MongoDB does something interesting that we aren’t used to in the Java world – if you use the
MongoClient object, it manages the pooling for you on the server side. This is different than how most databases connect, which leaves pooling up to the application or the Java server.
Anyway, I was working with a Grails application that used
MongoClient and the Mongos kept running out of memory. Therefore tickets were opened, words were said, and fingers were pointed. In the past, they had used Grails connection pooling for Mongo, ran into these problems, and switched to a single
MongoClient. They said it was happening again but we developers needed evidence to show that (or, disprove it).
Doing some research showed that the Java Driver for Mongo has some good JMX endpoints to monitor. But the problem was that we needed to track these values over a long period of time – more than just what VisualVM would show. And I didn’t have access to the server but they did have the Grails console running inside the applications.
So I wrote a simple Groovy script that queries the Mongo’s JMX values every second and prints it out in something resembling a CSV file. I wrote this to run inside a Grails console, but there is no Grails in it – it’s pure Groovy. You may have to make minor changes to get it to run on the command line. Anyway, here is the Gist:
Also published on Medium.