You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@hbase.apache.org by Sever Fundatureanu <fu...@gmail.com> on 2012/08/19 13:51:24 UTC
OOME cause
Hello,
I have a RegionServer which OOMEd. The memstore is set to 0.4 and the
blockCache to 0.4 of total memory. Although the memstore was empty
(confirmed with Ganglia), the server OOMEd.
I noticed the block cache is doing a lot of evictions, which caused a
full GC pause. However, I don't see why this would cause an OOME, if
the memstore is empty. Can someone please explain this?
You can find the last lines of .out and .log files here:
http://pastebin.com/NjV7Wrjj
http://pastebin.com/Px4McVDt
Thank you in advance,
--
Sever Fundatureanu
Vrije Universiteit Amsterdam
Re: OOME cause
Posted by Sever Fundatureanu <fu...@gmail.com>.
Hi Kevin,
The heap was close to 100% as expected. The version is 0.92.0.
I can also add I am using the options: HBASE_OPTS="-XX:NewSize=64m
-XX:MaxNewSize=64m -XX:+UseConcMarkSweepGC -XX:+UseParNewGC
-XX:CMSInitiatingOccupancyFraction=70"
If I allocate .8 for memstore+blockCache, will HBase use only .2 even
if the .8 is not fully used ? Or will it use more, if for example the
memstore is empty?
Thanks,
Sever
On Sun, Aug 19, 2012 at 7:08 PM, Kevin O'dell <ke...@cloudera.com> wrote:
> Sever,
>
> What is the heap set at? I would not recommend exceeding .7 with
> memstore and block cache combined as this does not leave much room for
> hbase itself. What version is this on?
> On Aug 19, 2012 7:51 AM, "Sever Fundatureanu" <fu...@gmail.com>
> wrote:
>
>> Hello,
>>
>> I have a RegionServer which OOMEd. The memstore is set to 0.4 and the
>> blockCache to 0.4 of total memory. Although the memstore was empty
>> (confirmed with Ganglia), the server OOMEd.
>> I noticed the block cache is doing a lot of evictions, which caused a
>> full GC pause. However, I don't see why this would cause an OOME, if
>> the memstore is empty. Can someone please explain this?
>>
>> You can find the last lines of .out and .log files here:
>> http://pastebin.com/NjV7Wrjj
>> http://pastebin.com/Px4McVDt
>>
>> Thank you in advance,
>> --
>> Sever Fundatureanu
>> Vrije Universiteit Amsterdam
>>
--
Sever Fundatureanu
Vrije Universiteit Amsterdam
E-mail: fundatureanu.sever@gmail.com
Re: OOME cause
Posted by Kevin O'dell <ke...@cloudera.com>.
Sever,
What is the heap set at? I would not recommend exceeding .7 with
memstore and block cache combined as this does not leave much room for
hbase itself. What version is this on?
On Aug 19, 2012 7:51 AM, "Sever Fundatureanu" <fu...@gmail.com>
wrote:
> Hello,
>
> I have a RegionServer which OOMEd. The memstore is set to 0.4 and the
> blockCache to 0.4 of total memory. Although the memstore was empty
> (confirmed with Ganglia), the server OOMEd.
> I noticed the block cache is doing a lot of evictions, which caused a
> full GC pause. However, I don't see why this would cause an OOME, if
> the memstore is empty. Can someone please explain this?
>
> You can find the last lines of .out and .log files here:
> http://pastebin.com/NjV7Wrjj
> http://pastebin.com/Px4McVDt
>
> Thank you in advance,
> --
> Sever Fundatureanu
> Vrije Universiteit Amsterdam
>