hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dan Burkert <danburk...@gmail.com>
Subject Re: Thinking about 0.98
Date Fri, 23 Aug 2013 23:03:39 GMT
Yep.  To support descending sort index scans Honeycomb has to maintain a separate descending
index, which slows down inserts/updates/deletes and doubles the storage overhead of indices.
We would gladly pay a performance hit for descending scans if it meant not having to store
indices twice. 

Descending scans also make sense WRT the new data types API. Currently you can pick whether
you want the type to sort ascending or descending, but if you need both the only option is
to duplicate.  Obviously some people may prefer to scan their data primarily in descending
sort, so unless descending scan speed becomes on-par with ascending it still makes sense to
have the choice.

-- Dan

On Aug 23, 2013, at 1:15 PM, Stack <stack@duboce.net> wrote:

> On Thu, Aug 22, 2013 at 8:41 PM, Dan Burkert <danburkert@gmail.com> wrote:
>> As an interested bystander (user) I would love to see the reverse scan
>> feature (HBASE-4811 (https://issues.apache.org/jira/browse/HBASE-4811))
>> make 0.98.  There has been a lot of talk about HBase indexes lately, and
>> the ability to reverse scan an index opens up a lot of possibilities.
> Do you need it for you mysql'ing Dan?
> St.Ack

View raw message