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

Switch to Threaded View
HBase, mail # user - got HBASE-2516 on 0.20.6


Copy link to this message
-
Re: got HBASE-2516 on 0.20.6
Andrey Stepachev 2010-09-17, 21:25
Thanks for comment.

2010/9/18 Jean-Daniel Cryans <[EMAIL PROTECTED]>:
> That exception is "normal", it just means a region is either moving or
> being split, unless your client has to retry 10 times and then gets a
> RetriesExhaustedException. This has been discussed a few times on this
> mailing list.

I have custom RetryResultIterator, which reopen scanner from last known
postition.

>
> Also you say you got HBASE-2516, but if you really got it then you'd
> see IOException instead of NotServingRegionException.

I got it in different situations.

>
> Are you actually having issues importing your data? If so, can you be
> more specific about how it fails?

I have several families and 70% of data should be changed (change in
json object). I open scanner and go through, make new changed puts
(they are placed in same key/family/qualifier) and put them in the same
table/family.

I wrote a very small tool (looks like mr job, but simply java app)
which spawns serveral threads,
gets key ranges and each thread works with only one family at a time
and one key range.
Tool do something like this (executor pool has fixed number of threads):

families.each { |f|
    keyranges.each { |range|
        executorPool.startThread(f, range)
    }
}

If I run tool on two familes, and looks like region splits on first
family, when i process second family, i got
http://pastebin.com/M4xzaHAg.
If I start tool only with failed family, i got "region closed".
http://pastebin.com/9zFLi9Pn
I really can't figure out what is going on. But I think that here is
some "race" when region begins
to split.

>
> Regarding 0.89, since you are running HBase on only one machine, I
> don't expect that you'll run into any major issues. Here at
> StumbleUpon we have been running on 0.89.20100830 (the release
> candidates) since the first week of september in our production
> environment.

Thanks. Interesting. I'll try to move my test server and look how it works.
0.89 has too many features (like timestamp scan) that i need.

>
> J-D
>
> On Fri, Sep 17, 2010 at 1:36 PM, Andrey Stepachev <[EMAIL PROTECTED]> wrote:
>> I need to massive data rewrite in some family on standalone server. I
>> got org.apache.hadoop.hbase.NotServingRegionException
>> or java.io.IOException: Region xxx closed if I write and read at the same time.
>>
>> What i shall do in 0.20.6 version? One thing that i can try - write to
>> another family and then copy changed data
>> back to original family. But this is not really good solution.
>> Another option - got to 0.89, but i don't know where it will be
>> released, so it is very questionable, is it really usable
>> (small standalone server with up to 5 nodes in the future).
>>
>