hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Yu <yuzhih...@gmail.com>
Subject Re: Efficiently scan table segments with coprocessor
Date Wed, 09 Apr 2014 03:07:30 GMT
bq.  a single user's data is stored across a set of rows

Have you looked at HBASE-9488 'Improve performance for small scan' ?


On Tue, Apr 8, 2014 at 8:30 PM, David Quigley <dquigley89@gmail.com> wrote:

> Hi,
> We have a short and tall hbase table structure where a single user's data
> is stored across a set of rows in hbase. New events are inserted into a
> user's data, and the sorting of the rows handles the structure of the users
> data. Most of our user base is inactive at any given time. When a user's
> data is updated, we want to run a computation on the user's data and update
> the user's stats. Because of our userbase pattern, running a full scan is
> very inefficient. We basically want to stream updates every time a user
> takes an action.
> I am thinking that the best way to do this is through a RegionObserver
> coprocessor, using either prePut or postPut. Only problem is that we would
> need to instantiate a new scan each time pre or postPut is called, which
> might we inefficient. Since hbase already has a pointer to the user's data
> via the Put, is there any way to leverage this to scan that local data more
> efficiently than using a new InternalScanner?
> Thanks,
> Dave

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message