non pk present before pk in RVC cause wrong execute plan

[孟庆义(孟庆义)]

Hi, Dear

 

Create table t (a integer not null, b integer not null, c integer not null,
d integer constraint pk primary key (a, b, c));

 

SQL1 : select * from t where (a,d) in ( (1,4) , (2, 3))

SQL2 : select * from t where (d,a) in ( (4,1) , (3,2))

 

SQL1 and SQL2 have different execute plan, the SQL2’s plan is not correct.
Found in branch 3.2

 

See below log:

 

0: jdbc:phoenix:localhost> explain select * from t where (a,d) in ((1,4),(2,
3));
+------------------------------------------+
|                   PLAN                   |
+------------------------------------------+
| CLIENT 1-CHUNK PARALLEL 1-WAY SKIP SCAN ON 2 KEYS OVER T [1] - [2] |
|     SERVER FILTER BY (A, D) IN
([128,0,0,1,128,0,0,4],[128,0,0,2,128,0,0,3]) |
+------------------------------------------+
2 rows selected (0.032 seconds)
0: jdbc:phoenix:localhost> explain select * from t where (d,a) in ((4,1),(3,
2));
+------------------------------------------+
|                   PLAN                   |
+------------------------------------------+
| CLIENT 1-CHUNK PARALLEL 1-WAY SKIP SCAN ON 2 KEYS OVER T [3] - [4] |
+------------------------------------------+
1 row selected (0.031 seconds)

 

Daniel meng