parallel apply got wrong result when there are multiple columns as primary key

[guo-shaoge]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

  drop table if exists t1;
  create table t1 (c1 bigint, c2 int, c3 int, c4 int, primary key(c1, c2), index (c3));
  insert into t1 values(1, 1, 1, 1), (1, 2, 3, 3), (2, 1, 4, 4), (2, 2, 2, 2);

  set @@tidb_enable_parallel_apply = off;
  select (select /*+ NO_DECORRELATE() */ sum(c4) from t1 where t1.c3 = alias.c3) from t1 alias where alias.c1 = 1;

  set @@tidb_enable_parallel_apply = on;
  select (select /*+ NO_DECORRELATE() */ sum(c4) from t1 where t1.c3 = alias.c3) from t1 alias where alias.c1 = 1;

2. What did you expect to see? (Required)

Results are same for parallel apply and serial apply.

3. What did you see instead (Required)

Results are different

4. What is your TiDB version? (Required)

v7.5.0

[guo-shaoge]

I think this bug is because PhysicalIndexLookupReader.Clone() miss clone CommonHandles, check code

[seiya-annie]

/found customer