[Discussion] Support Secondary Index on Carbon Table

[Indhumathi]

Hi Community,

Currently we have datamaps like,* default datamaps* which are block and
blocklet and *coarse grained datamaps* like bloom, and *fine grained
datamaps* like lucene which helps in better pruning during query. What if we
introduce another kind of datamap which can hold blockletId as index? Initial level,
we call it as index which will work as a child table to the main table like we have
MV in our current code.

Yes, lets introduce the secondary index to carbon table which will be the
child table to main table and it can be created on column like we create
lucene datamap, where we give index columns to create index. In a similar way,
we create secondary index on column, so indexes on these column will be blocklet IDs
which will help in better pruning and faster query when we have a filter query on the
index column.

Currenlty we will take it as index table and then later part we will make
it inline to datamap interface.

So design document is attached in JIRA, please give your suggestion/inputs.

JIRA Link: CARBONDATA-3680
<https://issues.apache.org/jira/browse/CARBONDATA-3680>

Thanks & Regards,
Indhumathi M

[ravipesala]

+1

Regards,
Ravindra.

On Wed, 5 Feb 2020 at 8:03 PM, Indhumathi M <[hidden email]> wrote:

> Hi Community,
>
> Currently we have datamaps like,* default datamaps* which are block and
> blocklet and *coarse grained datamaps* like bloom, and *fine grained
> datamaps* like lucene
> which helps in better pruning during query. What if we introduce another
> kind of datamap which can hold blockletId as index? Initial level, we call
> it as index which
> will work as a child table to the main table like we have MV in our current
> code.
>
> Yes, lets introduce the secondary index to carbon table which will be the
> child table to main table and it can be created on column like we create
> lucene datamap,
> where we give index columns to create index. In a similar way, we create
> secondary index on column, so indexes on these column will be blocklet IDs
> which will
> help in better pruning and faster query when we have a filter query on the
> index column.
>
> Currenlty we will take it as index table and then later part we will make
> it inline to datamap interface.
>
> So design document is attached in JIRA, please give your suggestion/inputs.
>
> JIRA Link: CARBONDATA-3680
> <https://issues.apache.org/jira/browse/CARBONDATA-3680>
>
> Thanks & Regards,
> Indhumathi M
>

--
Thanks & Regards,
Ravi

[Ajantha Bhat]

+1

On Wed, 5 Feb, 2020, 8:02 pm Ravindra Pesala, <[hidden email]> wrote:

> +1
>
> Regards,
> Ravindra.
>
> On Wed, 5 Feb 2020 at 8:03 PM, Indhumathi M <[hidden email]>
> wrote:
>
> > Hi Community,
> >
> > Currently we have datamaps like,* default datamaps* which are block and
> > blocklet and *coarse grained datamaps* like bloom, and *fine grained
> > datamaps* like lucene
> > which helps in better pruning during query. What if we introduce another
> > kind of datamap which can hold blockletId as index? Initial level, we
> call
> > it as index which
> > will work as a child table to the main table like we have MV in our
> current
> > code.
> >
> > Yes, lets introduce the secondary index to carbon table which will be the
> > child table to main table and it can be created on column like we create
> > lucene datamap,
> > where we give index columns to create index. In a similar way, we create
> > secondary index on column, so indexes on these column will be blocklet
> IDs
> > which will
> > help in better pruning and faster query when we have a filter query on
> the
> > index column.
> >
> > Currenlty we will take it as index table and then later part we will make
> > it inline to datamap interface.
> >
> > So design document is attached in JIRA, please give your
> suggestion/inputs.
> >
> > JIRA Link: CARBONDATA-3680
> > <https://issues.apache.org/jira/browse/CARBONDATA-3680>
> >
> > Thanks & Regards,
> > Indhumathi M
> >
> --
> Thanks & Regards,
> Ravi
>

[kumarvishal09]

+1
-Regards
Kumar Vishal

On Wed, 5 Feb 2020 at 8:08 PM, Ajantha Bhat <[hidden email]> wrote:

> +1
>
> On Wed, 5 Feb, 2020, 8:02 pm Ravindra Pesala, <[hidden email]>
> wrote:
>
> > +1
> >
> > Regards,
> > Ravindra.
> >
> > On Wed, 5 Feb 2020 at 8:03 PM, Indhumathi M <[hidden email]>
> > wrote:
> >
> > > Hi Community,
> > >
> > > Currently we have datamaps like,* default datamaps* which are block and
> > > blocklet and *coarse grained datamaps* like bloom, and *fine grained
> > > datamaps* like lucene
> > > which helps in better pruning during query. What if we introduce
> another
> > > kind of datamap which can hold blockletId as index? Initial level, we
> > call
> > > it as index which
> > > will work as a child table to the main table like we have MV in our
> > current
> > > code.
> > >
> > > Yes, lets introduce the secondary index to carbon table which will be
> the
> > > child table to main table and it can be created on column like we
> create
> > > lucene datamap,
> > > where we give index columns to create index. In a similar way, we
> create
> > > secondary index on column, so indexes on these column will be blocklet
> > IDs
> > > which will
> > > help in better pruning and faster query when we have a filter query on
> > the
> > > index column.
> > >
> > > Currenlty we will take it as index table and then later part we will
> make
> > > it inline to datamap interface.
> > >
> > > So design document is attached in JIRA, please give your
> > suggestion/inputs.
> > >
> > > JIRA Link: CARBONDATA-3680
> > > <https://issues.apache.org/jira/browse/CARBONDATA-3680>
> > >
> > > Thanks & Regards,
> > > Indhumathi M
> > >
> > --
> > Thanks & Regards,
> > Ravi
> >
>

