Thanks, Maryann.

Yes, offset and fetch was set to null.

Finally I achieved EnumerableSort rewriting with my implementation
(ArchiveEnumerableSort) by unregistering EnumerableSort from planner
in TableScan implementation (while keeping registration of my
converter rule).

  planner.removeRule(EnumerableRules.ENUMERABLE_SORT_RULE);

After this planner chose as expected ArchiveEnumerableSort.

Seems that Calcite prefer EnumerableRules despite if any other presented.

And in my implementation I was able to remove sorting of column, which is
already sorted in my dataset.

On Wed, Nov 4, 2015 at 6:29 AM, Maryann Xue <[email protected]> wrote:
> Hi Alexander,
>
> Not sure if it's the problem you have, but it might be worth checking if
> your ArchiveEnumerableSort contains "collation" only, i.e. offset==null &&
> fetch==null. The SortRemoveRule deals with sort-only Sort rels, if either
> offset or fetch is not null this rule won't apply. Let me know if it helps.
>
>
> Thanks,
> Maryann
>
> On Tue, Nov 3, 2015 at 11:09 AM, Alexander Reshetov <
> [email protected]> wrote:
>
>> Thanks Maryann.
>>
>> Trying to debug what's going on I created ArchiveConverterRule (and
>> registered this rule).
>> Rule implements ConverterRule and add convert(), which returns
>> ArchiveEnumerableSort.
>>
>> So basically ArchiveEnumerableSort should be my replacement of
>> EnumerableSort.
>> But when I call explain plan for select I get EnumerableSort anyway,
>> while I expect to see
>> implemented ArchiveEnumerableSort.
>>
>> Should it work as I expect, or there is something other that I need to do?
>>
>> On Sat, Oct 31, 2015 at 3:26 AM, Maryann Xue <[email protected]>
>> wrote:
>> > Oh, it's not on master branch yet. It's the "calcite" branch, and for
>> > PhoenixTable specifically, you can go:
>> >
>> https://git1-us-west.apache.org/repos/asf?p=phoenix.git;a=blob;f=phoenix-core/src/main/java/org/apache/phoenix/calcite/PhoenixTable.java;h=4be745036335443bc1bcaae07acf1f32fda6c59a;hb=refs/heads/calcite
>> > .
>> >
>> > On Fri, Oct 30, 2015 at 4:36 PM, Alexander Reshetov <
>> > [email protected]> wrote:
>> >
>> >> Could you please point me to PhoenixTable in repo? Unfortunately I
>> >> could not find it.
>> >>
>> >> On Fri, Oct 30, 2015 at 10:33 PM, Maryann Xue <[email protected]>
>> >> wrote:
>> >> > SortRemoveRule.INSTANCE is already in the default rule set. I think
>> what
>> >> > you should do is just make your table return the right collation
>> trait.
>> >> For
>> >> > example in Phoenix we have a Table implementation called PhoenixTable
>> >> which
>> >> > overrides getStatistic() and implements getCollations() of Statistic.
>> >> >
>> >> > On Fri, Oct 30, 2015 at 1:40 PM, Alexander Reshetov <
>> >> > [email protected]> wrote:
>> >> >
>> >> >> Is it correct that I need to override
>> >> >> org.apache.calcite.rel.AbstractRelNode.register and add rule
>> >> >> SortRemoveRule.INSTANCE ?
>> >> >>
>> >> >> But while it checks RelCollation, then I should also somehow set it.
>> >> >> But I could not find how to do it.
>> >> >>
>> >> >> On Fri, Oct 30, 2015 at 4:23 PM, Alexander Reshetov
>> >> >> <[email protected]> wrote:
>> >> >> > Here some clarification of my intends.
>> >> >> >
>> >> >> > Calcite choosing next plan:
>> >> >> >
>> >> >> > +------+
>> >> >> > | PLAN |
>> >> >> > +------+
>> >> >> > | EnumerableSort(sort0=[$0], dir0=[DESC])
>> >> >> >   ArchiveTableScan(table=[[MAIN, SUB]], fields=[[0, 1, 2, 3, 4, 5,
>> 6,
>> >> >> > 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21]])
>> >> >> >  |
>> >> >> > +------+
>> >> >> >
>> >> >> >
>> >> >> > The problem is invocation of EnumerableSort. Because data already
>> >> >> > sorted by one column. So I need to force exclusion of
>> EnumerableSort
>> >> >> > from planner.
>> >> >> >
>> >> >> > Do I need implement some interface or override some method to
>> achieve
>> >> >> this?
>> >> >> >
>> >> >> > On Fri, Oct 30, 2015 at 9:51 AM, Julian Hyde <[email protected]>
>> >> wrote:
>> >> >> >> Yes, this is possible. In fact Phoenix does exactly this.
>> >> >> >>
>> >> >> >> Take a look at https://issues.apache.org/jira/browse/CALCITE-818
>> and
>> >> >> https://issues.apache.org/jira/browse/CALCITE-71, which were fixed
>> >> along
>> >> >> the way.
>> >> >> >>
>> >> >> >>> On Oct 29, 2015, at 1:16 AM, Alexander Reshetov <
>> >> >> [email protected]> wrote:
>> >> >> >>>
>> >> >> >>> Hello,
>> >> >> >>>
>> >> >> >>> I have kinda theoretical question. But it should help to
>> understand
>> >> >> >>> overall Calcite's possibility.
>> >> >> >>>
>> >> >> >>> Suppose there is some abstract format/table, which supports only
>> >> >> >>> sorting by one column.
>> >> >> >>>
>> >> >> >>> Let's say there is next query passed to calcite:
>> >> >> >>>
>> >> >> >>>    select * from table order by id
>> >> >> >>>
>> >> >> >>> Is it possible to rewrite query tree so it will change this
>> request
>> >> >> into this?
>> >> >> >>>
>> >> >> >>>    select * from table
>> >> >> >>>
>> >> >> >>> In this way after logical query rewrite Calcite should not be
>> >> >> >>> concerned about sorting at all.
>> >> >> >>
>> >> >>
>> >>
>>

Reply via email to