Are there already some ideas or plans to feed JMX data of Java processes into ES?
Maybe Jolokia is an option as it provides an HTTP/JSON interface to the MBeanServer within a JVM:
Maybe a new input_type (like http) in FileBeat where it pulls for all/filtered JMX data or a separate Beat?
Logstash itself is too heavy to install as an agent on the servers where the applications run and the jmx input plugin in logstash might not scale that good to grab the JMX data from central logstash instances from hundreds of Java processes. IMHO, the data should be fetched locally and then fed into ES (via Logstash for filtering etc.)