Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
4.13.0
-
None
-
None
Description
Given following table:
CREATE TABLE test_table ( PK1 INTEGER NOT NULL, PK2 INTEGER NOT NULL, PK3 INTEGER NOT NULL, DATA INTEGER, CONSTRAINT TEST_PK PRIMARY KEY (PK1,PK2,PK3))
and a sql:
select * from test_table t where (t.pk1 >=2 and t.pk1<5) and ((t.pk2 >= 4 and t.pk2 <6) or (t.pk2 >= 8 and t.pk2 <9))
Obviously, it is a typical case for the sql to use SkipScanFilter,however, the sql actually does not use Skip Scan, it use Range Scan and just push the leading pk column expression (t.pk1 >=2 and t.pk1<5) to scan,the explain sql is :
CLIENT PARALLEL 1-WAY RANGE SCAN OVER TEST_TABLE [2] - [5] SERVER FILTER BY ((PK2 >= 4 AND PK2 < 6) OR (PK2 >= 8 AND PK2 < 9))
I think the problem is affected by the WhereOptimizer.KeyExpressionVisitor.orKeySlots method, in the following line 763, because the pk2 column is not the leading pk column,so this method return null, causing the expression
((t.pk2 >= 4 and t.pk2 <6) or (t.pk2 >= 8 and t.pk2 <9)) is not pushed to scan:
757 boolean hasFirstSlot = true; 758 boolean prevIsNull = false; 759 // TODO: Do the same optimization that we do for IN if the childSlots specify a fully qualified row key 760 for (KeySlot slot : childSlot) { 761 if (hasFirstSlot) { 762 // if the first slot is null, return null immediately 763 if (slot == null) { 764 return null; 765 } 766 // mark that we've handled the first slot 767 hasFirstSlot = false; 768 }
For above WhereOptimizer.KeyExpressionVisitor.orKeySlots method, it seems that it is not necessary to make sure the PK Column in OrExpression is leading PK Column,just guarantee there is only one PK Column in OrExpression is enough.
Attachments
Attachments
Issue Links
- is related to
-
PHOENIX-3328 Optimize ORed leading pk column range comparisions to SkipScan
- Resolved