Hi Navdeep, I believe you will need to: * implement RegionScanner that would apply aggregation on Cell level * extend from BaseRegionObserver to force using your RegionScanner in preGet and preScan
I don't have a simple example in front of me, but maybe the following will give you some pointers. We use versions of a Cell to store delta-values when performing append-style increments (you put delta in next version of a cell instead of incrementing existing). Then, during scanning those deltas got summed up into a single value. I assume you want to do something along those lines, so you may learn some from that code. Here's the RegionScanner implementation [1]. Note that in next(List<Cell> cells, int limit) you'll need to check for crossing the boundary of the cell (i.e. cells given to you may have e.g. 3 versions of cell of column1 and 2 versions of a cell of column2). Here's the BaseRegionObserver implementation [2]. On a side note, be sure to not overuse the versions of a Cell. Many times using columns is a better schema design. Cheers, Alex Baranau -- http://cdap.io - open source framework to build and run data applications on Hadoop & HBase [1] https://github.com/caskdata/cdap/blob/develop/cdap-hbase-compat-0.98/src/main/java/co/cask/cdap/data2/increment/hbase98/IncrementSummingScanner.java [2] https://github.com/caskdata/cdap/blob/develop/cdap-hbase-compat-0.98/src/main/java/co/cask/cdap/data2/increment/hbase98/IncrementHandler.java [3] http://hbase.apache.org/book.html#schema.versions On Thu, May 14, 2015 at 5:37 AM, Navdeep Agrawal < navdeep_agra...@symantec.com> wrote: > Hi, > I am trying to use co processor to do some aggregations(eg topn) over all > versions of a cell and return it . I found most of the aggregation > implementation with coprocessors are done on column . how we can achieve > for every cell in that column ,any ideas ,links ??? > > Use case - if I want to dom some aggregation over all versions of cell and > return single value for that cell given row key and column . > > > Thanks, > Navdeep >
