Troubles with perfomance when use zabbix proxy

[altervast]

Hi everybody!

I have troubles with proxy usage.
My infrastructure: 3 remote networks sites A,B,C
Site A is monitored by Zabbix server itself,
Sites B&C are monitored by zabbix proxies in these sites.

Here are the data from zabbix server:
Number of hosts (monitored/not monitored/templates) 283 192 / 19 / 72
Number of items (monitored/disabled/not supported) 26414 24803 / 13 / 1598
Number of triggers (enabled/disabled) [problem/ok] 5426 3339 / 2087 [26 / 3313]
Required server performance, new values per second 208.4 -

The problems are:
1. Zabbix queue is 6K elements now
2. Values processed by Zabbix server per second are less than 0.5K

The graphs are similar to all the servers (zabbix server and two proxies)

But QUEUE OF ITEMS TO BE UPDATED is different!!!
Proxy 5 seconds 10 seconds 30 seconds 1 minute 5 minutes More than 10 minutes
R-proxy-01 1089 3964 3280 704 314 483
S-proxy-01 52 303 124 72 67 17
Server 33 1 0 0 0 0

In case of decreasing the item new values request time, the queue could reach 20K on all servers quickly.

We have provided a few experiments:
1. We have reconfigured the hosts to report to zabbix server instead of proxy in one site. The queue went down below 0.5K as result.
2. We have disabled from the monitoring turned off hosts. The queue became equivalent to processed value graph.

The question is: What should I configure to be able to use proxy correctly:
1. without data leaks
2. with correctly processed switched off hosts with enabled monitoring.

Proxy configs and my.cnf attached below.

my.cnf of R-proxy-01
max_connections=300
innodb_flush_log_at_trx_commit=2
innodb_buffer_pool_size=4G
innodb_thread_concurrency=8
innodb_flush_method=O_DIRECT
innodb_data_file_path=ibdata1:10M:autoextend
innodb_file_per_table=1
innodb_log_file_size=512M
query_cache_size=128M
join_buffer_size=128M
table_cache=128M
thread_cache_size=128M

zabbix_proxy.conf of R-proxy-01
ProxyMode=0
ProxyLocalBuffer=1
ProxyOfflineBuffer=12
HeartbeatFrequency=60
ConfigFrequency=600
DataSenderFrequency=1

StartPollers=100
StartIPMIPollers=5
StartPollersUnreachable=20
StartTrappers=10
StartPingers=20
CacheSize=500M
StartDBSyncers=20
HistoryCacheSize=500M
HistoryTextCacheSize=500M
Timeout=3
TrapperTimeout=300
UnreachablePeriod=45
UnavailableDelay=60
UnreachableDelay=15

[LenR]

Do the agents list their proxy's IP to allow access?

If you're 2.2, add the zabbix_proxy template to the proxies, check for busy internal and data gathering processes. Also check the internal items on the server.

[Palmertree]

If you use the Zabbix internal monitoring make sure that the proxy server you are monitoring is processed by the same proxy server or not another zabbix server. The metrics returned are for which every device does the polling for Zabbix internal monitoring for zabbix metrics.