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. > >> >> >> > >> >> > >> >
