Plumbing memory leaks

Transcript

1. Memory leaks Plumbr
2. Who am i? Nikita Salnikov-Tarnovski, @iNikem! Plumbr, Tartu start-up! 12 years of JVM development
3. Agenda Assumptions! Reality! Tools! Solution
4. Java and GC We were promised hassle free memory-management! Why we crash with OOM?! Is GC a hoax?
5. Promises No size calculation and explicit malloc Just call new No need to free Just forget and let GC handle it! Confined heap
6. OOM: Java Heap Space! PermGen space! GC overhead limit exceeded! requested XX bytes for Chunk::new. Out of swap space?! unable to create new native thread! allocLargeObjectOrArray! nativeGetNewTLA
7. Why OOM? Your code is too greedy!! Tries to use too much memory at once! Inefficient algorithm/data structure
8. Quiz How much memory is needed to put 10M integers into a java.util.HashSet?! 40M! 100M! 300M! 600M 547M
9. GC A subsystem of JVM, which periodically checks which objects in memory are still being used and which are not. Unused objects can be discarded and memory reclaimed and reused again.! Not a mind-reading magician
10. Memory leak A situation where some objects are not used by application any more, but GC fails to recognize them as unused! The notion of “object is not used by application any more” is totally, absolutely, 100% application specific!! http://plumbr.eu/blog/what-is-a-memory-leak
11. Examples Unbounded “cache”! String.substring! Dynamic class generation! Streams left opened! Immortal threads
12. Tools Profilers! Heap dump analyzers! APM! command line tools of JDK
13. Results crashes, sometimes together with the application;! tons of flickering dashboards, graphics, buttons, bells and whistles, but no answer;! console full of cryptic text and java class names;! … or application adapted for snails and turtles, with response times increased 5-350 times.! http://plumbr.eu/blog/solving-outofmemoryerror-no-tools-will-help-you
14. Memory Analyzer (MAT) From Eclipse! It really works! If you have experience! http://www.eclipse.org/mat/
15. Plumbr
16. About us http://plumbr.eu, @JavaPlumbr! A spin-off of the biggest software company in Baltics! Totally different approach
17. Demo
18. How it works Java agent with native part! Runs alongside your application in the same JVM! Monitors objects lifecycle! Makes decision based on per-class data
19. Glooms Doesn’t work in synthetic test applications! Find leaks, not explosions! Doesn’t help with native memory problems
20. Shines Gives you the answer, not just data! Before customers start to suffer! Small overhead compared to profilers! No upfront investment