On 12/28/2013 12:53 AM, Dylan Hutchison wrote:
/1. Should the transpose table be built as part of ingest code or as an accumulo combiner?/ I recommend ingest code for much greater simplicity, though it may be possible to build a combiner to automatically ingest to a second table. When inserting (row,col,val) triples, do another insert to the transpose with (col,row,val). Use summing combiners to create the degree tables.
Using a combiner is likely to be much more hassle than it's worth. When your Combiner gets invoked server-side, you have no notion of lifecycle management and the only way to write to another table is to instantiate a Connector and BatchWriter.
As such, it's very difficult, and possibly impossible with the current API, to write to a separate table inside of a Combiner without leaking resources inside of the TabletServer.
Definitely implement it in your ingest code :)
