I am running Logstash 1.5.4 on a number of Windows servers as an agent to collect eventlog and other data. I've done a decent job of tweaking the configuration such that it minimizes the use of expensive filtering and other code, but the memory still blows right past the heap size allocated to the JVM in setup.bat.
This is a memory leak -- under normal baseline nothing-happening-on-Windows conditions, the Java process climbs very, very slowly and inexorably toward a wall after which it bombs out (and is restarted by NSSM). Under stress test, the memory climbs much faster, and when the stress test stops, it does not clear the memory.
I have seen rumblings that this is due to a memory leak in JRuby, and that there is a fix soon to be available shipped with Logstash 2.1.1. When is that coming out, and in the meantime does anyone have any stopgaps that directly address the memory issue in question?