Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
HBase, mail # user - HBase Region Server crash if column size become to big


Copy link to this message
-
Re: HBase Region Server crash if column size become to big
John 2013-09-11, 17:03
wow, you're right ... I changed the setBatch Value to 50000 and now it
seems to work!

Big thanks to all! :)

btw. @ Bryan: there was no other error/warning message after the warning.
And there was no other error in the syslog or so. You can check it out by
yourself. Load one big rowkey with ~600000 columns in the HBase and then
try to get/scan the key. I've tested it with my real cluster (6 machines)
and with my virtual machine (1 virtual machine) and got the error on both
cluster. Here is my full programm http://pastebin.com/KzRiRe25 ... just
execute it with args "600000"

thanks again :)
2013/9/11 Bryan Beaudreault <[EMAIL PROTECTED]>

> @John, I think you're going to want to limit your batch, as opposed to
> raise it.  How much memory does the RegionServer get?  Are you sure the row
> is only 70MB?  You could check HDFS directly by ls'ing the region
> directory, or use the HFile tool.
>
> The "errors" you have been posting are simply WARNs.  There are arbitrary
> limits defined for responseTooLarge, responseTooSlow, operationTooLarge,
> just to give you the ability to debug bad client calls.  They don't mean
> the RS will necessarily have an issue returning the result.
>
> Are there literally no other logs at all after that WARN?  Have you checked
> the system logs for oom killer invocations?
>
>
>
>
> On Wed, Sep 11, 2013 at 11:58 AM, John <[EMAIL PROTECTED]> wrote:
>
> > @Kevin, I'm using Apache Pig to execute my programm. I wrote my own HBase
> > Load UDF and added now  scan.setBatch(10000000), but it is still
> crashing.
> >
> > @Dhaval: I'm using Cloudera 4.4.0. Its nearly the default Installation
> from
> > the cloudera manager. I have no idea why there is now log file
> >
> > Does anyone test the Java program and execute a get in the hbase shell?
> >
> >
> >
> >
> > 2013/9/11 Dhaval Shah <[EMAIL PROTECTED]>
> >
> > > @Mike rows can't span multiple regions but it does not cause crashes.
> It
> > > simply won't allow the region to split and continue to function like a
> > huge
> > > region. We had a similar situation long back (when we were on 256mb
> > region
> > > sizes) and it worked (just didn't split the region).
> > >
> > > Sent from Yahoo! Mail on Android
> > >
> > >
> >
>