JVM BUG(s) - Hadoop's threads hanging

[Dmitry Sivachenko]

On 21 июня 2014 г., at 15:41, Dmitry Sivachenko <trtrmitya at gmail.com> wrote:

> 
> On 21 июня 2014 г., at 14:17, Ronald Klop <ronald-lists at klop.ws> wrote:
>> 
>> 
>> Mmm, that makes it more though to solve. It would still be nice to know if running Hadoop on another OS has the same problem. That gives a pointer where to look.
>> And it would be nice to reduce your setup to a reproducible testcase, so people can replay it. Otherwise it is just a lot of guessing from my side what is happening internally in the application. I don't know how easy it is to create that.
>> Any news on the mailing-lists of Hadoop?
>> 
> 
> No news from Hadoop mailing-list.
> I will have test results with Linux soon (next week).


Okay, I tried Linux: no surprise it just works.
I tested hadoop-1.2.1 with Oracle JDK-1.7, OpenJDK-1.6, OpenJDK-1.7.


> 
> If anyone want to dig into it, I can send my config files for hadoop (almost no special tuning) and a command to run hadoop's bundled TeraSort benchmark.
> I have a cluster with 40 machines (1 for namenode, 1 for secondarynamenode and the rest are task nodes), but I am sure you don't need as much h/w to reproduce this.
>