Memory Leak Detection For Java Collections Of Cloud Foundry Java Application

Memory leak occurs when Java programs keep some unneeded objects references and these objects would never be accessed again. Such memory leaks are difficult to identity the root cause because of the complexity of the references between objects, and because we cannot ensure some object was useless.In this thesis, we implement a memory leak detecting system for Cloud Foundry Java applications based on memory leak detection using container profiling. We calculate a confidence value of memory leaking for each container object from the memory usage and the elements usage. First we identity a leaking region through program’s memory consumption data after every garbage collection. Second we calculate the memory contribution through the reference map and objects size. Third we collect the elements usage in every container through modifying the bytecode of Java collections and calculate the elements usage contribution. At last we rank the confidence values of all monitored container instances.We implement a memory leak detecting system using Java Instrument, JVM Interface Tools, ASM and so on. The system has three parts:Java Agent, Native Agent and confidence calculation module. Java Agent collects the elements usage data and Native Agent collects memory data. The memory leak detecting system is part of application performance management in Cloud Foundry. We also detect a memory leaking example program with the system and test the system runtime overhead.