Thanks steven, I went ahead and filed https://issues.apache.org/jira/browse/DRILL-3270 for the time being.
On Tue, Jun 9, 2015 at 8:31 PM, Steven Phillips <[email protected]> wrote: > It's currently not supported. It would require an enhancement to the > storage plugin. > > Feel free to file an enhancement jira, or f you want to experiment with it > yourself, you can take a look at the HBaseGroupScan code, which is where > the assignment logic is located. Currently, we don't split the work within > a region, but it shouldn't be too difficult to modify the code to do this. > The difficult part is that we don't necessarily have a good idea what the > distribution of keys is within a region, so we can't be sure that we split > split the region evenly. The easy way to handle it would be to simply > assume the keys are evenly distributed between the start and end key, and > split half way between (or 1/3 or whatever, depending on how many fragments > per region you want). > > On Tue, Jun 9, 2015 at 11:20 AM, Andy Pernsteiner < > [email protected] > > wrote: > > > I've been doing some experimentation with Drill (1.0) against MaprDB. I > > believe I'd see similar behavior vs HBASE, although I haven't tried yet. > > > > I have a 11GB table, which is split into 8 regions (not perfectly > > balanced, some have 2x the # of records as others). > > > > When I run a drill query which results in a full table scan (eg: select > > *..), I notice in the profile that there are only 8 minor fragments > > assigned. This seems to coincide with other tests which I've ran that > also > > seem to assign one minor fragment per region. > > > > Is this expected? Are there any ways of increasing the parallelism on > > table scans? (eg: every XX records, or assign XX fragments per region) > > > > > > > > -- > > Andy Pernsteiner > > Manager, Field Enablement > > ph: 206.228.0737 > > > > www.mapr.com > > > > Now Available - Free Hadoop On-Demand Training > > < > > > http://www.mapr.com/training?utm_source=Email&utm_medium=Signature&utm_campaign=Free%20available > > > > > > > > > -- > Steven Phillips > Software Engineer > > mapr.com > -- Andy Pernsteiner Manager, Field Enablement ph: 206.228.0737 www.mapr.com Now Available - Free Hadoop On-Demand Training <http://www.mapr.com/training?utm_source=Email&utm_medium=Signature&utm_campaign=Free%20available>
