> On 20 Jul 2015, at 10:27, Tilman Hausherr <[email protected]> wrote:
> 
> Am 20.07.2015 um 18:12 schrieb Allison, Timothy B.:
>> All,
>>   While integrating 2.0.0 trunk into Tika and running against govdocs1, I'm 
>> finding two issues that are difficult to reproduce.
>> 
>> Background:
>> Tika-batch has a parent process that kicks off a Tika processor in a child 
>> process, if that dies unexpectedly, the parent kicks it off again.  I'm 
>> running with 10 consumer/parser threads and -Xmx5g on an (8 cpu/8GB vm); 
>> RHEL 7, Linux cloud-server-02 3.10.0-123.20.1.el7.x86_64 #1 SMP Wed Jan 21 
>> 09:45:55 EST 2015 x86_64 x86_64 x86_64 GNU/Linux)
>> 
>> Two problems:
>> 
>> 1)      The child process exits with value 1. I'm catching Throwable around 
>> the primary execution call in the child process and logging it; nothing 
>> shows up in the log files from that part of the code. From the parser log 
>> files (at trace), I can tell which 10 files were being processed at the 
>> time, but I'm not seeing any other information about what caused the exit.  
>> When I run against just those 10 files, all is ok.
>> 
>> 2)      The OS is killing the child far more often than it does with 1.8.9 
>> (exit code 137).
>> 
>> For the second problem, I'll wait until the optimizations to the caching are 
>> completed before I start worrying about that.  However, do you have any 
>> recommendations on how to figure out what's going on with 1)?
> 
> I'm also having some problem with that system... with my test software, I 
> have observed that java uses more and more space, despite it being told not 
> to use more than a certain amount with -Xmx. After some time, the "process 
> killer" kills the application.

Xmx doesn’t limit native memory, so if there’s a leak associated with AWT, 
ImageIO C libraries, or some other JNI library, the process can grow without 
limit. Such a leak could be due to a bug, or us not calling close() somewhere.

— John

> Seems something changed in java memory management:
> http://karunsubramanian.com/websphere/one-important-change-in-memory-management-in-java-8/
>  
> <http://karunsubramanian.com/websphere/one-important-change-in-memory-management-in-java-8/>
> 
> I did some investigation on this a few months ago, but gave up out of 
> frustration.
> 
> Tilman
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [email protected] 
> <mailto:[email protected]>
> For additional commands, e-mail: [email protected] 
> <mailto:[email protected]>

Reply via email to