Description
In cases where the optimizer chooses to do index multi-probing for an IN list (DERBY-47), the presence of a LIKE condition whose argument is a parameter marker leads to an ASSERTion failure with sane jars. With insane jars, the same query returns incorrect results (it only returns rows matching the first argument in the IN list).
As an example, if we have the following DDL:
create table ct (i int, c1 char(25), c2 char(40));
insert into ct values 1, 2, 3, 4, 5, 6, 7, 8, 9;
insert into ct values 0, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19;
insert into ct select 7 * i from ct;
insert into ct select 13 * i from ct;
update ct set c1 = cast(i as char(25));
update ct set c2 = c1 || c1;
create index idx2 on ct (c1, c2);
Then we see the following results:
– This one works.
select i,c1,c2 from ct where c1 in ('1','2') and c2 like '%' order by i;
– These all work, too.
prepare p1 as 'select i,c1,c2 from ct where c1 in (''2'',''1'') and c2 like ''%''';
execute p1;
prepare p1 as 'select i,c1,c2 from ct where c1 in (?,?) and c2 like ''%''';
execute p1 using 'values (''1'', ''2'')';
– But these all fail. Something about having a parameter in the LIKE is triggering the assertion...
prepare p1 as 'select i,c1,c2 from ct where c1 in (?,?) and c2 like ? order by i';
execute p1 using 'values (''1'', ''2'', ''%'')';
prepare p1 as 'select i,c1,c2 from ct where c1 in (?,?) and c2 like ?';
execute p1 using 'values (''1'', ''2'', ''%'')';
prepare p1 as 'select i,c1,c2 from ct where c1 in (''2'',''1'') and c2 like ?';
execute p1 using 'values (''%'')';
prepare p1 as 'select i,c1,c2 from ct where c1 in (''2'',?) and c2 like ?';
execute p1 using 'values (''1'', ''%'')';
The actual assertion is thrown because multiprobing expects the start and stop keys to be same value. But something about the LIKE parameter violates that expectation:
ERROR XJ001: Java exception: 'ASSERT FAILED All multi-probing result sets are expected to have a single key that is both the start key AND the stop key, but that is not the case.: org.apache.derby.shared.common.sanity.AssertFailure'.
This is a regression introduced by the DERBY-47 changes; everything runs fine in 10.2 and earlier.
Attachments
Attachments
Issue Links
- is related to
-
DERBY-47 Some possible improvements to IN optimization
- Closed
- relates to
-
DERBY-3061 Wrong results from query with two conjuncts
- Closed