Trustin Lee wrote:
> 
> On Nov 21, 2007 7:27 PM, Matt Phillips <[EMAIL PROTECTED]> wrote:
>> Trustin Lee wrote:
>> >
>> > You might be writing too fast so that the write request queue piles
>> > up.  Could you run jmap to dump your server's memory map?
>> >
>>
>> Do you mean you'd like a hprof memory dump?
> 
> Yes.
> 

It looks like I spoke too soon: in production both heap and direct buffers
leak, and ironically the direct buffers seem far worse, in direct
contradiction to what the tight loop test indicated. This is certainly a
weird situation. I just wish I had access to a profiler in order to track
the origin...

In the direct case the histogram looks like:

  8218080 513630  java.util.concurrent.ConcurrentLinkedQueue$Node
  2622280 4955    int[]
  1247320 12293   * ConstMethodKlass
  1107480 15942   char[]
  986880  12293   * MethodKlass
  947200  29600   java.lang.ThreadLocal$ThreadLocalMap$Entry
  658160  17045   * SymbolKlass
  625560  26065  
java.util.concurrent.locks.ReentrantReadWriteLock$Sync$HoldCounter
  ...

I've dropped two hprof files at the following locations. The "direct" one is
using direct heap buffers, the "heap" one is using heap buffers, both taken
after at least 12 hours of light usage. These will be auto deleted in 24
hours, so please grab a copy asap.

  ftp://publicftp.dsto.defence.gov.au/avis-direct.hprof.bz2
  ftp://publicftp.dsto.defence.gov.au/avis-heap.hprof.bz2

Hope this is enough. I might see if I can use the profiler in NetBeans to
give me a better idea of what's going on.

Matthew.
-- 
View this message in context: 
http://www.nabble.com/Memory-leak-in-MINA-1.1.4-tf4849077s16868.html#a13888344
Sent from the Apache MINA Support Forum mailing list archive at Nabble.com.

Reply via email to