[ 
https://issues.apache.org/jira/browse/GORA-65?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13173607#comment-13173607
 ] 

Keith Turner commented on GORA-65:
----------------------------------

Oh one other major difference is paritioning a query.  I will explain this by 
example. Suppose I have an gora-accumulo table where the key is a long.  
Suppose this table has the following splits.  There are two reasons these split 
points are not 8 bytes.  Users can insert split points.  Also, when Accumulo 
chooses a split point, it will choose the shortest possible string.  So split 
point do not always come from rows in the table.

{noformat}
0x20
0x60
{noformat}

When partitioning a query over the entire table, I have to take the 2 splits 
above and generate three partition queries that cover the entire table but do 
not overlap. Also the queries need to align exactly with accumulo tablets (like 
HBase regions) to maintain locality.   The fact that the split points are not 8 
byte also makes it tricky, they will not decode as a long.  So I generate three 
gora queries with the following start and end keys.  These queries cover all 
possible longs in the table.  The queries map to accumulo tablets which have 
exclusive start and inclusive end keys.

{noformat}
query 1  start key=null               end key=0x1fffffffffffffff
query 2  start key=0x2000000000000000 end key=0x5fffffffffffffff
query 3  start key=0x6000000000000000 end key=null  
{noformat}


When the key is not fixed width (like a long or int) and its a String, then its 
much easier to create the partition queries.


                
> Implement Accumulo datastore for Gora
> -------------------------------------
>
>                 Key: GORA-65
>                 URL: https://issues.apache.org/jira/browse/GORA-65
>             Project: Gora
>          Issue Type: New Feature
>            Reporter: Keith Turner
>         Attachments: GORA-65-1.patch
>
>
> Accumulo is a BigTable implementation similar to HBase.  Accumulo would be a 
> great fit for another Gora datastore.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: 
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Reply via email to