avro-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Scott Carey <sc...@richrelevance.com>
Subject Re: OOMs in java hadoop tests?
Date Fri, 30 Apr 2010 01:50:25 GMT
On Apr 29, 2010, at 4:54 PM, Doug Cutting wrote:

> Scott Carey wrote:
>> Yes, I see the same thing.
>> 
>> Adding maxmemory="128m"
>> to the junit ant task fixed it for me.
> 
> I wonder if what we're actually seeing is an exhaustion of file 
> descriptors due to some leak.  As I recall, this also shows up as an OOM.
> 
> Another way to fix this, rather than adding maxmemory, is to remove 
> forkmode="once", so that a new JVM is forked per test.  That also fixes 
> it for me.
> 

How much longer do the tests take with that setting?  Generating a new JVM for each unit test
will increase the run time.

Also, if destroying the JVM makes it work, then there is probably some sort of leak in certain
tests -- static variables or something.  We could force the JVM to create a heap dump on exit
and poke around with the heap dump to see.

I'm fairly sure that running out of file descriptors has a slightly different message.


> Doug


Mime
View raw message