hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Lars George <lars.geo...@gmail.com>
Subject Re: HBase 0.89.0-20100924 region assignment woes
Date Thu, 25 Nov 2010 18:50:14 GMT
Apparently it got stripped. My GMail webclient shows it though but
that seems misleading. Here is the log:

http://pastebin.com/pLDp9p0D

Lars

On Thu, Nov 25, 2010 at 3:59 PM, Ted Yu <yuzhihong@gmail.com> wrote:
> Lars:
> Did you forget to attach the log (pastebin) ?
>
> I assume this scenario could happen in 0.20.6 as well.
>
> On Thu, Nov 25, 2010 at 2:49 AM, Lars George <lars.george@gmail.com> wrote:
>
>> Hi,
>>
>> Talking to Lars we found a weird series of events that led to dying
>> task attempts during the bulk job they are running. See the attached
>> log. Region in question is:
>>
>>
>> raw_occurrence_record,28256992928,1290626317732.86e3cd5c5d0f22430debb36f1668d3fc
>>
>> After a split the region is assigned to a new server (c1n6) but then
>> immediately ripped out underneath it because the server is deemed to
>> be too loaded. Then it releases that region (after a few minutes) and
>> a few others and then that new daughter region, which did not have
>> time yet to serve anything really, is reassigned to.... you guessed it
>> probably, the same server as that is now deemed (?) to be the least
>> loaded one. So that new daughter is not available for minutes causing
>> the task attempt to die.
>>
>> I see that in trunk this is all replaced. No RegionManager but an
>> AssignmentManager who does random node selection. Is that all obsolete
>> now and this should be improved on trunk?
>>
>> Lars
>>
>

Mime
View raw message