In a situation when agent is being intensively polled by various servers, proxies and zabbix_get commands (at a rate which exceeds ^{number of listeners} / _{time per check}) a queue of connections to be accepted builds up. This means that listeners will accept connections when some fraction of the server/proxy/zabbix_get timeout has already passed and this will make it even harder for them to keep up.

Problem can escalate uncontrollably and soon listeners will accept connections that have been timed out on server/proxy/zabbix_get side, will perform a check nobody has interest for and will attempt to send back a result nobody is waiting for.

Currently agent does not check if connection has been closed by peer and it takes a very long time to go through awaiting connection queue. In very severe cases only agent restart can help.