[kunalkapoor]

+1

Thanks
Kunal Kapoor

On Wed, Feb 5, 2020, 5:33 PM Indhumathi M <[hidden email]> wrote:

> Hi Community,
>
> Currently we have datamaps like,* default datamaps* which are block and
> blocklet and *coarse grained datamaps* like bloom, and *fine grained
> datamaps* like lucene
> which helps in better pruning during query. What if we introduce another
> kind of datamap which can hold blockletId as index? Initial level, we call
> it as index which
> will work as a child table to the main table like we have MV in our current
> code.
>
> Yes, lets introduce the secondary index to carbon table which will be the
> child table to main table and it can be created on column like we create
> lucene datamap,
> where we give index columns to create index. In a similar way, we create
> secondary index on column, so indexes on these column will be blocklet IDs
> which will
> help in better pruning and faster query when we have a filter query on the
> index column.
>
> Currenlty we will take it as index table and then later part we will make
> it inline to datamap interface.
>
> So design document is attached in JIRA, please give your suggestion/inputs.
>
> JIRA Link: CARBONDATA-3680
> <https://issues.apache.org/jira/browse/CARBONDATA-3680>
>
> Thanks & Regards,
> Indhumathi M
>

[Jacky Li]

+1&nbsp;
Thanks for proposing this :)


Regards,
Jacky




------------------&nbsp;原始邮件&nbsp;------------------
发件人:&nbsp;"Kunal Kapoor"<[hidden email]&gt;;
发送时间:&nbsp;2020年2月6日(星期四) 凌晨2:13
收件人:&nbsp;"dev"<[hidden email]&gt;;

主题:&nbsp;Re: [Discussion] Support Secondary Index on Carbon Table



+1

Thanks
Kunal Kapoor

On Wed, Feb 5, 2020, 5:33 PM Indhumathi M <[hidden email]&gt; wrote:

&gt; Hi Community,
&gt;
&gt; Currently we have datamaps like,* default datamaps* which are block and
&gt; blocklet and *coarse grained datamaps* like bloom, and *fine grained
&gt; datamaps* like lucene
&gt; which helps in better pruning during query. What if we introduce another
&gt; kind of datamap which can hold blockletId as index? Initial level, we call
&gt; it as index which
&gt; will work as a child table to the main table like we have MV in our current
&gt; code.
&gt;
&gt; Yes, lets introduce the secondary index to carbon table which will be the
&gt; child table to main table and it can be created on column like we create
&gt; lucene datamap,
&gt; where we give index columns to create index. In a similar way, we create
&gt; secondary index on column, so indexes on these column will be blocklet IDs
&gt; which will
&gt; help in better pruning and faster query when we have a filter query on the
&gt; index column.
&gt;
&gt; Currenlty we will take it as index table and then later part we will make
&gt; it inline to datamap interface.
&gt;
&gt; So design document is attached in JIRA, please give your suggestion/inputs.
&gt;
&gt; JIRA Link: CARBONDATA-3680
&gt; <https://issues.apache.org/jira/browse/CARBONDATA-3680&gt;
&gt;
&gt; Thanks &amp; Regards,
&gt; Indhumathi M
&gt;

[niuge]

+1

I have a suggestion.

Comparatively, query hint will be better to avoid queries push downed to SI
table.
Example:
SELECT /*disable_si*/ * FROM main_table WHERE name='abc'

Regards,
Zhi Liu



--
Sent from: http://apache-carbondata-dev-mailing-list-archive.1130556.n5.nabble.com/

[David CaiQiang]

+1



-----
Best Regards
David Cai
--
Sent from: http://apache-carbondata-dev-mailing-list-archive.1130556.n5.nabble.com/

[manishgupta88]

+1

Regards
Manish Gupta

On Thu, 6 Feb 2020 at 1:50 PM, David CaiQiang <[hidden email]> wrote:

> +1
>
>
>
> -----
> Best Regards
> David Cai
> --
> Sent from:
> http://apache-carbondata-dev-mailing-list-archive.1130556.n5.nabble.com/
>