You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@kudu.apache.org by MattK <bs...@gmail.com> on 2016/11/18 02:16:11 UTC

Drill on Kudu

If I understand Kudu correctly, it manages its own cluster, and does not rely on ZK for coordiantion (?).

When using the Kudu storage configuration in Drill 1.7/8, is it necessary to run multiple Drillbits, and is there a concept of data locality with Drill on Kudu?

Or is a single standalone Drillbit used when only A Kudu cluster is the target, and the Kudu handles the work pushed down from the single Drillbit?


Re: Drill on Kudu

Posted by Todd Lipcon <to...@cloudera.com>.
On Thu, Nov 17, 2016 at 6:16 PM, MattK <bs...@gmail.com> wrote:

> If I understand Kudu correctly, it manages its own cluster, and does not
> rely on ZK for coordiantion (?).
>
> Correct.


> When using the Kudu storage configuration in Drill 1.7/8, is it necessary
> to run multiple Drillbits, and is there a concept of data locality with
> Drill on Kudu?
>

I believe Drill does expose locality. However, on the Kudu side I don't
think anyone has tested Drill integration in nearly a year, so it's
possible that things aren't working as expected anymore.


>
> Or is a single standalone Drillbit used when only A Kudu cluster is the
> target, and the Kudu handles the work pushed down from the single Drillbit?
>
>
Should be possible, but I don't think the Drill/Kudu integration
implemented much in the way of filter pushdown, so may not actually work.

In general I think the Drill mailing list is a better spot for this
question, since the Drill/Kudu integration is managed in their source
repository and part of their releases.

-Todd
-- 
Todd Lipcon
Software Engineer, Cloudera