We have configuration that includes Apache and Tomcat servers connected by mod_jk. There are several tomcat servers and request are distributed by the Round-Robin algorithm. Just recently the number of AJP threads suddenly started growing and the tomcat servers crashed.
We do not see anything unusual in the logs, and the load was below average. The Garbage collection started about 5 minutes before the sudden increase as shown by New Relic software.
What is the best approach in dealing with this crash? I am thinking about simulating the load with JMeter, what can be other ideas?