5 sec]
zabbixs 73644 73533 0 12:10 ? 00:01:42 /opt/OpsMonitor/zabbix-server/sbin/zabbix_server: history syncer #3 [synced 17 items in 0.003793 sec, idle 5 sec]
zabbixs 73645 73533 0 12:10 ? 00:01:31 /opt/OpsMonitor/zabbix-server/sbin/zabbix_server: history syncer #4 [synced 19 items in 0.005030 sec, idle 5 sec]

[westbrook]

Hi, kloczek,

Maybe you're right, when I use testtools to send packages(with datas about other host) to the server, the server may consider the request was from a proxy?

So how I can do to cheat the server?

[glebs.ivanovskis]

Do you send data in large batches?

[westbrook]

Do you send data in large batches?

Hi,

Yes, I send data in large batches.

The larger I send, the history cache (free), go faster down directly to 0%. When 0%, keep on sending data, parts of requests failed, and it stays at 0%.

Though history cache free is 0%, the history syncer can only reach the top 25% and keep on 25%.

[glebs.ivanovskis]

History syncer needs to process triggers too and for each trigger only one new value can be processed at a time to guarantee that all PROBLEM and OK events are generated and all they come in correct order. This is why sometimes not all history syncer processes are fully loaded.

I think you are sending data for one item or a small set of items and you have a trigger on them. You will achieve better parallelism if you send a mixture of different items not bound by common triggers.