Query:
SELECT X.a FROM X LEFT JOIN Y ON X.b=Y.c INNER JOIN (SELECT X.a FROM X) t ON
X.a=t.a;
102:LogicalProject(a=[$0])
101:LogicalJoin(condition=[=($0, $3)], joinType=[inner])
98:LogicalJoin(condition=[=($1, $2)], joinType=[left])
96:EnumerableTableScan(table=[[X]])
97:EnumerableTableScan(table=[[Y]])
100:LogicalProject(a=[$0])
99:EnumerableTableScan(table=[[X]])
Root {kind: SELECT, rel: EnumerableProject#173, rowType: RecordType(BIGINT a),
fields: [<0, a>], collation: []}
173:EnumerableProject(a=[$1])
172:EnumerableJoin(condition=[=($0, $1)], joinType=[inner])
170:EnumerableProject(a=[$0])
96:EnumerableTableScan(table=[[X]])
171:EnumerableJoin(condition=[=($1, $2)], joinType=[left])
96:EnumerableTableScan(table=[[X]])
97:EnumerableTableScan(table=[[Y]])
Schema:
Table X:
Column Type
a bigint not null
b bigint not null
Table Y:
Column Type
c bigint not null
On 2018/05/14 20:36:28, Vitalii Diravka <[email protected]> wrote:
> Check yours intermediate RelNode plan, which is going to be improved by
> this rule. The operands and their ordering are important in matching of the
> rules.
> https://github.com/apache/calcite/blob/master/core/src/main/java/org/apache/calcite/rel/rules/JoinPushThroughJoinRule.java#L87
>
> From your case the rule should match. Could you share please the plan?
>
> Kind regards
> Vitalii
>
>
> On Tue, May 8, 2018 at 4:24 PM [email protected] <
> [email protected]> wrote:
>
> > Okay let me rephrase my question.
> >
> > I am executing a query having inner and left join. The plan generated by
> > calcite for that query is like
> >
> > inner
> > / \
> > left C
> > / \
> > A B
> >
> > But it would be better if it proceeds inner join first, as the row count
> > will be reduced further to proceed left join. So the plan needed is,
> >
> > left
> > / \
> > inner B
> > / \
> > A C
> >
> > So I read through the documentation and found JoinPushThroughJoinRule will
> > do so. So I tried applying them (Till now, I am giving only row count to
> > statistics). But no change in the plan.
> >
> > Then I found onMatch() in JoinPushThroughJoinRule.java has
> > not been called at all.
> >
> > How can this be resolved?
> >
> >
> > On 2018/05/04 14:23:00, Michael Mior <[email protected]> wrote:
> > > I have no immediate answers, but it would be helpful if you could
> > provide a
> > > complete working example of code that exhibits the problem.
> > >
> > > --
> > > Michael Mior
> > > [email protected]
> > >
> > >
> > > Le ven. 4 mai 2018 à 03:58, Valli Annamalai <[email protected]> a
> > > écrit :
> > >
> > > > I wanted to apply JoinPushThroughJoinRule.LEFT and
> > > > JoinPushThroughJoinRule.RIGHT to the plan for optimization. So I added
> > > > those 2 rules to Program and ran it as a sequence of programs.
> > > >
> > > > But I am not getting plan applied with these rules
> > > >
> > > > I did read [here
> > > > <
> > > >
> > https://issues.apache.org/jira/browse/CALCITE-457?focusedCommentId=14205221&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14205221
> > > > >]
> > > > that adding rules does not mean that the plan will be better. So sanity
> > > > checks may fail. In my case, during onMatch() checks (in matchRecurse()
> > > > from ValcanoRuleCall.java), it fails because of no parents for the
> > subsets
> > > > (getParentRels() returns empty list) .
> > > >
> > > > Thus there are no successors and the recursion has not been trigerred.
> > So
> > > > the rules has not been fired (onMatch() in
> > JoinPushThroughJoinRule.java has
> > > > not been called at all). What can be done for that?
> > > >
> > > > Any other flags need to be enabled or rules to be added??
> > > >
> > > >
> > > > Thanks in advance
> > > >
> > >
> >
>
