Steps to reproduce:

1. Use 50+ passive proxies with required amount of proxy pollers with configuration sync interval every 1 minute;
2. Get some unplanned network outage, all proxies are not reachable;
3. OOM killer;

Result:
PSS consumption:

Proxy poller are jumping from 8mb to 150mb+:

Private + Shared:
  8.3 MiB + 129.5 KiB =   8.4 MiB	/usr/sbin/zabbix_server: proxy poller #7 [exchanged data with 0 proxies in 0.000017 sec, idle 1 sec]

150.1 MiB + 139.5 KiB = 150.2 MiB	/usr/sbin/zabbix_server: proxy poller #6 [exchanged data with 0 proxies in 0.000014 sec, idle 1 sec]

From performance perspective:

1. not needed work for database to retrieve full configuration each time for each proxy;
2. unpredictable memory consumption;

Expected:
Server continue to work, monitoring system should not kill itself with OOM just because some proxies are died or not reachable.