hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Yu <yuzhih...@gmail.com>
Subject Re: One Region Server fails - all M/R jobs crash.
Date Fri, 22 Nov 2013 18:05:18 GMT
Thanks Dhaval for the analysis.

bq. The HBase version is 0.94.6

David:
Please upgrade to 0.94.13, if possible. There have been several JIRAs
backporting patches from trunk where jdk 1.7 is supported.

Please also check your DataNode log to see whether there was problem there
(likely there was).

Cheers


On Sat, Nov 23, 2013 at 2:00 AM, Dhaval Shah <prince_mithibai@yahoo.co.in>wrote:

> You logs suggest that you are overloading resources
> (servers/network/memory). How much data are you scanning with your MR job,
> how much are you writing back to HBase? What values are you setting for
> setBatch, setCaching, setCacheBlocks? How much memory do you have on your
> region servers? 1 server crashing should not cause a job to fail because it
> will move on to the next one (given the right parmas for retries and retry
> interval are set). Your region server logs suggest that its way more
> complicated than that.
>
> 2013-11-17 09:58:37,513 WARN
> org.apache.hadoop.hbase.regionserver.HRegionServer: Received close for
> region we are already opening or closing; e54b8e16ffbe2187b9017fef596c62aa
>
> looks like some state inconsistency issue
>
> I also see that you are using Java 7. Though some people have had success
> using it, I am not sure if Java 7 is currently the recommended version
> (most people use Java 6!)
>
> 2013-11-18 18:01:47,959 INFO org.apache.zookeeper.ClientCnxn: Unable to
> read additional data from server sessionid 0x342654dfdd30017, likely server
> has closed socket, closing socket connection and attempting reconnect
>
> This line is suggesting a problem with your zookeeper. If zookeeper screws
> up, HBase will and hence your MR job over HBase will.
>
> 2013-11-21 06:54:01,105 WARN org.apache.hadoop.hdfs.DFSClient: Failed to
> connect to /XXX.XXX.XXX.XXX:50010 for block, add to deadNodes and continue.
> java.net.ConnectException: Connection refused
>
> And this suggests datanode crashed. So many processes (don't know if they
> belong to the same server or not) crashing at the same time seems to be a
> load issue or a network issue to me.
>
>
>
> Regards,
> Dhaval
>
>
> ________________________________
>  From: David Koch <ogdude@googlemail.com>
> To: user@hbase.apache.org
> Sent: Friday, 22 November 2013 12:35 PM
> Subject: Re: One Region Server fails - all M/R jobs crash.
>
>
> Here you go:
>
> Task log: http://pastebin.com/VePTLHEk
> Region Server log: http://pastebin.com/iu8y0VYL
>
>
>
> On Fri, Nov 22, 2013 at 6:27 PM, Ted Yu <yuzhihong@gmail.com> wrote:
>
> > Attachment didn't go through.
> >
> > Can you pastebin their contents ?
> >
> > Thanks
> >
> > On Nov 23, 2013, at 12:55 AM, David Koch <ogdude@googlemail.com> wrote:
> >
> > > Sorry for the previous message, I attach the equired log files.
> > >
> > > Regards,
> > >
> > > David
> > >
> > >
> > > On Fri, Nov 22, 2013 at 5:53 PM, David Koch <ogdude@googlemail.com>
> > wrote:
> > >>
> > >>
> > >>
> > >> On Fri, Nov 22, 2013 at 4:17 PM, Ted Yu <yuzhihong@gmail.com> wrote:
> > >>> Can you pastebin snippet of:
> > >>> 1. task logs which show failure
> > >>> 2. region server log shortly before the crash
> > >>>
> > >>> Thanks
> > >>>
> > >>>
> > >>> On Fri, Nov 22, 2013 at 7:14 AM, David Koch <ogdude@googlemail.com>
> > wrote:
> > >>>
> > >>> > Hello,
> > >>> >
> > >>> > We experience reliability problems when running M/R jobs over
HBase
> > tables.
> > >>> > Specifically, it suffices for one Region Server to crash in order
> to
> > fail
> > >>> > all M/R jobs.
> > >>> >
> > >>> > My guess is that this is not normal with a replication factor
of 3.
> > >>> >
> > >>> > The HBase version is 0.94.6 installed as part of of Cloudera 4.4.
> > HBase
> > >>> > settings are pre-sets. Cluster size is 30 machines.
> > >>> >
> > >>> > What steps can I follow to improve the situation?
> > >>> >
> > >>> > Thank you,
> > >>> >
> > >>> > /David
> > >>> >
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message