You are viewing a plain text version of this content. The canonical link for it is here.

Posted to user@drill.apache.org by Marcus Rehm <ma...@gmail.com> on 2016/07/15 20:48:26 UTC

Pushdown Capabilities with RDBMS

Hi all,

I started to teste Drill and I'm very excited about the possibilities.

By now I'm trying to map ours databases running on Oracle 11g. After try
some queries I realized that the amount of time Drill takes to complete is
bigger than a general sql client takes. Looking the execution plan I saw
(or understood) that Drill is doing the join of tables and is not pushing
it down to the database.

Is there any configuration required to it? How can I tell Drill to send to
Oracle the task of doing the join?

Thanks in Advance.

Best regards,
Marcus Rehm

Re: Pushdown Capabilities with RDBMS

Posted by Marcus Rehm <ma...@gmail.com>.

Hi Sudheesh and Zelaine,

Thank you both for the reply.

Unfortunately I wasn't able to recreate the scenario with a small set of
tables and dummy records, but I created the issue 4818 -
https://issues.apache.org/jira/browse/DRILL-4818 on Jira and attached the
query, the physical plan and the Json Profile related to the case.

Again, thank you very much for the support.

Best,
Marcus



2016-07-15 17:48 GMT-03:00 Marcus Rehm <ma...@gmail.com>:

> Hi all,
>
> I started to teste Drill and I'm very excited about the possibilities.
>
> By now I'm trying to map ours databases running on Oracle 11g. After try
> some queries I realized that the amount of time Drill takes to complete is
> bigger than a general sql client takes. Looking the execution plan I saw
> (or understood) that Drill is doing the join of tables and is not pushing
> it down to the database.
>
> Is there any configuration required to it? How can I tell Drill to send to
> Oracle the task of doing the join?
>
> Thanks in Advance.
>
> Best regards,
> Marcus Rehm
>

Re: Pushdown Capabilities with RDBMS

Posted by Zelaine Fong <zf...@maprtech.com>.

Note though that the decision on whether to push down the join is dependent
on the cost of doing so relative to other options.  I believe there are
currently limitations in the costing model of the JDBC storage plugin that
prevent the optimizer from choosing the "correct" plan in all cases.

But as Sudheesh noted, it would be good to capture your scenario in a Jira
ticket.

-- Zelaine

On Fri, Jul 15, 2016 at 5:26 PM, Sudheesh Katkam <sk...@maprtech.com>
wrote:

> Hi Marcus,
>
> I am glad that you are exploring Drill! Per RDBMS storage plugin
> documentation [1], join pushdown is supported. So the scenario you
> described is likely a bug; can you open a ticket [2] with the details on
> how to reproduce the issue?
>
> Thank you,
> Sudheesh
>
> [1] https://drill.apache.org/docs/rdbms-storage-plugin/ <
> https://drill.apache.org/docs/rdbms-storage-plugin/>
> [2] https://issues.apache.org/jira/browse/DRILL <
> https://issues.apache.org/jira/browse/DRILL>
>
> > On Jul 15, 2016, at 1:48 PM, Marcus Rehm <ma...@gmail.com> wrote:
> >
> > Hi all,
> >
> > I started to teste Drill and I'm very excited about the possibilities.
> >
> > By now I'm trying to map ours databases running on Oracle 11g. After try
> > some queries I realized that the amount of time Drill takes to complete
> is
> > bigger than a general sql client takes. Looking the execution plan I saw
> > (or understood) that Drill is doing the join of tables and is not pushing
> > it down to the database.
> >
> > Is there any configuration required to it? How can I tell Drill to send
> to
> > Oracle the task of doing the join?
> >
> > Thanks in Advance.
> >
> > Best regards,
> > Marcus Rehm
>
>

Re: Pushdown Capabilities with RDBMS

Posted by Sudheesh Katkam <sk...@maprtech.com>.

Hi Marcus,

I am glad that you are exploring Drill! Per RDBMS storage plugin documentation [1], join pushdown is supported. So the scenario you described is likely a bug; can you open a ticket [2] with the details on how to reproduce the issue?

Thank you,
Sudheesh

[1] https://drill.apache.org/docs/rdbms-storage-plugin/ <https://drill.apache.org/docs/rdbms-storage-plugin/>
[2] https://issues.apache.org/jira/browse/DRILL <https://issues.apache.org/jira/browse/DRILL>

> On Jul 15, 2016, at 1:48 PM, Marcus Rehm <ma...@gmail.com> wrote:
> 
> Hi all,
> 
> I started to teste Drill and I'm very excited about the possibilities.
> 
> By now I'm trying to map ours databases running on Oracle 11g. After try
> some queries I realized that the amount of time Drill takes to complete is
> bigger than a general sql client takes. Looking the execution plan I saw
> (or understood) that Drill is doing the join of tables and is not pushing
> it down to the database.
> 
> Is there any configuration required to it? How can I tell Drill to send to
> Oracle the task of doing the join?
> 
> Thanks in Advance.
> 
> Best regards,
> Marcus Rehm