[
http://issues.apache.org/jira/browse/DERBY-805?page=comments#action_12366736 ]
Jeff Lichtman commented on DERBY-805:
-------------------------------------
Here is my reply to Army's reply (also previously posted to derby-dev):
> > How could a predicate be pushable to only one side of a union? Can you
> > provide an example of a predicate that can be pushed only to one side?
> >
> If we take something along the lines of:
>
>
> select ... from
> t2,
> (select * from t1 union values (1,2), (3,4), (5,6)) X1 (a,b)
> where X1.a = t2.i;
>
>
> In this case the predicate X1.a = t2.i could be pushed to the left ("select *
> from t1") and used when reading T1, but couldn't be pushed to the VALUES
> clause because there's no underlying table.
OK. One way to deal with this is to put a ProjectRestrictNode between the union
node and the values clause as a place to park the predicate. To make things
simple, you might want to always put ProjectRestrictNodes under both sides of
the union during preprocessing (i.e. after binding but before optimization). In
some cases the extra nodes won't be needed, but ProjectRestrictNodes (and the
corresponding ProjectRestrictResultSets) are cheap. Also, you could eliminate
unnecessary ProjectRestrictNodes at the end of optimization (possibly in
modifyAccessPaths()).
This approach would give better performance in some cases, and could simplify
the code (since you wouldn't have to figure out when the predicates are
pushable).
> Push join predicates into union and other set operations. DERBY-649
> implemented scalar (single table) predicate pushdown. Adding join predicate
> push down could improve performance significantly.
> --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: DERBY-805
> URL: http://issues.apache.org/jira/browse/DERBY-805
> Project: Derby
> Type: Sub-task
> Components: SQL
> Versions: 10.1.2.0, 10.2.0.0
> Environment: generic
> Reporter: Satheesh Bandaram
> Assignee: A B
> Fix For: 10.2.0.0
> Attachments: DERBY-805.html
>
> Fix for DERBY-649 implemented scalar (single table) predicate push down into
> UNIONs. While this improves performance for one set of queries, ability to
> push join-predicates further improves Derby performance by enabling use of
> indices where possible.
> For example,
> create view V1 as select i, j from T1 union all select i,j from T2;
> create view V2 as select a,b from T3 union all select a,b from T4;
> insert into T1 values (1,1), (2,2), (3,3), (4,4), (5,5);
> For a query like
> select * from V1, V2 where V1.j = V2.b and V1.i =1;
> If the join order choosen is V1,V2, V1 can use index on V1.i (if present)
> following fix for DERBY-649. But if there is a index on V2.b also, Derby
> currently can't use that index. By pushing join predicate, Derby would be
> able to use the index and improve performance. Some of the queries I have
> seen (not the one shown here...) could improve from 70-120 seconds to about
> one second.
> Note there is a good comment by Jeff Lichtman about join-predicate push down.
> I am copying parts of it here for completeness of this report: (Modified)
> If predicate push down is done during optimization, it would be possible to
> push joins into the union as long as it's in the right place in the join
> order.
> For example:
> create view v as select * from t1 union all select * from t2;
> select * from v, t3 where v.c1 = t3.c2;
> In this select, if t3 is the outer table then the qualification could be
> pushed into the union and optimized there, but if t3 is the inner table the
> qualification can't be pushed into the union.
> If the pushing is done at preprocess time (i.e. before optimization) it is
> impossible to know whether a join qualification like this can be safely
> pushed.
> There's a comment in UnionNode.optimizeIt() saying:
> /* RESOLVE - don't try to push predicated through for now */
> This is where I'd expect to see something for pushing predicates into the
> union during optimization.
> BTW, the business of pushing and pulling predicates during optimization can
> be hard to understand and debug, so maybe it's best to only handle the simple
> cases and do it during preprocessing.
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
