Lately, I have been finding the need to get into the bowels of Grails in order to increase the performance of large batch type features of any application. Most of the time the issue appears to be with the Hibernate session becoming over-saturated with references. Performance degrades significantly even when limiting the batch size to ~100 records. Luckily, Oracle ships a pretty decent profiler with Java, VisualVM. The trick is getting it to work correctly with a Grails application.
Here are the tricks, I’ve found
- Add the command ‘-Xverify:none’ to your GRAILS_OPTS environment variable before launching your application. VisualVM relies on class replacement to instrument your code for profiling, and this switch allows that to work correctly.
- Disable the auto reloading agent in Grails by passing -Dauto.recompile.disable=true on the command line when you start your Grails application. This mostly a safety measure. Since VisualVM is replacing classes, you don’t want Grails replacing over the top of them. Also, there is a significant performance hit when using the Reloading Agent once it recompiles any class since it it basically proxying every class reference.
- Launch VisualVM with a bunch of memory. Since Grails needs a good chunk of memory, the profiler is going to need a lot. You can increase memory to VisualVM by passing the commands -J-Xms<memory size> and -J-Xmx<memory size> on the command line when launching VisualVM (i.e. jvisualvm -J-Xms1G -J-Xmx1G).
- While your there, you should add your Grails application classes to VisualVM’s classpath so you can make better sense of the profiling results. You can prepend/append classes to VisualVM’s classpath by using the –cp:p and –cp:a arguments, respectively (i.e. jvisualvm -J-Xms1G -J-Xmx1G –cp:a <path to grails app>/target/classes