Problem description
There is no connection pool and no bulk processing. Every check will create a new connection to specified endpoint and will hold it until all the data is received. This leads to the problem when 100 JMX metrics for the same endpoint can cause 200 open connections on Java Gateway (one for incoming connection and one for connection to endpoint).

Possible fix
The problem can be fixed by increasing the limits (ulimit or fs.file-max in /etc/sysctl.conf), setting bigger update intervals or by scaling Java Gateways so there are not so many active connections required at once.