Friday, December 27, 2013

I have a production server running with the following flag: -XX:+HeapDumpOnOutOfMemoryError
Last night it generated a java-38942.hprof file when our server encountered a heap error. It turns out that the developers of the system knew of the flag but no way to get any useful information from it.
Any ideas?
share|improve this question
add comment

5 Answers

up vote66down voteaccepted
If you want a fairly advanced tool to do some serious poking around, look at the Memory Analyzer projectat Eclipse, contributed to them by SAP.
Some of what you can do is mind-blowingly good for finding memory leaks etc -- including running a form of limited SQL (OQL) against the in-memory objects, i.e.
SELECT toString(firstName) FROM com.yourcompany.somepackage.User
Totally brilliant.
share|improve this answer
8 
I'd just like to add a +100 for Eclipse Memory Analyzer. I'm currently attempting to sift thru a 400mb+ heap dump file, and it took jhat more than 70 minutes to read the file, before it caused a complete JVM crash. EMA is able to open it up in < 5 minutes. –  matt b Jun 18 '09 at 18:37
 
Too bad MAT often fails to read heaps... :-( –  Vladimir Dyuzhev Apr 1 '10 at 14:55
2 
I keep getting parsing errors when opening HPROF files using the Eclipse Memory Analyzer (that were in fact also dumped by Eclipse!). Unfortunate.. sigh. –  lost_bits1110 Dec 14 '11 at 22:16 
add comment

I suspect we have a major memory leak in our ActiveMQ connection bridge - we're seeing typical memory leak patterns (app loads fine, slows down if it runs for prolonged periods of time or is restarted over and over again over short periods of time). I looked up modern best practices for finding Java memory leaks and a lot of developers seem to be abandoning traditional tools like jhat/jmap in lieu of the new(er)jvisualvm.
Upon launching this tool (and spending a few hours reading over its tutorial) I am able to take profiler snapshots for both CPU and memory.
I'm just sort of stuck at this point - how do I analyze these snapshots to identify the leak? There's a plethora of documentation out there as to how to use jvisualvm to produce snapshots, but very little documentation as to how to actually make sense of them.
Thanks in advance.
share|improve this question
 
Have you tried running FindBugs over the source? I find that pretty effective at identifying code that could cause memory leaks. –  Jivings Feb 6 '12 at 2:01
 
Thanks for the suggestion - but two things: (1) I've been told over and over again that FindBugs is good for NullPointerExceptions but not memory leaks, and, more importantly, (2) I really want to get familiar with jvisualvm, so unless it is clearly not the right tool for this job, I'm interested to see how it is used as a diagnostic tool. Thanks again for the good suggestion though! –  IAmYourFaja Feb 6 '12 at 2:17
1 
One little thing: Tools/Plugins/Available Plugins -- install Visual GC -- the plugin shows better graph of your heap memory and GC status. –  coolcfan Feb 6 '12 at 3:28
add comment
Analysing memory leak using visualvm is not that easy. It has a tool/plugin called 'sampler'. This can be used to sample memory or cpu. You can take snapshot at regular interval and look for possible leaks. Here is some details on how to use it- Obtained from quick search
More effective way will be to get a heap dump (say when the application has slowed down or when OOM happens). VisualVM help you to take heapdump (using Heap Dump on Monitor tab)
This file can be analysed by MAT- Some details are here at How do I analyze a .hprof file? andhttp://memoryanalyzer.blogspot.in/
share|improve this answer
add comment

No comments: