Reactor hanging on to a 1.6GB LinkedTransferQueue

[Andras Szerdahelyi]

hello list,

i have observed what seems to be a memory leak ( at least in it's effects ) in my application, an XMemcached Reactor object seems to be holding a 

1.6GB LinkedTransferQueue. I started browsing the queue from the heap dump i aquired and could find only "ENABLE_WRITE" events in it. This also only occurs on the same two hosts out of my 30-40 node deployment of this application. Is this some kind of memcached connectivity issue? How do i get the root cause of this?

Thanks!

Andras

ps: my xmemcached version is 1.4.1

The thread com.google.code.yanf4j.nio.impl.Reactor @ 0x6003d15c0 Xmemcached-Reactor-1 keeps local variables with total size 1?693?437?896 (85,76%) bytes.

Accumulated Objects in Dominator Tree 

Class Name Shallow Heap Retained Heap Percentage

com.google.code.yanf4j.nio.impl.Reactor @ 0x6003d15c0 Xmemcached-Reactor-1

168 1?693?437?896 85,76%

+

com.google.code.yanf4j.util.LinkedTransferQueue @ 0x6003fe0c8

24 1?693?433?120 85,76%

+

java.lang.ThreadLocal$ThreadLocalMap @ 0x6003d16d8

24 4?416 0,00%

+

char[20] @ 0x6003d1668 Xmemcached-Reactor-1

56 56 0,00%

+

java.util.concurrent.locks.ReentrantLock @ 0x6003fe1d0

16 48 0,00%

+

java.util.HashMap$KeyIterator @ 0x6095e1748

40 40 0,00%

+

java.lang.Object @ 0x6003fe098

16 16 0,00%

+

java.util.concurrent.atomic.AtomicInteger @ 0x6003fe0a8

16 16 0,00%

+

java.util.concurrent.atomic.AtomicBoolean @ 0x6003fe0b8

16 16 0,00%