hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Stack <st...@duboce.net>
Subject Re: Recommendation for migrating region server implementations
Date Thu, 24 Mar 2011 20:33:43 GMT
I'm not sure.  If it came up -- 'seems to work' -- then it looks like
we just ignore the extra stuff (though, that seems a little odd... I'd
expect the deserializing of these 'exotic's to throw an exception).
Test more I'd say.  The shell you are using below is for sure from an
untarnished hbase -- there is no indexedhbase in the CLASSPATH?

St.Ack

On Thu, Mar 24, 2011 at 1:30 PM, George P. Stathis <gstathis@traackr.com> wrote:
> Ah, it seems to work, yes. I was thinking all along that it didn't because I
> had setup a simple unit test that kept throwing this:
> java.io.IOException: Unknown protocol to name node:
> org.apache.hadoop.hbase.ipc.IndexedRegionInterface
> at
> org.apache.hadoop.hbase.regionserver.HRegionServer.getProtocolVersion(HRegionServer.java:2400)
> I had upgraded locally on top of my old 0.89 hbase.root and it seemed to
> work OK at first. Startup was clean and IRB shell was working. That unit
> test kept throwing me off though. Then, I remembered that we have a custom
> hbase-site.xml for our maven unit tests that was still referencing the
> old IndexedRegionInterface. Removed that and problem solved.
> So, does this mean that we can use the existing table definitions as-is?
> They still contain old index definitions. E.g.
> hbase(main):022:0> describe 'monitors'
> DESCRIPTION
>                                                        ENABLED
>
>  {NAME => 'monitors', INDEXES =>
> 'fooattributes:foo=org.apache.hadoop.hbase.client.tableindexed.IndexKeyGe
> true
>  neratorEorg.apache.hadoop.hbase.client.tableindexed.RowBasedIndexKeyGeneratorattributes:fooorg.apache.hadoop.io.Writable
>
>  0org.apache.hadoop.io.ObjectWritable$NullInstance'org.apache.hadoop.io.WritableComparable',
> FAMILIES => [{NAME => ...
> The INDEXES dictionary is still there. Could it create issues?
> -GS
> On Thu, Mar 24, 2011 at 3:54 PM, Stack <stack@duboce.net> wrote:
>>
>> Do you have disk space to spare?  I'd think that all that is different
>> about indexed hbase is the WAL format.  If you had an hbase.rootdir
>> that was the product of a clean shutdown with no WALs to process, I'd
>> think you could just 0.90.x on top of it.  If you had the disk space
>> you could give it a go?
>> St.Ack
>>
>> On Thu, Mar 24, 2011 at 12:15 PM, George P. Stathis
>> <gstathis@traackr.com> wrote:
>> > Hey folks,
>> >
>> > What would be the best approach for migrating away from a given region
>> > server implementation back to the default out-of-the box one? My goal
>> > here
>> > is to upgrade our cluster to 0.90 and migrate away from
>> > IndexedRegionServer
>> > back to the default HRegionServer.
>> >
>> > The only options that I know of at the moment are:
>> >
>> >   - Export/Import - pros: straightforward, cons: tedious, takes a long
>> > time
>> >   - org.apache.hadoop.hbase.mapreduce.CopyTable - not sure about the
>> > pros
>> >   and cons here because I've never used it. It seems to require that
>> > tables
>> >   are moved to a different setup. Can it be used in place?
>> >
>> > If folks know of a different way to do this, please let me know.
>> >
>> > Thank you in advance for your time.
>> >
>> > -GS
>> >
>
>

Mime
View raw message