In hadoop 2.4.0 NM always have 100+ AppLogAggregatorImpl threads running, Normally,only 20+ containers will running all time,so if there is any issue about the AppLogAggregator leak. I observe these threads is all waiting,because the logic of code is: While(If not appFinshed) Wait(1s);
Like this: [cid:image001.png@01D2978E.628DB510]