Why does zabbix monitor cpu utilization have a negative value？

[Nathan]

Everyone, I have encountered a strange thing, the details are as follows:
background:
Zabbix version is 5.0.2, both server and client are used in production environment
There are 2 java projects running on the client host
The client and server of zabbix transmit information through the cross-public network in the active mode

problem:
The server shows that the CPU utilization monitoring value of the host suddenly becomes a large negative value in an instant (see the attachment for the monitoring legend). I am not sure if it is because the zabbix server and the client are transmitted through the public network. The data is caused, but the cause cannot be determined. How can I analyze the cause of the problem or if you know what is causing it, please tell me
thanks to everyone who answered

[Nathan]

The monitoring item item has always used zabbix's default cpu monitoring template,
Both the client and the server are CentOS 7, and the data is sent to two zabbix servers on the client host. I checked another monitoring platform and it shows that there is no problem. The monitoring cpu utilization has always been a non-negative value. , Only this time is a negative value. I think the problem should not lie in the client configuration item or monitoring template. I have an assumption that when the client program is running, the value obtained by the agent is correct, but is it possible When the network is abnormal or fluctuates, the value received by the server appears to be a non-positive integer

[cyber]

Currently too little info to make any assumptions...
item config? Host OS etc...?